Enhancing the Electric Grid Reliability through Data Analytics
1. Enhancing the Electric Grid
Reliability through Data
Analytics
Mrinalini Sharma
7/23/2017
2. GTx Introduction to Analytics Modeling 2017
Abstract
The Midwest Independent Transmission System Operator (MISO) was the nation’s first
approved Regional Transmission Organization (RTO) to create an operations solution that
would enhance the electric production and distribution system by integrating the region’s
utilities into one system. This paper attempts to propose a solution similar to the one
implemented by MISO in 2009 for the day-ahead market using basics of data analytics and
modeling. The real-time market is outside the scope of this paper. The models proposed
will need to be trained, tested and validated on actual data and adjusted to reflect the
actual daily operations of the electric grid system.
Keywords
Data Analysis, Clustering, Linear Regression, Auto Regressive Integrated Moving Average
(ARIMA), Optimization, Reliability, Energy Efficiency, Operations Research
Background
For many years, electric utilities had a localized focus in serving their customers.
Utilities served electric demand in their service territory with their own power plants and
transmission lines. This led to each utility maximizing its power plant and transmission line
usage at the expense of the larger electric system. In addition, the utilities had to hold
enough capacity so as to manage their operational risk. Therefore, the utilities were
optimizing their assets to serve the territory’s demand while carrying extra investment to
manage the risk.
In the 1990s, the Federal Energy Regulatory Commission (FERC) pushed to reduce
the individual utilities’ market power and encouraged the formation of RTOs that could
ensure that power travels to where it is needed at the lowest possible cost with the highest
level of reliability, considering the physical constraints of the electricity system.
Regulations were passed to allow open access to the transmission system that further
supported the broad wholesale power competition.
3. GTx Introduction to Analytics Modeling 2017
MISO used Operations Research (O.R.) methods to design and launch the energy and
ancillary services market in 2009. It is a commodities market that facilitates real-time and
day-ahead trade transactions in energy and ancillary services. The O.R. products allowed
MISO to improve reliability and to increase efficiencies in the utilization of power plants
and the transmission grid assets while also reducing the need for future assets.
The utilities participating in MISO hold the physical control of the power plant and
the transmission lines. Functionally, however, these are controlled by MISO which sends
dispatch signals to the utilities to tell them how much power to produce and when to
produce. MISO manages the real-time balancing of energy demand and supply by
coordinating supply offers and demand bids received via a process called market clearing.
Supply Offers reflect the production capabilities and the cost to produce of each power
plant. Demand Bids reflect the projected consumption quantities and the willingness to pay
price of each demand center. MISO determines which supply offers and demand bids are
accepted and at what price to ensure that power gets delivered to where it is needed at the
lowest cost while ensuring the highest level of reliability. Reliability is heavily stressed here
to ensure that the transmission lines don’t get overloaded and no equipment is damaged.1
Research Question
How can, the market clearing process, be optimized to ensure that the power gets delivered
to where it is needed at the lowest possible cost, while ensuring highest reliability?
Proposed Solution
The solution is divided into three parts:
1. Identification of the power plants that are scheduled to be running the following day
2. Determination of the projected consumption quantities as well as the On-Peak and
Off-Peak demand for each demand center on the following day.
3. Sequencing the available power plants to ensure that the power is delivered at the
lowest possible cost while maintaining the reliability.
Solution Format: Given {data}, use {model} to {result}
The model will need be refreshed and re-run every day for the day-ahead projections.
1
Brian Carlson, Yonghong Chen, Mingguo Hong. (2012) MISO Unlocks Billions in Savings Through the
Application of Operations Research for Energy and Ancillary Services Markets
4. GTx Introduction to Analytics Modeling 2017
PART 1
Identification of the power plants that are scheduled to be running the following day
Given: ‘On’ and ‘Off’ commitment data from each power plant
If a certain power plant, say Pi, will be ‘On’ at least partly during the following day, then
Pi = 1, else 0
Use: Clustering
To: Identify clusters of power plants that will be ‘On’ during the following day
PART 2
Determine the projected consumption quantities as well as the On-Peak and Off-Peak
demand for each demand center for the following day
Approach 1
Given: Previous energy use and demand data from each demand center
a. Daily energy use data
b. On-Peak Demand data (15 min or 5 min interval)
c. Off-Peak Demand data (15 min or 5 min interval)
Use: ARIMA (Auto-regressive Integrated Moving Average)
To: Predict the energy consumption quantity and the On-Peak as well as Off-Peak demand
of each demand center based on the historical energy use and demand data.
Approach 2
Given:
a. Weather data (Dry bulb and Wet bulb Temperature data)
b. Production data for industrial facilities
c. Operation data, like hours of use, for commercial and residential facilities
Use: Multivariate Linear Regression
To: Predict the energy consumption for each demand center
Verify the results of ARIMA against those from the Linear Regression and make changes to
the models accordingly. Take the more conservative energy consumption amount
predicted from the two models to protect against any operational risk.
5. GTx Introduction to Analytics Modeling 2017
PART 3
Determine the sequencing amongst the available power plants to ensure that the power is
delivered at the lowest possible cost while maintaining the reliability.
Given:
1. Supply Offers from each power plant - the amount of energy each power plant can
supply and the cost at which it can supply during On-Peak as well as Off-Peak hours
2. Projected energy consumption amounts from each demand center during On-Peak
and Off-Peak hours
3. Physical constraints of the transmission lines (for example, power must be supplied
at 60 Hz)
Use: Optimization
To: Minimize the cost of power delivery while ensuring that the transmission lines are not
overburdened.
Optimization Function
Variables:
Xip = Amount that Power Plant i can provide during peak hours p
Xio = Amount that Power Plant i can provide during the off peak hours o
Yip = Cost at which Xip can be provided
Yio = Cost at which Xio can be provided
Dip = Energy Demand from Demand Center i during peak hours p
Dio = Energy Demand from Demand Center i during off-peak hours o
Constrains:
∑Xip = ∑Dip
∑Xio =∑ Dio
Xi >= 0 (for both peak and off-peak hours)
Di >= 0 (for both peak and off-peak hours)
While the constraints for the optimization problem will include the physical constraints of
the transmission grid assets, I am not completely equipped to translate those into
mathematical equations.
6. GTx Introduction to Analytics Modeling 2017
Objective Function:
Min ∑Yip
Min ∑Yio
The objective function is going to ensure that first the more-efficient power plants are
brought online in order to minimize the associated cost. This will lead to the minimization
of not only financial costs but also environmental and other societal costs.
An underlying assumption of this optimization algorithm is that transmission charges for
power delivery are fixed and/or insignificant and are not affected by the distance that
energy has to travel to reach the end use customer. If, however, this is not the case,
constraints can be added to the optimization model to minimize the total cost of production
and distribution.
Conclusion
This paper presents a combination of analytical models that can be applied to the supply
and demand data of the electric industry to create an optimal solution with the least
societal costs and high reliability. The models described in this paper are purely from a
conceptual stand point and will need to be tested to reflect the actual daily operations.