Establishing best practice approaches for developing credible electricity demand and energy forecasts for network planning

670
Establishing best practice approaches for
developing credible electricity demand and
energy forecasts for network planning
Working Group
C1.32
December 2016

Members
G. Ancell NZ
S. Avdakovic BA
J. Breedt ZA
T. Bugten NO
A.R. Carrasco ES
G. Carruthers AE
Z. Meng CN
D. Pilenieks RU
S. van den Waeyenberg BE
WG C1.32
Copyright © 2016
“All rights to this Technical Brochure are retained by CIGRE. It is strictly prohibited to reproduce or provide this publication in
any form or by any means to any third party. Only CIGRE Collective Members companies are allowed to store their copy on
their internal intranet or other company network provided access is restricted to their own employees. No part of this
publication may be reproduced or utilized without permission from CIGRE”.
Disclaimer notice
“CIGRE gives no warranty or assurance about the contents of this publication, nor does it accept any responsibility, as to the
accuracy or exhaustiveness of the information. All implied warranties and conditions are excluded to the maximum extent
permitted by law”.
WG XX.XXpany network provided access is restricted to their own employees. No part of this publication may be
reproduced or utilized without permission from CIGRE”.
Disclaimer notice
“CIGRE gives no warranty or assurance about the contents of this publication, nor does it accept any responsibility, as to the
ESTABLISHING BEST PRACTICE
APPROACHES FOR DEVELOPING CREDIBLE
ELECTRICITY DEMAND AND ENERGY
FORECASTS FOR NETWORK PLANNING
ISBN : 978-2-85873-373-6

ESTABLISHING BEST PRACTICE APPROACHES FOR DEVELOPING CREDIBLE ELECTRICITY DEMAND AND ENERGY FORECASTS
FOR NETWORK PLANNING
Page 3
ESTABLISHING BEST
PRACTICE APPROACHES
FOR DEVELOPING CREDIBLE
ELECTRICITY DEMAND AND
ENERGY FORECASTS FOR
NETWORK PLANNING
Table of Contents
EXECUTIVE SUMMARY................................................................................................................. 5
Survey findings ........................................................................................................................... 5
Recommendations and conclusions........................................................................................... 5
Suggestions for future CIGRE work ........................................................................................... 5
Chapter 1 Description of the Working Group ................................................................................ 7
Background................................................................................................................................. 7
Scope.......................................................................................................................................... 7
Work methodology...................................................................................................................... 8
Context with other C1 working groups ....................................................................................... 8
Document structure .................................................................................................................... 8
Chapter 2 Overview of load and energy forecasting................................................................... 10
Introduction............................................................................................................................... 10
Who is involved in load and energy forecasting?..................................................................... 10
What is being forecast? ............................................................................................................ 11
Inputs ........................................................................................................................................ 12
Load and energy forecasting methodologies ........................................................................... 12
Evaluation and process review................................................................................................. 16
Chapter 3 Survey Design............................................................................................................ 18
Introduction............................................................................................................................... 18
Design....................................................................................................................................... 18
Round 1 .................................................................................................................................... 18
Round 2 .................................................................................................................................... 19
Round 3 .................................................................................................................................... 19
Round 4 .................................................................................................................................... 20
Round 5 .................................................................................................................................... 22
Chapter 4 Analysis of survey results........................................................................................... 23

Page 4
Introduction............................................................................................................................... 23
Survey respondents.................................................................................................................. 23
Electrical Network characteristics............................................................................................. 23
Forecast purpose...................................................................................................................... 23
Forecast properties................................................................................................................... 26
Data collection .......................................................................................................................... 27
Forecast methodology & tools.................................................................................................. 27
Forecast accuracy & methodology revisions............................................................................ 28
Top three issues in current forecasting methodologies............................................................ 29
Current challenges in forecasting............................................................................................. 29
Future challenges in forecasting............................................................................................... 29
Chapter 5 Discussion .................................................................................................................. 31
Introduction............................................................................................................................... 31
Best Practice............................................................................................................................. 31
Future challenges ..................................................................................................................... 32
Active load ................................................................................................................................ 32
Paradigm shifts in demand forecasting .................................................................................... 32
Other Techniques ..................................................................................................................... 34
Summary .................................................................................................................................. 34
Chapter 6 Conclusions and recommendations ........................................................................... 35
Suggestions for future CIGRE work ......................................................................................... 35
Bibliography/References ............................................................................................................. 37
Appendix 1 - Summary of survey answers ................................................................................. 38
Appendix 2 - List of Members ..................................................................................................... 56
Appendix 3 - Definitions .............................................................................................................. 57

Page 5
EXECUTIVE SUMMARY
Working Group C1.32 carried out a survey of international utilities to establish best practice approaches for
developing electricity load and energy forecasts. The final survey was sent around to CIGRE members in October
2015. By the end of January 2016, a total of 29 fully completed surveys were available for analysis.
The respondents represent 18 countries. The survey was filled in more than once for Australia, New Zealand, and
Japan as there are multiple TSOs in the country each covering different areas. China and Belgium filled in the
survey multiple times but for different types of forecast (e.g., forecasts for long-term grid development versus short-
term operational planning). Most respondents come from Oceania (10 responses), Asia, and Europe (8 responses
each). One response was received from Africa, North America, and South America each.
Survey findings
The survey results show that the forecasts are mainly used for long-term grid planning; in 40% of these 23 cases
the responses also apply for short-term operational planning. In 18% of the cases, the forecast was also used for
generation development; in 21%, for security of supply; and in 29% also for generation adequacy studies. The
timing of the peak load is evenly spread: 50% of the respondents work in a country or region with a peak load in
winter and 44% in summer.
In most cases both load and energy are forecast: Thirty six percent of the forecasts were for load only. No forecasts
focused solely on energy. Almost all forecasts are required by regulation, yet in most cases regulation does not
prescribe the methodology. Most respondents used load and energy forecasting software that was developed in-
house.
Most respondents do not explicitly forecast reactive power. Reactive power is often indirectly forecast based on
assumed or historic values of power factor.
Most forecasting teams (56%) consist of a small group of up to 5 people.
The forecast methodology is frequently revised. Most respondents revised it in the last 2 years (53%) and of this
group almost all are also planning to revise again in the next 2 years (88%). Of those who reviewed the
methodology more than 2 years ago, most also plan to revise the forecast methodology again in the next 2 years.
Given that forecast methodologies are frequently reviewed, it seems that best practice in forecast methodologies is
not widely agreed (at least amongst the survey respondents). Some aspects where there are common approaches
are the modelling of load by type (e.g. residential, commercial, industrial), using a combination of top down and
bottom approaches to forecasting and forecast horizons (e.g. most long term forecasts look ahead 5 to 15 years).
Recommendations and conclusions
The most important aspects to improve in the forecast method are:
 Input from external sources (such as economic growth, population, etc.);
 Measurement data;
 Input from the distribution level.
The survey responses indicate the most important influences needing to be incorporated into load forecasting in
the next 10 years are:
 Penetration of Renewable Energy Sources (RES);
 Demand side response management;
 Storage and electric vehicles.
Suggestions for future CIGRE work
The Working Group identified a large number of areas where further work could be undertaken by CIGRE. The
Working Group elected to focus on two areas for immediate work.

Page 6
Survey of the capabilities and performance of inhouse load and energy forecasting tools
This work stream follows on from this Working Group. The work is a survey of the capabilities and performance of
demand forecasting tools. The survey will include questions on:
 Forecasting methodologies incorporated in the forecasting tools.
 Developed models for electric vehicles, storage, RES, demand side management etc.
 Accuracy of the forecasts.
This work stream is an input into the next work stream.
Best practice models for load and energy forecasting.
This work stream will identify or develop best practice models for:
 Penetration of Renewable Energy Sources (RES).
 Balancing Supply and Demand Models in new energy mix.
 Demand side response management.

Page 7
Chapter 1 Description of the Working Group
The terms of reference for Working Group C1.32 were approved on 24 May 2014. A list of members is contained in
Appendix 2.
Background
Network owners need accurate forecasts of electricity consumed by loads to make prudent investment decisions.
Network owners can then assess the capability of their networks to meet the forecast electricity consumed by loads
and identify the necessary changes to the network to meet system reliability and adequacy targets.
The following issues make the task of producing accurate load forecasts challenging:
 changes to customer behaviour responding to increases in electricity prices;
 availability of new embedded generation systems such as roof-top PV arrays;
 the fact that some system operators may not yet have complete information about the installed generation
capacities beyond the customers' meters, e.g. roof-top PV;
 even less system operators may have hourly or better metering of consumption and distributed generation;
 government policies encouraging energy efficiency;
 government and regulatory policies in tariff requirements (e.g. flat fee or dynamic pricing);
 government policies encouraging embedded generation in distribution networks;
 uncertainty regarding economic activity;
 uncertainty regarding exchange rates;
 uncertain demand from trade exposed industries.
CIGRE WG C1-24 examined the increasing use of market simulations to determine economic benefits of
transmission augmentations and hence justify those augmentations. This work identified that load forecasts need to
have sufficient granularity to support an effective assessment of the economic impacts of network augmentation.
The WG C1-24 report identified that a number of demand conditions need to be studied to develop a robust
assessment of the economic benefit. Load forecasts must therefore facilitate study of a variety of network loading
conditions, and a variety of adequacy assessment needs.
The increasing installation of embedded generation in distribution networks and the characteristics of that
generation (particularly roof-top PV) are changing the utilization of the transmission and distribution networks
across time. For example, high penetration of rooftop PV on distribution feeders in Australia, California, Germany
and elsewhere has already suppressed the midday peak network utilization, and peak demand is moving to the
early evening. As PV may continue to become less expensive, such changes may increase in scale and become
more wide spread world-wide. Such changes to the timing of the peak demand net of distributed generation, and
that the peak network utilization may be driven by peak insolation rather than demand, present another challenge
for forecasters. The scope of the working group is to examine best practice approaches from around the world and
emerging trends.
Scope
This working group aims to examine the demand and energy forecasting techniques currently being employed by
network companies around the world. The working group will seek to identify:
1. What are the key issues and challenges that need to be addressed in producing load forecasts to support
network planning and system adequacy activities?
2. What methodologies are employed in developing forecasts? Including
a. How are uncertain future developments such as the electric car, heat pump or rooftop PV penetration
being accounted for in energy and load forecasting?
b. What time granularities (hourly all year or even shorter intervals), time horizons (how many years into
the future), and scenario handling are employed in developing forecasts?
c. How are transmission and distribution system operators cooperating in developing forecasts for loads
and for distributed generation?

Page 8
d. What is the relationship between data used for operational time scale load and generation forecasting
and planning timescale forecasting?
3. What approaches are employed to assess the accuracy of forecasts, and to adjust them in reaction to
observed developments?
4. Those best practice techniques that tend to produce the most accurate forecasts and that meet emerging
needs and applications for demand forecasts.
5. What issues need to be overcome to adopt best practice techniques? These may include better forecasting
tools, improved data and data systems.
6. The impacts of demand side response on demand forecasting techniques, and what this means for best
practice.
The scope will be addressed by developing and executing an electronic survey of network companies to identify
current forecasting issues and best practice approaches.
Work methodology
The Working Group had meetings at the 2014 Paris Session, the 2015 Lund Symposium and the 2016 Paris
Session. The Working Group also had regular teleconference meetings. A web survey tool (Survey Monkey) was
used to develop and carry out a survey of CIGRE members.
Context with other C1 working groups
Study Committee C1 has a strategic plan vision and focus to anticipate and plan a system that best fits the
paradigm shift brought about by rapid evolution in generation patterns and economics, demand response,
Information & Communications Technology (ICT), and in social, environmental, regulatory frameworks and
expectations.
There are six C1 Working Groups that have published or are in the process of publishing Technical Brochures in
2016 that deal with issues relating to distribution side generation, planning and development. These six Working
Groups complement each other and focus on different aspects of the same subject. The summary below should
help readers understand the differences among these working areas:
 C1.18/C2/C6 deals with solutions for coping with limits for very high penetrations of renewable energy
solutions.
 C1.20 focuses on how to accommodate high load growth and urban development in future plans.
 C1.27 looks at the definition of reliability in light of new developments in various devices and services that
offer customers and system operators new levels of flexibility. The focus is on how new developments
should change the definition of reliability and adequacy used with generation and transmission planning.
The Working Group suggested necessary changes to the definitions of reliability and adequacy.
 JWG C1.29 looks at the requirement for a change in the conventional planning criteria for future
transmission networks as a result of an increased level of distributed energy resources at MV and LV
levels. The Working Group also assessed the adequacy of currently adopted, and/or those in the process
of being delivered, transmission planning-methods.
 C1.30 addresses technical risks and solutions from periodic, large surpluses or deficits of available
renewable generation in a particular area. The working group defined a so called risk-solution matrix to
find and illustrate the total situation of risks and solutions which appears in utilities today.
 C1.32 examines the demand and energy forecasting techniques currently being employed by power
systems around the world.
Document structure
Chapter 1 provides the reader with an overview of the working group, what led to it and the processes followed
within the work group. Chapter 2 provides an overview on load forecasting theory and application of load
forecasting within the electricity supply industry. This chapter provides details on the purpose of forecasting,

Page 9
modeling techniques used and data collection methods and discusses components of spatial forecasting and
selection of a target network for forecasting purposes when applying forecasting theory.
Chapter 3 describes the design of the survey. The survey results are displayed in Chapter 4 with statistics and
assumptions made. Chapter 5 discusses the results of the survey, considers best practice in the context of the
survey and examines future challenges. Chapter 6 contains the conclusion of the document and suggests future
work to be done within CIGRE.
Appendix 1 contains the survey questions and a summary of responses including respondent comments.
Appendix 2 contains a list of members in the working group. Appendix 3 has definitions for some of the terminology
used in this report.

Page 10
Chapter 2 Overview of load and energy forecasting
Introduction
Load and energy forecasting is fundamental to electricity utility operations and planning. Most power system
investment decisions, from scheduling and dispatch to development of new generation plant and transmission
infrastructure, are underpinned by load forecasting. Forecasting is a process that is focused on predicting future
events or conditions by combining current facts with future market and evolving technology trends, cycles and
seasonality.
Forecasts of electrical load or electrical energy consumed or produced are made by a number of different entities.
The forecasts are used for a range of purposes, from real-time operation of the power system, to determining the
required long term generation, transmission and distribution development plans.
Some of the drivers of load and energy growth or decline are changing. The uptake of distributed generation such
as rooftop photo voltaic generation at consumers’ premises means that consumers will supply some of their
electricity needs themselves and will at times inject excess generation into the power system. Power flow on
transmission and distribution systems will fundamentally change.
This chapter provides a discussion of load and energy forecasting aspects that are considered in the survey.
Appendix 3 contains a table of terms and definitions used.
Who is involved in load and energy forecasting?
There are two types of organization who provide input to or make use of load and energy forecasts. The first of
these types are organizations directly involved in the production, transmission and use of electrical industry. The
second type are organizations which use load and energy forecasts for monitoring and reporting or developing
other forecasts (e.g. GDP growth in countries).
The first type of organization includes distribution owners and operators, transmission owners and operators,
system operators, generators, electricity market participants (e.g. those involved in hedge trading). These
organizations can be vertically integrated (e.g. generation and transmission) or quite separate such as independent
system operators. This group uses forecasting for operational purposes such as scheduling and dispatch, short
term outage planning and in the longer term, generation and network capacity enhancement decision making.
The second type of organization includes government bodies, regulators and private organizations which provide
services to the first group of organizations such as providing weather forecasts.
The first questions asked in the survey are aimed to identify the respondent and what they make forecasts for.
These questions are summarised in Table 1.
National Context Number of TSOs in the respondent’s country
The company’s scope of responsibilities in terms of:
Geographical area
Voltage levels
Role in the electrical cycle from generation to consumption
Power System
and regulatory
environment
Electrical network characteristics of the geographical area in the TSO scope
Size of the peak load
Timing of the peak load
Amount of installed production capacity not directly connected to the TSO grid: total +
division per fuel type
Relationship between short-term and long-term forecasting
Regulatory requirements
Table 1 - SURVEY QUESTIONS FOR IDENTIFICATION

Page 11
What is being forecast?
Load and energy forecasts are made for a range of purposes. For electricity utilities, these purposes range from
scheduling and dispatch decisions to generation and transmission investment decision making.
Of interest is what quantity is being forecast. The forecast could be a time series for the load over the next 24 hours
at half hour intervals. The forecast could be peak or minimum demand over a year period for the next 3 years. Or
the forecast could be annual energy consumption. Other quantities such as reactive power or power factor can be
part of the forecast.
The forecast can be of a net or gross nature. A net forecast is the combination of the underlying load and
distributed generation and is typically measured at a certain point such as the connection to a distribution or
transmission network. This measurement is a net measurement of load and generation beyond the meter. The load
beyond the meter can be made of distinct components such as residential load, commercial load and industrial
load. The load behind the meter can also have distributed generation which will reduce the metered net load
amounts. A load and energy forecast can be for the net metered point or for the components of load beyond the
meter (gross forecast).
Load and energy forecasts can be made on a system, region substation, voltage or customer connection point
basis.
Load and energy forecasts have many characteristics:
 Time horizon (how far the forecast looks ahead);
 Time granularity (e.g. intervals for time series, single instant);
 Geographic granularity (area, region or load level).
In terms of time horizon (how far the forecast looks ahead) load and energy forecasting can be categorized into the
following groups:
 Short term load forecasting;
 Medium term load forecasting;
 Long term load forecasting.
The definition of short, medium and long term will be different for different organizations. An ISO might consider short
term to mean the next 24 hours, medium term to be the next six weeks and long term to be the next three years. A
transmission grid owner might consider anything less than 3 years to be short term, 10 years to be medium term and
30 years to be long term.
Each of these forecasts can be analyzed and produced with different granularities. An ISO may forecast load at each
connection to the grid at half hour intervals over the coming 24 hours. A transmission system planner may forecast
loads at peak demand over a year for each of the next 30 years.
Some questions asked in the survey are concerned with forecast purpose and properties. These questions are
summarized in Table 2.

Page 12
Purpose of the
forecast
General purpose of the forecast discussed in the survey (multiple surveys will be possible
per TSO)
Short description of the forecast
Forecasting of load versus energy
Forecast
properties
Forecasted moment (peak, minimum, other)
Net versus gross load forecasting
Time horizon of the forecast (separate question for short-term and for long-term
forecasting)
Time granularity of the forecast (separate question for short-term and for long-term
forecasting)
Geographical granularity of the forecast (national, regional, substation, client, voltage
level)
Forecasting of the power factor or of the apparent power
Forecasting of reactive power
Table 2 - FORECAST PURPOSE AND PROPERTIES
Inputs
Inputs for load and energy forecasting come from a range of areas. These areas include historic measurements of
electrical and other quantities such as temperature. Other forecasts such as national GDP may also be inputs.
Assumptions about the future effects of programs (e.g. energy efficiency, RES subsidies) can be inputs.
Information about known changes to load (e.g. connections of new customers or permanent load shifts between
substations) inform load and energy forecasts.
The quality of the data sets collected is of utmost importance. The aspects that should be considered [1] are:
 Accuracy and reliability of the source of the data;
 Adequacy of the data to the represented phenomenon, accordance with past cycles and trends with a
complete time range data;
 Timelines of data collection and processing should meet the forecasters needs;
 Consistency of the data, regularly updated.
The accuracy and validation of historic electrical data affect the usefulness of forecasts. For example, historic load
data can be measured in SCADA or by revenue metering. The revenue metering data is likely to be better as
revenue meters are generally have a higher accuracy class and are more likely to be calibrated than the voltage
and current transformers used for SCADA measurements. The revenue meter data is checked for errors (validated)
while SCADA is not.
Source of measurement data
Time granularity of measurement data used for forecasting
Use of data from external sources
Collaboration with the DSO: input data, DSO load forecasting, frequency of information exchanges between TSO
and DSO
Information of load type
Table 3 - INPUT DATA
Load and energy forecasting methodologies
There are many different types of load forecasting techniques and methodologies. Most forecasting techniques can
be classified as either qualitative or quantitative (see Figure 1). Qualitative forecasting techniques are subjective,
based on the opinion and judgment of informed parties. Qualitative forecasting is appropriate when past data are not

Page 13
available. Quantitative methods forecast future data as a function of past data. They can be used when past numerical
data is available and when it is reasonable to assume that some of the patterns in the data are expected to continue
into the future.
Figure 1: Quantitative and qualitative forecasting
QUANTITATIVE FORECASTING
Quantitative techniques and methodologies are most often grouped as statistical and deterministic approaches.
These can include models such as:
 Multiple linear regressions models: load or some transformation of load is usually treated as the dependent
variable, while weather, macroeconomic and calendar variables are treated as independent variables.
 Univariate models: these models forecast the evolution of a variable based on the past observations of the
same variable over time. ARMA models (autoregressive moving-average models) are widely used, especially
for short term forecasting. We can find in this group trend projection methods, which focus on patterns,
pattern changes, and disturbances caused by random influences.
 Artificial Neural Networks: ANN is a soft computing technique that does not require the forecaster to explicitly
model the underlying physical system. By simply learning the patterns from historical data, a mapping
between the input variables and the electricity demand can be constructed, and then adopted for prediction.
 Deterministic methods which incorporate the identification and explicit determination of relationships
between the factors being forecasted and influence of other factors on these forecasts.
Companies can develop their own tool for forecasting models or use existing commercial tools such as SAS1,
eVIEWS2, MATLAB3, and LoadSEER4.
QUALITATIVE FORECASTING
To produce an informative forecast of future demand and energy needs it is of great importance to understand the
area on which the forecast is done. The impacting market cycles and developmental trends are the main influencing
factors. A good base for collating of information for the area is based on three functions: scanning, tracking and
1SAS Institute Inc., USA, http://www.sas.com/en_us/software/foundation.html.
2 IHS Inc., USA, http://www.eviews.com/home.html.
3 Mathworks, USA, http://www.mathworks.com/products/matlab/.
4 Integral Analytics, USA, http://www.integralanalytics.com/products-and-services/spatial-growth-
planning/loadseer.aspx.

Page 14
monitoring. All relevant information on the area should be scanned for future changes, any concerns or changing
patterns should be tracked for validity and continuity, and then continuously monitored to derive trends.
Five main criteria have been identified by Payne [2] and can be described to inform the market intelligence needed
to derive accurate forecasts. These Five Market Intelligence factors to utilize are shown in Figure 2.
Figure 2: Market INTELLIGENCE framework
First, there is a need to derive the amount of demand or energy needed in a specific geographical area. Second, it
should be determined when this load is anticipated or needed, third, it should be determined where this load should
be allocated spatially. The fourth factor indicates why the demand will be growing and lastly what is the primary factor
driving this load.
FORECASTING FOR A TARGET MARKET
Long term load forecasting can be an iterative process. Transmission planners can determine needed changes to
the grid to accommodate future load and generation. The parties making investment decisions for generation and
load may then change their own development plans considering the indicated grid changes. The parties making
forecasts need to understand this factor.
The drivers for load and generation investment vary from country to country. In some countries, there will be strong
political drivers to provide electricity to people who don’t have electricity supply. There is no existing infrastructure so
the power system can be developed from scratch. These countries can be called Developing Countries in terms of
their electricity infrastructure. In other countries, the electricity infrastructure may be well established and the drivers
are based more on meeting reliability and adequacy targets. These countries can be called Developed Countries.
The power system is developed on an incremental basis making the best use of existing infrastructure.

Page 15
In Developing Countries, there are plentiful opportunities for growth and development into the future. This increases
the uncertainty levels as there are many parties which can influence the scenario that will play out into the future. In
Developed Countries, the levels of uncertainty are significantly less with regards to economic growth and
development and the demand forecasts associated with these networks.
Figure 3 shows how technology and forecast methods can be influenced by the nature of the target market.
Figure 3: forecast on target market
SCENARIOS
Long term forecasts often have a number of scenarios. These scenarios reflect different assumptions. For example,
a high growth scenario may determine load growth under very beneficial economic conditions. The high growth
scenario can be contrasted with a low growth scenario.
Each scenario may be associated with probabilities but this is not always necessary. For example, an expected
(average) load growth scenario can be determined. The probability of future load exceeding this scenario is expected
to be 50%. Similarly, a high load growth scenario with a probability of exceedance of 10% can be determined.
Competing transmission or generation expansion options can be tested against a range of different scenarios. An
expansion option that is ideal for one scenario may perform poorly under another scenario.
When defining the scenarios, experts usually have to identify the important assumptions that have an effect on load
growth. These assumptions can vary from one country to another, or even in the same country, from one historical
moment to another.
TOP-DOWN LOAD FORECASTING
In this approach, forecasts are developed from big geographical areas, with the possibility to taper down to a certain
level of granularity. Even though data may come from different sources, they are added so they can be related by
the models to general indices. These indices are based in weather and macroeconomic inputs (heating degree days,

Page 16
average daily temperature, Gross Domestic Product, population, prices of electricity, housing stock, and so forth).
These models are usually consistent and efficient.
This approach can be a combination of different models for each sector or specific customer type. Depending on the
sector, the relationship between the consumption and the indices might vary and it is thus convenient to model them
separately if possible. External role players such as government decisions, system operator structures and country
development status can have an influence on this approach.
BOTTOM-UP LOAD FORECASTING
In contrast to top-down load forecasting, where a forecast is made at a high level, the bottom-up approach consists
of forecasting at a lower level, and then adding the obtained forecasts in combination with diversification factors
applicable. These methods are based on knowledge of the end-use consumption, and consider each type of
customer separately. As a result, these methods are very detailed and complex.
HIERARCHICAL LOAD FORECASTING
Hierarchical load forecasting is a new trend in forecasting that tries to preserve the pros of top-down approach while
supplying forecasts at sub-regional level as well. It provides load forecasts at various levels of the hierarchy
(geographic, temporal, circuit connection or revenue class hierarchy). This offers the utilities more insights into the
power system and customer usage patterns than the traditional top-down or bottom-up load forecasts.
ENSEMBLE FORECASTING
Ensemble forecasts are produced by combining several forecasts that are made by using different methodologies.
SPATIAL FORECASTING
Spatial forecasting is when the growth patterns in a specified area, whether a region or a country is matched with its
physical geographical properties. This can be a very informative approach to identifying growth patterns and
overlaying it with different trends such as sectorial development, mineral availabilities, urban and rural developments
and population growth cycles.
Some questions asked in the survey are concerned with the above forecast purpose and properties. These
questions are summarized in Table 4.
Bottom-up versus top-down approach for forecasting
Number of forecasts for different scenarios
Impact of temperature, electric vehicles, heat pumps, renewable energy sources, air conditioners,
micro grids, storage, demand side response management, electric efficiency
Name of the tool used for load forecasting
Creator of the tool
Number of people in the company working on the forecasting
Drivers of the load in the country
Table 4 - FORECAST METHODOLOGY AND TOOLS
Evaluation and process review
Forecasting is a dynamic process where models and tools must be periodically reviewed and improved to adapt them
to the actual circumstances. Methodologies that once might have been valid may not be valid anymore in the future.

Page 17
Frequency of revisions of the forecast
Frequency of revisions of the forecast methodology
Future plan for a review of the forecast methodology
Barriers to improve the accuracy of the forecast
Table 5 - FORECAST ACCURACY AND REVISIONS

Page 18
Chapter 3 Survey Design
Introduction
The purpose of the Working Group was to establish a view on the current forecasting methods and expected
challenges for load forecasting in the future. The idea was to gather information from as many and as varied
countries as possible to be able to draw valid conclusions and to analyse the impact of country and grid
characteristics on forecasting needs and methods. Keeping in mind the purpose of the research and the
practicalities of data collection, the group chose to collect data via a written survey.
Design
The survey was designed progressively in several rounds after the CIGRE 2014 Paris Session:
• Round 1 – October 2014. Internal survey of the load forecasting methodology among the Workgroup
members.
• Round 2 – November-December 2014. Qualitative analysis of the internal survey responses.
• Round 3 – January-March 2015. Creation and testing of the draft survey (focus: content).
• Round 4 – April-May 2015. Creation and testing of the online survey (focus: format).
• Round 5 – May 2015. Finalization of the survey (Lund Meeting).
Round 1
October 2014. Internal survey of the load forecasting methodology among the Workgroup members.
During the first meeting the Workgroup decided to start with a survey among the Workgroup members only to
identify the most important aspect of load forecasting. An open-ended questions survey based on the Terms of
Reference of the Workgroup was sent around by the convenor to the other members.
The questions were rapidly defined as the group agreed that the Terms of Reference included the most relevant
questions that would allow to refining the scope of the survey. In addition to questions on types of forecasting and
data collection, the survey questions would concern differences between countries in terms of economic and
electrical load stability as well as a list of uncertainties that all seemed to be confronted with (whether economic
uncertainties, such as GDP evolution, or technological uncertainties, such as Rooftop PV, Electrical Vehicles, etc.).
Although the stakes appear to be very different, all were expected to cope with similar uncertainties and
challenges.
The clearly defined Terms of Reference and a fruitful discussion during a Workgroup teleconference resulted in the
following open-ended questions. The questions were kept short and relatively easy to answer as to stimulate a
quick progress of the group’s work:
 The purpose of the demand forecast (e.g. long term planning or short term system operation)?
 What are the key issues and challenges that need to be addressed in producing the load forecast?
 What methodologies are employed in developing the forecasts?
 How are uncertain future developments such as the electric car, heat pump or rooftop PV penetration being
accounted for in energy and load forecasting?
 What time granularities (hourly all year or even shorter intervals), time horizons (how many years into the
future), and scenario handling are employed in developing forecasts?
 How are transmission and distribution system operators cooperating in developing forecasts for loads and
for distributed generation?
 What is the relationship between data used for operational time scale load and generation forecasting and
planning timescale forecasting?
 What approaches are employed to assess the accuracy of forecasts, and to adjust them in reaction to
observed developments?
 The impacts of demand side response on demand forecasting techniques.

Page 19
Round 2
November-December 2014. Qualitative analysis of the internal survey responses.
The main topics and forecasting characteristics were identified through a qualitative analysis of the internal survey
responses, showing the Workgroup members’ own expertise in load forecasting. Eight working group members
responded on the internal survey. The analysis entailed a reading and encoding of the answers; both by
respondent (to get a view of all aspects of load forecasting for each member) and by question (to discover strong
(dis)similarities across members). Frequently returning terminology on the one hand and varying accents laid in
responses on the other aided in creating a broad list of more concrete questions. The analysis provided an idea on
the multitude of methods that could be used and warned for the importance of well defining and scoping the subject
matter.
Round 3
January-March 2015. Creation and testing of the draft survey (focus: content)
The qualitative analysis of the Workgroup members’ description of their own forecasts brought forward the
high-level structure of the survey. The main topics were clear. Next, during several feedback rounds, the work
group described specific questions within each category to create a draft version of the actual survey that was to be
sent around to CIGRE members. At this stage the content validity of the survey remained high as all questions
proposed by Workgroup members were maintained to make sure all relevant issues were included.
An online survey tool Survey Monkey5 was chosen to create the questionnaire. The sought benefits of working with
an online tool were its user friendliness (both for the creators of the survey as for the respondents) and the
facilitation of the analysis of the responses afterwards. The questionnaire included a combination of closed and
open-ended questions. Closed questions were preferred in case the workgroup has already identified main
response categories while open-ended questions would better serve in case possible responses were less clear or
it was expected that respondents would provide richer answers if they could formulate them freely. Open-ended
questions were also used to allow respondents to add response categories that were not previously identified or to
request additional explanation. However, some answers to open questions may be difficult to interpret and answers
may be spread over many categories, making it hard to analyse them.
The draft survey finally included 10 sections. Table 6 shows how the main topics from the qualitative analysis
(round 2) were restructured in the draft survey (round 3).
5 Survey Monkey is a web based application for designing and carrying out surveys.
https://www.surveymonkey.com/.

Page 20
Main topics Qualitative Analysis Sections of the Draft Survey
Regulatory and organisational background
Size of the organization and forecasting staff
Size of the power system
Interconnections with other grids
Level of economic development
General information
Electrical network characteristics
Forecast properties
Purpose: short term operational planning, outage
planning, long term grid reinforcement,
generation adequacy, …
Subject: load, energy, ...; peak, minimum, ...
Granularity and level, accuracy
Driving factors of load
Purpose of the forecast
Forecast properties
Forecasting techniques & methodologies
Tools and databases
Different methodologies per level
Input data
Customer properties and demand composition
Customer mix
Specific effects (air conditioning, heat pumps,
electric vehicles, ...)
Forecaster development & training
Forecaster experience and development
Array of forecaster within the organization
working from different perspectives?
Data collection
Forecast methodology and tools
Renewable Energy
Impact of local generation on load forecast taken
into account?
Use of production information
Forecast accuracy and revisions
Future challenges in forecast
Other remarks
Table 6 - SELECTION AND STRUCTURING OF SURVEY TOPICS (ROUNDS 1-2-3)
Round 4
April-May 2015. Creation and testing of the online survey (focus: format)
In the next round the group paid attention to the formulation of questions, the logical grouping of questions in
sections covering a common topic, the possible response formats, and the mandatory requirement to get a
response. The focus was on the reliability of the questions (i.e., to make sure each question would be interpreted in
the same way by all respondents) and on the validity of the questions (i.e., to give the group the information it was
after) (Office of Quality Improvement, 2010).
Table 7 details the specific topics in each section of the draft survey. At this stage the draft survey was created
online and sent to all the members of the working group for a first major test. An extra open question was included
per section for the respondents to give any feedback on their tests. The feedback included the usefulness of
questions, the clarity of the formulation and need for definitions, propositions on response formats, or remarks on
the mandatory character of some questions. Also during the test phase of the survey the use of the online tool
proved convenient as the comments of the workgroup testers were centrally collected and exportable to excel
format in a structured way.

Page 21
Sections of the Draft Survey (mandatory questions in italic)
1. General information
Identification and contact information of the respondent
Identification of the respondent’ company (TSO)
Number of TSOs in the respondent’s country
The company’s scope of responsibilities in terms of:
- Geographical area
- Voltage levels
- Role in the electrical cycle from generation to consumption
2. Electrical network characteristics of the geographical area in the TSO scope
Size of the peak load
Timing of the peak load
Amount of installed production capacity not directly connected to the TSO grid: total + division
per fuel type
Relationship between short-term and long-term forecasting
Regulatory requirements
3. Purpose of the forecast
General purpose of the forecast discussed in the survey (multiple surveys will be possible per
TSO)
Short description of the forecast
Forecasting of load versus energy
4. Forecast properties
Forecasted moment (peak, minimum, other)
Net versus gross load forecasting
Time horizon of the forecast (separate question for short-term and for long-term forecasting)
Time granularity of the forecast (separate question for short-term and for long-term forecasting)
Geographical granularity of the forecast (national, regional, substation, client, voltage level)
Forecasting of the power factor or of the apparent power
Forecasting of reactive power
5. Data collection
Source of measurement data
Time granularity of measurement data used for forecasting
Use of data from external sources
Collaboration with the DSO: input data, DSO load forecasting, frequency of information
exchanges between TSO and DSO
Information of load type
6. Forecast methodology and tools
Bottom-up versus top-down approach for forecasting
Number of scenarios
Impact of temperature, electric vehicles, heat pumps, renewable energy sources, air
conditioners, micro grids, storage, demand side response management, electric efficiency
Name of the tool used for load forecasting
Creator of the tool
Number of people in the company working on the forecasting
Drivers of the load in the country
7. Use of production information
More detailed questions on the impact of local production
Input data on local production
Correction for unmeasured production
8. Forecast accuracy and revisions
Frequency of revisions of the forecast
Frequency of revisions of the forecast methodology
Future plan for a review of the forecast methodology
Barriers to improve the accuracy of the forecast
9. Future challenges in forecast

Page 22
Selection of the most important changes that are needed in forecasting in the next 10 years
Future challenges for the delivery of accurate load forecasts
10. Other remarks
Table 7 - FINAL SURVEY FORMAT
Round 5
May 2015. Finalization of the survey (Lund Meeting)
During the CIGRE meeting at Lund the group rigorously went through all questions and test feedback received
from 11 workgroup members in order to finalize the survey.
The number of questions was considered too large and reduced from 52 to 34 to stimulate survey participation.
Some questions were simply deleted as evaluated “not relevant enough” or requiring too many research by the
respondent (e.g., detailed numerical information that would also be available in ENTSO-E databases) while other
questions were merged to avoid too much overlap or repetition.
Some questions were added at the request of WG 1.23 on long-term planning and different scenario’s in the future
for generation and load, depending on the context of the country.
The final survey (including 36 questions of which 16 are mandatory) was sent for Study Committee approval during
the summer. Approval was granted in August 2015. The final survey was sent around to CIGRE members on
October 5, 2015.

Page 23
Chapter 4 Analysis of survey results
Introduction
This chapter presents an analysis of the survey results. The survey questions and a summary of responses are
contained in Appendix 1.
The final survey was sent around to CIGRE members on October 5, 2015, with the request to respond by
November 30, 2015, and a reminder sent in December. By the end of January, a total of 29 fully completed
surveys were available for analysis.
Survey respondents
The distribution of respondents by organisation is shown in Figure 4.
Figure 4: Survey responses by company
The respondents represent 18 countries. The survey was filled in more than once for Australia, New Zealand, and
Japan as there are multiple TSOs in the country each covering different areas. China and Belgium filled in the
survey multiple times but for different types of forecast (e.g., forecasts for long-term grid development versus short-
term operational planning). Most respondents come from Oceania (10 responses), Asia, and Europe (8 responses
each). One response was received from Africa, North America, and South America each.
Electrical Network characteristics
The timing of the peak load is evenly spread: 45% of the respondents work in a country or region with a peak load
in winter; 52% in summer; 1 respondent indicated no seasonal difference. The timing of the peak is expected to be
most likely related to country-specific characteristics (cultural habits, lifestyles, etc. in response to the changing
weather conditions throughout the year).
Forecast purpose
Table 8 shows the survey respondents’ load and energy forecast purposes. Note the many respondents indicated
multiple purposes. The forecasts for which the survey was filled in are mainly used for long-term grid planning.
18%
24%
35%
9%
15%
What type of company do you work for?
Independent System Operator
Transmission grid owner
Integrated transmission grid
owner and system operator
Distribution
Vertically integrated
generation and transmission

Page 24
How would you describe the purpose of the forecast(s) for which you fill in this
survey?
Answer Options
Response
Percent
Response
Count
Short-term operational planning (dispatching,
outages, maintenance, ...)
55.9% 19
Long-term grid development 70.6% 24
Long-term generation development 17.6% 6
Security of supply 20.6% 7
Generation adequacy 29.4% 10
Table 8 – Forecast purpose
Together all respondents listed 40 different types of forecasts, ranging from forecasts on the minimum, peak or
average value, calculated on an hourly, daily, monthly, or yearly basis, and calculated for the next day to the next
30 years. The majority of the forecasts focus on the peak moment. The type of forecast is shown in Figure 5.
Figure 5: Forecast type
The granularity of the forecasts is shown in Figure 6. Short-term forecasts focus on hourly or daily values for the
next few days; long-term forecasts mainly calculate yearly values for 5 to 15 years ahead although forecasts up to
20 to 30 years in the future are not exceptional.
Average
20%
Minimum
20%
Peak
53%
Other time of
interest
7%
Forecast type
Average Minimum Peak Other time of interest

Page 25
Figure 6: Forecast GRANULARITY
The time horizon of the forecasts is shown in Figure 7. The majority of the forecasts study periods of one year, but
this is logical considering the majority of the forecasts for which the survey was filled in are for long-term grid
development.
Figure 7: Forecast Horizon
The relationship between forecasts used for short term operational planning and long term grid expansion is shown
in Figure 8. A quarter of the respondents indicated that there was no relationship between forecasts for short-term
23%
15%
4%
11%
47%
Forecast granularity
per hour per day per week per month per year
28%
7%
7%33%
18%
7%
Forecast horizon
1-14 days ahead 1 month ahead 1 year ahead
5-15 years ahead 20-30 years ahead Specific years

Page 26
and long-term purposes. In cases where there was, it mostly concerned the use of the same metering data (55% of
the respondents), and to a lesser extent the same level of analysis (24% of the respondents) and the same data on
connected entities (consumption/generation). As expected the respondents confirmed that the forecasting
methodology is usually different.
Figure 8: Relationship between short term and long term forecasts
Forecast properties
In most cases (66%) both load and energy are forecast. 34% of the forecasts are concerned with load only. No
forecast focuses solely on energy.
The level (national, regional, substation, client, voltage) is shown in Table 9. One forecast may include forecasts at
different levels. For example, load and energy forecasts can be made at the substation or lower level and then
aggregated into regional and national forecasts taking into account diversity between substation loads.
On which geographical or client level do you do the forecast?
Answer Options
Response
Percent
Response
Count
national level 45.5% 15
regional level 48.5% 16
substation level 54.5% 18
client level 18.2% 6
voltage level 6.1% 2
Other (please specify) 18.2% 6
Table 9 - Level at which forecasts are made
Most respondents (80%) had forecasts at the substation level or lower.
Two thirds of the respondents acknowledged a spatial aspect to the forecast.
36%
16%
16%
14%
11%
7%
Relationship between forecasting for short-term
operational planning and long-term grid reinforcement
Same historic measured data
No relationship
Same level of analysis (e.g.,
substations)
Same data provided by connected
parties (e.g. distribution company)
Same generation data
Similar forecasting methodology

Page 27
The use of long-term forecasting for grid development is confirmed in the level of analysis: 76% of the respondents
who filled in the survey for long-term purposes indicated working at least on substation level.
The majority of the respondents do not forecast reactive power as such (71%), although occasionally reactive
power is indirectly forecasted based on assumed or historic values of power factors.
Data collection
The data used in general comes from the distribution system operators, the grid users (direct clients), and
government institutions, most often updated yearly or on demand (except for metering data).
Distribution system operators overall provide information on metering (48% client projects (20% of the survey
respondents), transfers in their grid (48%) and projects of their clients (24%), local growth rates (41%), and local
production (28%). They also give feedback on the forecast (24%) and validate its accuracy (24%).
Direct clients also provide information on client projects (62% of the survey respondents), metering (41%), local
growth rates (21%), and own forecasts (24%). To a lesser extent grid users give data on local production (14%)
and give feedback on the forecast (17%).
Certain statistics are typically collected from government institutions, such as national growth rates (52% of the
respondents), sectoral growth rates (34%), local production information (21%), population data (45%), weather
statistics (48%), and weather forecasts (38%).
Regulators sometimes validate the forecast accuracy (18%) and provide overall feedback (27%).
More than half of the respondents categorize load per type. For more than half of them (62%) the categories are
high-level, namely “residential”, “industrial”, or “commercial”. A small group (14%) considers more detail, such as
categories per sector.
The use of production information is ambiguous: the majority of the respondents answered not to correct for
measured production, half of the respondents work with net load values, half with gross load values. This seems to
indicate that in the case where gross values are used, the amount of decentralized production or at least the
amount of unmeasured production is negligible.
Forecast methodology & tools
More than half of the respondents use a mix of top-down and bottom-up approach (62%) and make forecasts for up
to 2 to 5 scenarios (66%). The tools used for load forecasting are mostly developed in-house (66%).
The use of load sector (e.g. residential, commercial, industrial) in the load and energy forecast is shown in Table
10.
Do you use information on the load sector in the load forecast?
Answer Options
Response
Percent
Response
Count
Yes: on the level of 'residential', 'industrial',
'commercial', but no more detail
62.1% 18
Yes: on more detailed level, such as per sector 13.8% 4
No 24.1% 7
Table 10 - use of load sector
Table 11 shows which load related components are not modelled in current forecasting methodologies. The most
infrequently modelled components are heat pumps, heating appliances and demand side response. The most
frequently modelled components are RES and temperature effects.

Page 28
Component
Not
modelled
penetration of renewable energy sources (RES) 4
the impact of temperature 6
air conditioners 10
increased electric efficiency 12
electric vehicles 14
storage 14
demand side response management 15
heating appliances 16
heat pumps 17
Table 11 - Modelling
Forecast accuracy & methodology revisions
The frequency at which long term forecasts are reviewed in shown in Table 12. Most long term forecasts are
reviewed every year.
How often are the load forecasts reviewed?
Answer Options
Response
Percent
Response
Count
Yearly 60.0% 18
Every 2 years 3.3% 1
Other 13.3% 4
Table 12 – Long term Forecast review frequency
The forecasting methodology is surprisingly frequently reviewed as seen in Table 13. Most respondents reviewed
their methodology in the last 2 years.
Have you reviewed your load forecast methodology in the last 5 years?
Answer Options
Response
Percent
Response
Count
No 14.3% 4
Yes, in the last 1 to 2 years 64.3% 18
Yes, in the last 3 to 5 years 21.4% 6
Table 13 – Forecast methodology review times
Most respondents are planning to revise it again in the next 2 years as seen in Table 14. For those who reviewed
the methodology more than 2 years ago, (24%) most also plan to revise again in the next 2 years (63%).

Page 29
Are there plans to review your load forecast methodology in the next 5 years?
Answer Options
Response
Percent
Response
Count
No 14.8% 4
Yes, in the next 1 to 2 years 74.1% 20
Yes, in the next 3 to 5 years 11.1% 3
Please feel free to elaborate 9
Table 14 - Plans to review forecasting methodology
Top three issues in current forecasting methodologies
The aspects that most respondents ranked among the top 3 of issues to tackle to improve the forecast method are:
1. Input from external sources (such as economic growth, population, etc.);
2. Measurement data;
3. Input from DSO level e.g. shifting of supply of load from one location to another.
Current challenges in forecasting
Table 15 shows how respondents ranked the following barriers to improving forecast accuracy. Regulatory stability
was ranked the highest indicating that changes is in regulation may have a significant effect on load and energy
forecasting methodologies.
Please rank in relative order which barriers are most important to overcome to
improve the accuracy of forecasts
Answer Options
Rating
Average
Response
Count
regulatory stability 4.55 20
climate data 4.32 25
IT software & databases 4.19 21
internal human resources 4.13 23
input from DSO level 3.64 22
improve measurement data 3.21 24
input from external sources, such as economic
statistics (correlation to GDP, population growth, …)
2.74 23
Table 15 – barriers to forecast accuracy improvement
The impact of temperature is well incorporated on local (34%) and/or national (31%) level, as is the penetration of
renewable energy (34% both on a local level and a national level).
Future challenges in forecasting
Table 16 shows how respondents ranked the most important aspects of their forecasts that need to change or be
incorporated in the next 10 years. The result is not surprising as the aspects ranked most important are
comparatively new compared with the lower ranked aspects (which are likely better understood and modelled in
current forecasting methodologies).

Page 30
What do you believe to be the three most important aspects of your forecasts
that you will need to change or incorporate in the next 10 years to improve
your forecasts?
Answer Options
Response
Percent
Response
Count
penetration of renewable energy sources (RES) 66.7% 20
demand side response management 46.7% 14
electric vehicles 43.3% 13
storage 43.3% 13
electric efficiency 33.3% 10
air conditioners 20.0% 6
temperature 20.0% 6
heat pumps 3.3% 1
answered question 30
skipped question 4
Table 16 - Improvements to forecasting methodologies
The most important influences to incorporate in the load forecasts in the next 10 years are:
1. Penetration of RES;
2. Demand side response management;
3. Storage and electric vehicles.
The open question on the future challenges of load forecasting confirmed this: the topics that were mentioned most
often were:
1. RES, electric vehicles and storage/batteries;
2. Demand side response and customer behaviour in general;
3. The impact of technological innovation.

Page 31
Chapter 5 Discussion
Introduction
This chapter discusses the survey and future challenges in load and energy forecasting.
Best Practice
Best practice is a technique or method that consistently yields results which are superior to those obtained by other
means. This section examines relative best practice (or lack thereof) based on survey responses. The survey had a
relatively small sample size and did not have responses from many countries. It is acknowledged that better
practices may exist in utilities or other organisations that did not respond to the survey. The following discussion
cannot be construed as absolutely representing best practice in load and energy forecasting.
Combination of top down and bottom up approaches
Most respondents used a combination of top down and bottom up approaches to forecasting. This combination
gives the benefit of forecasting at the global level and informing that forecast with detailed information about what
will happen with loads at different parts of the network.
Forecast Methodology
Most respondents reviewed their methodology in the last two years and were planning to review their methodology
again in the next two years. This suggests that best practice for forecasting methodologies has not been settled on
or that the components (e.g. RES models) are changing and are requiring frequent updates to forecasting
methodologies.
Regulators do not in most cases prescribe a forecast methodology. This further supports the suggestion that best
practice in forecasting methodologies is not agreed. If there were a single forecasting methodology that produced
superior forecasts to other means, then it is likely that most regulators would prescribe this methodology. Best
practice in forecasting methodologies is an area for further investigation.
Reactive power forecast
Most respondents did not explicitly forecast reactive power. Reactive power is forecast using historic or assumed
power factors. The nature of load is changing (for example the increase of inverter connected loads) and the long-
standing assumption that reactive power consumption increases as real power increases is becoming weaker than
in the past. Inverters have the ability to absorb and export reactive power in a controlled manner which needs to be
accounted for in demand forecasting.
Forecasting software and tools
Most organisations have developed their own forecasting software and tools in house. This could be due to a lack
of suitable commercial software and systems or a desire of forecasters to have software and tools which they
understand in detail.
Load modelling
More than half of the respondents categorize load by type. For more than half of them (62%) the categories are
high-level, namely “residential”, “industrial”, or “commercial”. A small group (14%) considers more detail, such as
categories per sector. The best practice based on usage seems to be to employ forecasting that has a number of
components (with different models) when preparing forecasts.
Long term forecasts
Most long term forecasters use time horizons of 5 to 15 years and a few have time horizons of up to 30 years. The
5 to 15-year time horizon seems appropriate given the future changes in technology (e.g. widespread RES) will
require changes to forecasting methodologies.

Page 32
Future challenges
RES, electric vehicles and storage/batteries are identified in the survey as one the most important future
challenges of demand forecasting. RES, often having variable output, makes forecasting harder especially in the
short term. Distributed RES is harder to monitor. Large scale RES production will require more complex methods
and data to forecast. These models will probably use more metrological parameters and a number of different
measurement data from various measuring stations in a wide area.
Individual electric vehicles can appear as loads on different parts of the power system as drivers travel and charge
their vehicles at different places. The widespread use of electric vehicles may lead to a significant change in the
existing load duration curves as owners take advantage of special price tariffs (e.g. owners are incentivised to
charge their vehicles overnight). However, a significant increase in the use of electric vehicles will take time and it
is expected that it will be possible to establish certain mathematical relations and better understand the behaviour
of electric vehicles users in terms of electricity consumption.
Energy storage is very beneficial in enabling development of a low-carbon electricity system. It also provides
flexibility and balancing to the grid as a backup to intermittent renewable energy. These systems can improve the
management of distribution networks, reducing costs and improving efficiency. In this way, well placed energy
storage will ease the market introduction of renewables, accelerate the decarbonisation of the electricity grid,
stabilise market prices and improve the security and efficiency of electricity transmission and distribution networks.
In the context of future challenges for the forecasting of power system load, it is clear that energy storage has
some influence, either directly or indirectly, and requires some development of appropriate mathematical models.
Active load
Some load and energy forecasting methodologies make the assumption that load cannot respond to conditions on
the power system. The load is assumed to be passive. Increasing amounts of active load which does interact with
the power system are appearing.
Demand Side Response (DSR) consists of a set of techniques, policies and market arrangements which are
designed to manage loading on the power system. In accordance with the basic postulates related to DSR, it is
clear that the future models for forecasting will be required to take into account certain social and economic
indicators, the customs of the population, demographic indicators, etc.
The rise of distributed smart internet-connected energy devices (e.g. RES, electric vehicles, batteries) which
operate in a coordinated manner will enhance the capability of DSR. Distributed smart load will react in very fast
time frames in response to conditions on the power system such as high electricity prices. Distributed smart load
may provide ancillary services. Load and energy forecasting methods should incorporate this response within
forecasts.
Power system load forecasters will also need to review current forecasting techniques and decide if they are fit to
tackling the challenges set out above. Will new techniques be required to deliver the required results in the future?
Paradigm shifts in demand forecasting
Figure 9 shows how the future supply chain of electricity might be influenced by growth of RES, and other forms of
power. The target market, market and energy regulators applicable to each country will hugely influence this model.
However, it is important to understand the implication and the purposes of the different forecasts associated with
such a supply chain.

Page 33
Figure 9: Future supply chain in electricity industry
The electricity supply chain functions the same irrespective of the functions of utilities within a country. There are
Generators, Transmission, Distribution and end consumers in every network. However, some of these functions are
separated into Generators, IPPs, TSOs and DSOs etc.
In future networks, RES, Embedded Generation and sources other than the conventional generation will be
incorporated into a more complex supply and demand model. This will be quite a paradigm shift for most
companies and the way networks are designed and refurbished will be carefully analysed and adapted accordingly.
As can be seen in Figure 9, different inputs on the supply chain in the separate stages will influence the supply and
demand equilibrium. The forecasts applicable to each stage can be seen in the diagram, as Sales and Revenue
forecasting will aid in generation production forecasts.
When considering an example of a country where the current supply is mostly delivered by the utility itself and then
distributed to the customers (Direct customers, TSOs or DSOs), Figure 10 shows how the profile for such a country
might change with penetration of RES and Embedded Generation sources.
Figure 10 shows how a single supplier, with limited to no alternative energy generation sources, needs to undergo
a paradigm shift towards an integrated supply system where RES, utility generation as well as customer self-
generation should be modelled to get the optimal network adequacy and capacity planning done.
This will again lead to the importance of different forecasts needed for different dimensions of a targeted network.
The paradigm shift of moving from a sole provider or distributor to a market player within the holistic supply and
demand model of electricity markets will affect how forecasting is carried out as there are now multiple parties with
information relevant to forecasts. There are still vast amounts of uncertainty involved in this regard, but as
highlighted earlier it is of utmost importance that the target market and paradigm in which your utility falls is
identified, analysed, and correctly applied for optimal network adequacy planning.

Page 34
FIGURE 10: COMBINATION OF RENEWABLE ENERGY SOURCES AND OTHER GENERATION
TECHNOLOGIES INFLUENCE PARADIGM SHIFT OF UTILITIES
Other Techniques
Other potential techniques, which may enhance future load forecasting:
 Multivariate analysis;
 Object oriented modelling;
 Multi-criteria decision analysis;
 Advanced data analysis techniques;
 Operations research – specifically to develop the load and energy forecast as a process;
 Spatial analysis to deal with complex studies;
 Business processes to control the input and output activities of the load and energy forecast process.
Summary
There are many challenges that must be met by load forecasters in the future to ensure efficient, reliable and
secure power system operation. Many of the future drivers of load and energy growth are inherently uncertain and
hence load forecasters must develop management strategies for this uncertainty to deliver robust planning
decisions while identifying key factors.
Load and energy forecasting processes and tools will need to evolve to co-ordinate input from a diverse range of
experts, for example industrial engineers, political operation research practitioners, statisticians, GIS (Geo-based
information systems) specialists, mathematicians, electrical engineers, market intelligence specialists, etc.

Page 35
Chapter 6 Conclusions and recommendations
WG C1.32 carried out a survey of international utilities to establish best practice approaches for developing
electricity load and energy forecasts at the end of 2015.
The survey results show that forecasts are mainly used for long-term grid planning (23 cases; 68% of the survey).
In 40% of these 23 cases the responses also apply for short-term operational planning. In 18% of the cases, the
forecast was also used for generation development; in 21% also for security of supply; and in 29% also for
generation adequacy studies. The timing of the peak load is evenly spread: 50% of the respondents work in a
country or region with a peak load in winter; 44% in summer.
In most cases both load and energy are forecast. Thirty six percent of the forecasts were for load only. No forecasts
focused solely on energy. Almost all forecasts are required by regulation, yet mostly the methodology is not
prescribed. Most respondents used load and energy forecasting software that was developed in-house.
Most respondents do not forecast reactive power as such although occasionally reactive power is indirectly
forecast based on assumed or historic values of power factor.
The forecast methodology is frequently revised. Most respondents revised it in the last 2 years (53%) and of this
group almost all are also planning to revise again in the next 2 years (88%). For those who reviewed the
methodology more than 2 years ago, most also plan to revise again in the next 2 years.
Given that forecast methodologies are frequently reviewed, it seems that best practice in forecast methodologies is
not widely agreed (at least amongst the survey respondents). Some aspects where there are common approaches
are the modelling of load by type (e.g. residential, commercial, industrial), using a combination of top down and
bottom approaches to forecasting and forecast horizons (e.g. most long term forecasts look ahead 5 to 15 years).
Most forecasting teams have a small number of people to carry out the forecasts and prefer customised in house
tools for forecasting. It is likely that many forecasting teams are highly dependent on the expertise and skills of a
few individuals and these individuals have limited ability to share knowledge with each other. There is an
opportunity for CIGRE to take a lead here.
The most important aspects to improve in the forecast method are, according to the respondents:
 Input from external sources (such as economic growth, population, etc.);
 Measurement data;
 Input from DSO level.
The survey responses indicate the most important aspects needing to be incorporated into load forecasting in the
next 10 years are:
 Penetration of Renewable Energy Sources (RES);
 Demand side response management;
Suggestions for future CIGRE work

Page 36
The Working Group identifed a large number of areas where further work could be undertaken by CIGRE. We
decided to focus on two areas for immediate work.
Survey of the capabilities and performance of inhouse load and energy forecasting tools
This work stream follows on from this working group. The work is a survey of the capabilities and performance of
demand forecasting tools. The survey will include questions on:
 Forecasting methodologies incorporated in the forecasting tools.
 Developed models for electric vehicles, storage, RES, demand side management etc.
 Accuracy of the forecasts.
This work stream is an input into the next work stream.
Best practice models for load and energy forecasting.
 Penetration of Renewable Energy Sources (RES).
 Balancing Supply and Demand Models in new Energy mix.
 Demand side response management.

Page 37
Bibliography/References
[1] Levenbach, Hans.; Clearly, James P.: Forecasting Practices & Process for Demand
Management, 2006.
[2] Payne, D.F.; “Modelling of different Long-Term Electrical Forecasts and its Practical
Applications for Transmission Network Flow Studies”, Rand Afrikaans University, 2004.

Page 38
Appendix 1 - Summary of survey answers
Cover letter.
Dear respondent,
CIGRE Working Group C1.32 aims to establish best practice in how load forecasts are determined today and the
challenges the forecasters face today and expect to face in the future. The terms of reference of WG C1.32 can be
found here.
We are primarily seeking responses from parties who prepare load and energy forecasts for transmission
purposes. However, other parties are welcome to respond to the survey if they have an interest in load forecasting.
The survey's purpose is to collect information regarding load forecasts, be it for load or energy, for short-term
operational planning or for long-term grid development. It is therefore also possible to fill in more than one survey
per organization (e.g., 1 for long term forecasting and 1 for short-term forecasting).
Mandatory questions are indicated with an asterix (*).
The time to fill in the survey is about 20 minutes. We would appreciate receiving your responses by 30/11/2015.
The survey is split into 10 sections
 General information
 Electrical network characteristics
 Purpose of the forecast
 Forecast properties
 Data collection
 Forecast methodology and tools
 Use of production information
 Forecast accuracy and revisions
 Future challenges in forecasting
 Other remarks
We would like to thank you in advance for your collaboration.
If you have questions about the survey please email Graeme Ancell (Working Group Convener)
Kind regards,
CIGRE Working Group C1.32.

Page 39
Question 1
What is the name of your company?
Answer Options
Response
Count
34
skipped question 0
Question 2
What type of company do you work for?
Answer Options
Response
Percent
Response
Count
Independent System Operator 17.6% 6
Transmission grid owner 23.5% 8
Integrated transmission grid owner and system
operator
35.3% 12
Distribution 8.8% 3
Regulator 0.0% 0
Generation 0.0% 0
Vertically integrated generation and transmission 14.7% 5
Other (please specify) 4
skipped question 0
Other: Vertically integrated distribution & transmission network company
Question 3
What is your company's country or regulation zone (spanning
multiple countries)?
Answer Options
Response
Count
34
skipped question 0
Question 4
Could you provide us your contact information in case we need to contact you to
clarify a response?
Answer Options
Response
Percent
Response
Count
Name: 100.0% 30
Email Address: 100.0% 30
skipped question 4

Page 40
Question 5
In your country or regulation zone, who does the electricity demand forecasting that
you use?
Answer Options
Response
Percent
Response
Count
My own company 76.5% 26
Transmission System Operator 35.3% 12
Distribution System Operator 20.6% 7
Regulator 2.9% 1
skipped question 0
Other: Outsourced to consultants, regulatory bodies, other groups in organisation.
Question 6
What is the scope of your company's responsibilities in the electrical cycle from
generation to consumption? (multiple answers possible)
Answer Options
Response
Percent
Response
Count
Transmission of electricity - grid owner and operator 79.4% 27
Transmission of electricity - system operations 52.9% 18
Generation of electricity 14.7% 5
Storage 2.9% 1
skipped question 0
Other: Distribution
Question 7
When is the peak load in the country or zone for which you are filling in this survey?
Answer Options
Response
Percent
Response
Count
Winter 44.1% 15
Summer 50.0% 17
Autumn/Spring 0.0% 0
No seasonal difference 5.9% 2
skipped question 0

Page 41
Question 8
How would you describe the purpose of the forecast for which you fill in this survey?
Answer Options
Response
Percent
Response
Count
Short-term operational planning (dispatching,
outages, maintenance, ...)
55.9% 19
Long-term grid development 70.6% 24
Long-term generation development 17.6% 6
Security of supply 20.6% 7
Generation adequacy 29.4% 10
Please use this box to add other purposes or explain
more about the answers given above.
14.7% 5
skipped question 0
Other: sales/revenue forecasting
Question 9
Please feel free to provide a short description of the purpose of the
forecast:
Answer Options
Response
Count
22
skipped question 12
Purpose of the Long Term Demand Forecast is to prompt the Transmission Grid Planning department to plan for a
developing country and the potential demand that the country might need going into the future in order to enable
economic growth and development in the country.
Demand forecasts from <company name> are predominantly for long-term grid development. May be used for
long-term system adequacy/generation development.
dispatching
There are three main purposes of the forecast: 1. Dispatching (short-term planning) 2. Maintenance of the
equipment (short-term and medium-term planning) 3. Grid and generation development (long-term planning)
The LF is a key input into the System Operators scheduling, pricing and dispatch software and is essential to
produce forward looking schedules of generation quantities/prices etc.
Long-term grid development
Inform network planners of load requirements as an input to investment decisions Provide short term forecasts to
inform contingency plans for outages
There are multiple forecasts that model maximum demand trend, monthly energy volumes plus SAIDI & SAIFI by
feeder category
Used to plan the transmission network. (110kV, 132kV, 275kV and 330kV)
maximum demand forecast primarily for network planning purposes but also used for seasonal readiness planning

Page 42
The load forecast is mainly used for long term planning and project evaluation.
Annual energy input to market models and CBA. Intra-day load profile to capture swing trade between countries
and price areas
The long-term load forecasts are used to identify investment needs in the grid: at the local level of a HV-MV
transformation station (TFO capacity to be expanded) or at the level the HV grid (expansion of TFO, cable, lines,
interconnection, ...)
Short term: used in short (three days horizon) and mid term (one week horizon) unit commitment, performance
evaluation process of demand forecasting , security of service and reliability studies and to update the short term
public database developed in PowerFactory (DIgSILENT). Long term: used to update the long term database
developed in Plexos and PowerFactory (DIgSILENT), integration studies of renewable energies, security of service
and reliability studies, transmission planning, development of generation major maintenance program.
We need load forecast for 1: grid extention planning (thus long term) and 2: the assessment of the Generation
adequacy and security of supply. That are two different exercises.
As mentioned in item 8
As a planner, i need to know the difference between generation capacity and load demand in a bulk power system.
The forecasts prepared for local system planning are joint initiatives between the System Planners (us) and
Distributors. Station level forecasts are generally provided by the Distribution companies, as they have the closest
relationship with end use customers and local planning offices. We work with the Distributors to ensure appropriate
and consistent assumptions when merging these forecasts to cover a larger geographic area for transmission
adequacy assessments. We also integrate other consideration, such as conservation targets, which are prepared
on a provincial level.
use in the process of establishing the power development plan and the transmission expansion plan
Determine the ability of the grid to meet future demand, justify future investments
dispatching
The LF is a key input into the System Operators scheduling, pricing and dispatch software and is essential to
produce forward looking schedules of generation quantities/prices etc.
Question 10
What is the relationship between forecasting data used for short-term operational
planning and long-term grid reinforcement?
Answer Options
Response
Percent
Response
Count
Use of the same historic transmission measured
data
58.1% 18
Use of the same data provided by generation data 16.1% 5
Use of the same data provided by connected parties
(e.g. distribution company)
19.4% 6
Same level of analysis (number of system nodes,
substations, ...)
22.6% 7
Similar forecasting methodology 9.7% 3
No relationship 25.8% 8
Please feel free to provide information in addition or explanation of
the answers above
12
skipped question 3

Page 43
Question 11
Is the forecast a regulatory requirement?
Answer Options
Response
Percent
Response
Count
Yes 90.0% 27
No 10.0% 3
Feel free to explain 11
skipped question 4
Question 12
Is the forecast methodology prescribed by regulation?
Answer Options
Response
Percent
Response
Count
Yes 3.2% 1
No 96.8% 30
Feel free to explain 11
skipped question 3
Question 13
Do you forecast load or energy (as the output of the forecast)?
Answer Options
Response
Percent
Response
Count
Load only 36.4% 12
Energy only 0.0% 0
Load and energy 63.6% 21
skipped question 1

Page 44
Question 14
Please indicate the forecasted moment in time, the granularity, and the horizon of the forecast?(e.g., forecast of the
peak (= moment) during the year (= granularity) for the next 10 years (= horizon))
Forecasted moment
Answer Options Peak Minimum Average
Other
times of
interest
Response
Count
Forecast 1 28 0 3 2 33
Forecast 2 4 7 4 1 16
Forecast 3 2 3 4 0 9
Forecast 10 0 0 0 0 0
Granularity
Answer Options per hour per day per week
per
month
per year
Response
Count
Forecast 1 8 3 1 1 20 33
Forecast 2 2 5 1 2 6 16
Forecast 3 2 0 1 3 3 9
Forecast 4 2 1 0 0 1 4
Forecast 5 1 0 0 0 1 2
Forecast 6 0 0 0 1 0 1
Forecast 7 0 1 0 0 0 1
Forecast 8 0 0 0 0 0 0
Forecast 9 0 0 0 0 0 0
Forecast 10 0 0 0 0 0 0
Horizon
Answer
Options
1 day
ahead
1
week
ahead
1
month
ahead
1 year
ahead
[number
of] days
ahead
[number
of]
weeks
ahead
[number of] months
ahead
[number
of]
years
ahead
for specific
days (e.g.
Christmas)
for
specific
periods
during
the
year
(e.g. a
season)
for
specific
years
Response
Count
Forecast
1
5 2 0 0 2 0 0 22 0 0 2 33
Forecast
2
3 1 0 1 1 0 1 7 0 0 2 16
Forecast
3
0 1 3 1 0 1 0 2 0 0 1 9
Forecast
4
1 1 0 0 1 0 0 0 0 0 1 4
Forecast
5
1 0 0 1 0 0 0 0 0 0 0 2
Forecast
6
0 0 1 0 0 0 0 0 0 0 0 1
Forecast
7
1 0 0 0 0 0 0 0 0 0 0 1
Forecast
8
0 0 0 0 0 0 0 0 0 0 0 0
Forecast
9
0 0 0 0 0 0 0 0 0 0 0 0
Forecast
10
0 0 0 0 0 0 0 0 0 0 0 0

Page 45
Question 15
On which geographical or client level do you do the forecast?
Answer Options
Response
Percent
Response
Count
national level 45.5% 15
regional level 48.5% 16
substation level 54.5% 18
client level 18.2% 6
voltage level 6.1% 2
skipped question 1
Other:
Feeder level (amps) for peak demand forecasts.
By customer type, load area and network wide.
Question 16
Do you incorporate a spatial forecast (forecast of where new significant loads e.g.
factory, mine might connect to the grid or where electric vehicles charge at different
times of the day) in your forecasts? A spatial forecast might include probabilities of
the load connecting at different locations.
Answer Options
Response
Percent
Response
Count
Yes. 66.7% 20
No. 33.3% 10
skipped question 4
Question 17
What tools and methodologies do you use for spatial planning?
Answer Options
Response
Percent
Response
Count
None 44.4% 12
Tools and methodology developed in house 55.6% 15
skipped question 7

Page 46
Question 18
Do you forecast reactive power? If so, could you elaborate on the purpose and the
methodology?
Answer Options
Response
Percent
Response
Count
yes 39.4% 13
no 60.6% 20
Comment (please specify) 15
skipped question 1
Power factors only
the main purpose is to determine whether the sources of reactive power are needed or not
in relation to active power
We forecast reactive power for planning of shunt reactor and shunt capacitor.
Reactive power forecasts are essential for identifying voltage limitations and, to a lesser extent, thermal limitations.
By connection point
Reactive power forecast is simply an extension of the most recent reactive power at times of peak, relative to the
change in peak. There is no actual forecasting of reactive component of new loads.
not specifically forecast, but output of forecast procedure
Generate reactive power requirements based on historic power factor and the maximum demand forecast
Currently, reactive power is indirectly forecasted but the method is much less advanced than for active power. We
are currently working at a specific method to forecast reactive power in the future.
NOT FOR SHORT TERM
For long term planning the DSO specifies the cos phi per connection point. From this we derive the reactive power
needs.
Reactive power is usually modeled assuming a constant power factor consistent with historical observed values. In
some cases (such as major anticipated shift in industrial customers) we may consider changing the value in our
models, but this would be in response to a specific event, and not part of regular planning.
Using historic power factor
Question 19
What is the source of the measurement data you use?
Answer Options
Response
Percent
Response
Count
SCADA (not validated, not calibrated) 60.0% 18
Revenue metering (validated and calibrated) 80.0% 24
skipped question 4

Establishing best practice approaches for developing credible electricity demand and energy forecasts for network planning

Establishing best practice approaches for developing credible electricity demand and energy forecasts for network planning

Recommended

Recommended

More Related Content

Similar to Establishing best practice approaches for developing credible electricity demand and energy forecasts for network planning

Similar to Establishing best practice approaches for developing credible electricity demand and energy forecasts for network planning (20)

More from Power System Operation

More from Power System Operation (20)

Recently uploaded

Recently uploaded (20)

Establishing best practice approaches for developing credible electricity demand and energy forecasts for network planning