Dayananda Sagar College of Engineering hosted a technical seminar on using artificial intelligence to support the integration of variable renewable energy sources into the power system. The presentation discussed how (1) AI can improve wind and solar power forecasting to reduce balancing costs, (2) demand response programs and storage solutions enabled by AI can help mitigate profile costs by increasing grid flexibility, and (3) AI applications like predictive maintenance can help lower grid-related infrastructure costs. While uncertainties remain regarding AI's value, the literature indicates it has potential to decrease renewable energy integration costs and thereby increase the economic viability and adoption of green energy sources.
ARTIFICIAL INTELLIGENCE TO SUPPORT THE INTEGRATION OF VARIABLE RENEWABLE ENERGY SOURCES TO THE POWER SYSTEM
1. DAYANANDA SAGAR COLLEGE OF ENGINEERING
An Autonomous Institute affiliate to VTU
Accredited by National Assessment & Accreditation council (NAAC) with A grade
Department of Electrical and Electronics engineering
A Technical Seminar on
ARTIFICIAL INTELLIGENCE TO SUPPORT THE INTEGRATION OF VARIABLE RENEWABLE
ENERGY SOURCES TO THE POWER SYSTEM
Submitted by
MURALIDHAR R
Electrical and Electronics Engineering
2. Introduction
The power sector is increasingly relying on variable renewable energy sources (VRE) whose share in
energy production is expected to further increase.
Artificial intelligence (AI) solutions and data-intensive technologies are already used in different parts
of the electricity value chain and, due to the growing complexity and data generation potential of the
future smart grid, have the potential to create significant value in the system.
However, different uncertainties or lack of understanding about its impact often hinder the commitment
of decision makers to invest in AI and data intensive technologies, also in the energy sector.
The goal of this article is to consider the value creation potential of AI in terms of managing VRE
integration costs. We use an economic model of variable renewable integration cost from the literature
to present a systematic review of how AI can decrease substantial integration costs.
Decarbonization has become a major issue, as climate change discussions and related policies also
indicate. To fulfil the increasing demand for energy in a sustainable way, there is a fundamental need for
greener and cheaper energy. Mainly due to renewable energy sources (RES), electricity has the potential
to become a key green and relatively cheap fuel of the future.
3. Global electricity generation from VRE (part of RES), such as wind and solar photovoltaics (PV) has
increased from 32 TWh to 1.857 TWh between 2000 and 2019.
The share of renewables in global electricity generation was near 28% in the first quarter of 2020, and
in several countries renewables have reached considerably higher penetrations. The penetration is
expected to significantly increase mainly due to solar PV and wind.
The integration costs reduce the business feasibility and economic viability (market value) of VRE
and increase System LCOE.
AI generally includes not only machine learning but also other tools, for example rule based systems
with rules possibly being hand-written instead of learned from data.
AI and data intensive technologies are already used in different parts of the electricity value chain and,
have the potential to create significant value in the system.
4. THE ROLE 0F AI TECHNOLOGY IN IMPROVING THE
RENEWABLE ENERGY SECTOR
Global energy demands are growing every year. And, fossil fuels won’t be able to fulfill our energy
needs in the future. Carbon emissions from fossil fuels have already hit an all-time high in 2018 due to
increased energy consumption.
On the other hand, renewable energy is emerging out as a reliable alternative to fossil fuels. It is much
safer and cleaner than conventional sources. With the advancements in technology, the renewable energy
sector has made significant progress in the last decade.
However, there are still a few challenges in this sector that can be addressed with the help of emerging
technologies.
● Technologies like AI and Machine Learning can analyze the past, optimize the present,
and predict the future. And, AI in the renewable energy sector can resolve most of the challenges
5. A FRAME WORK OF INTEGRATION COST OF VRE
There is a comprehensive literature on modelling the costs related to VRE integration and, with the
significant increase in penetration of VRE sources, there is also an increasingly better understanding of
the main cost drivers .
AI can impact investment costs, operation costs and integration costs simultaneously.
Three key VRE characteristics largely determining integration costs are uncertainty, variability and
location specificity.
The cost impact of uncertainty is being referred to as “balancing costs”, that of temporal variability as
“profile costs” and that of location as “grid-related costs”.
Uncertainty relates to the deviation between the forecasted VRE generation and the actual production,
which needs to be balanced in a short amount of time. Thus, balancing costs can be considered as
affecting the economic viability of VRE due to short-term deviations from generation schedule.
One reason for these deviations relates to supply forecast errors
6. Forecast errors depend on the installed VRE capacity, the forecasting tools, the time horizon, the type of
VRE technology and the geographical spread of VRE generation plant.
VRE costs are also variable because electricity is generated only under certain weather conditions
without necessarily always matching demand.
In case VRE sources are not able to fulfil demand, the output of conventional generation capacities will
have to flexibly adapt, ramping up and down possibly frequently.
The largest single factor of VRE integration costs are profile costs mainly due to the long-term reduced
utilization of capital invested in conventional generation capacities. Profile costs are estimated to be 15–
25 EUR/MWh at 30–40% penetration.
Finally, integration costs also arise because VRE sites are often located far from where electricity
demand occurs, while the increasing share of VRE in the grid can lead to power quality events. This
leads to grid-related costs as the transmission of electricity through the grid has high infrastructure costs,
and electricity is lost while being transmitted. Grid-related costs are anticipated to be in the single-digit
range of around 0 EUR/MWh–10 EUR/MWh.
Managing integration costs will become increasingly important as VRE penetration increases.
7. AI fostering the integration of VRE
Mitigating balancing costs
Supply and demand must equal at every moment in the electricity grid, however deviations from
contracted positions occur because of the short-term uncertainty related to VRE generation.
Today AI is widely used to support VRE generation forecasting [34–41], demand forecasting [42,43],
and more efficient balancing markets.
1.Generation forecasting
The value creation potential of wind and solar PV generation forecasting has been presented by
scholars mainly through model-based simulations and by companies through use-cases.
Machine learning has also been used to combine multiple meteorological models to improve the
accuracy of solar/wind power generation forecasts.
The machine-learning based model blending approach was shown to reduce “localized” error of the
individual models, indicating repeatedly over 30% improvement in solar power forecast accuracy
compared to forecasts based on the best individual meteorological model.
8. Wind generation forecasting can also be improved by AI, despite being more advanced compared to
solar power forecasting, since it uses similar techniques to meteorological forecasting.
2.Demand forecasting
Together with generation forecasting, demand forecasting is also crucial to balance the electricity grid.
Mainly due to the global deployment of smart meters, available data related to power consumption has
significantly increased helping to improve demand forecasting.
For a national or regional geographic context linear models were shown to be more
efficient than non-linear ones, with an average error of 2.04% for linear, 3.20% for
Artificial Neural Networks (ANN) based models and 3.14% for ANN based
hybrid models.
However at a smart city, smart grid level linear models seem to be
relatively less accurate.
9. 3.More efficient market design
Besides generation and demand forecasting, a more efficient market design can also help mitigating
balancing costs.
In European unbundled electricity markets, the market provides balancing of electricity demand and
supply in case an imbalance occurs.
AI can foster the efficiency of balancing markets. Profit-maximizing bidding strategies were shown to
reduce balancing costs by 50%.
For example, an algorithm, called EUPHEMIA, has been developed to determine energy allocation and
day-ahead electricity prices across Europe and allocate cross-border transmission capacity on a day-
ahead basis. The algorithm is used to calculate day-ahead electricity prices for 25 European countries.
10. MITIGATING PROFILE COST
Profile costs are mainly caused by the reduced utilization of costly back-up conventional generation
capacities in the long-term to compensate for the variable production of VRE.
They can be managed largely through increasing the flexibility of the power system in order to minimize
the need for setting up expensive back up energy generation systems.
1. Demand response
Demand response is increasingly used in power systems and refers to changes in end-users’
consumption patterns.
Besides influencing profile costs it also has complex effects on integration costs in general making it
also relevant for other cost components.
Digital solutions, smart infrastructure and AI are predicted to increase the volume of demand response
by 185 GW until 2040, creating an estimated value of USD 270 billion, by avoiding investments in
new electricity infrastructure such as power generation capacity, transmission and distribution.
11. AI has the potential to support demand response in several ways, such as through forecasting of demand
and future electricity prices, scheduling and control of loads at the aggregator and customer level, design
of incentive schemes, or customer segmentation.
For smart homes, Rocha et al. presented a new methodology based on AI techniques for energy
planning. This work considers electricity price fluctuations, priority in the use of equipment, operating
cycles and a battery bank and provides forecasts of distributed generation. The efficiency of the method
indicated a 51.4% cost reduction, when smart homes with and without distributed generation and battery
bank were compared.
A promising application area of demand response, also indicating the potential of these measures, is in
data centers.
They are highly automated and equipped with sensors, their IT equipment can be continuously
monitored, while many of their workloads can be programmed to finish before their required deadlines,
enabling very flexible management of power demand using innovative technologies.
12. 2.Storage solutions
In addition to demand response measures, storage solutions can become another very important source
of flexibility, with fast response time, to support the integration of VRE sources through profile costs
reduction, especially as the cost of storage technologies decreases.
Innovative storage solutions can also be enabled in the future by electric vehicles.
Electric vehicles can potentially impact profile costs, since these vehicles are both consumers of
electricity and mobile battery storage facilities, able to relatively easily change their consumption
pattern, hence also providing a special use-case for demand response.
They calculated cost savings using several techniques, showing that generally machine learning
techniques had the greatest impact, with deep neural networks providing the best performance.
In 2018, there were around 5,1 million electric vehicles globally and according to IEA [79] their number
is predicted to rise to 135–250 million by 2030.
Depending on the future number of electric vehicles the flexibility provided by “smart charging” may
save USD 100–280 billions of investment in new electricity infrastructure between 2016 and 2040.
13. AI can also help to model and accelerate the development, manufacturing and optimization of battery
systems .
Several different technologies can be used as storage solutions and in the future probably a mix of these
technologies will be deployed depending for example on the circumstances, resources available, or
economic aspects.
One advantage of battery storage systems is their fast response time, which is important when the
instantaneous optimization of variable and complex systems is required.
Mitigating grid-related costs
• Note that grid related costs differ depending on the VRE technology, as for instance building grid access
to off-shore windmills will likely have higher costs in general than connecting solar PV fields to the grid.
Power quality disturbance
The integration of VRE and distributed generation sources can create major power quality disturbances
in the power system.
VRE generation, especially solar PV, changes the behaviour of the power system from unidirectional to
bidirectional [95] and can cause different types of power quality events in the grid. AI methods can help
to improve the power quality of the grid, increasing the financial benefits for the system.
14. Predictive maintenance
Besides investment costs, scheduling maintenance or avoiding power grid failure events will be
especially crucial in VRE intensive regions where physical access to the grid infrastructure for
maintenance is also difficult.
AI can be used to optimize maintenance through better predictive maintenance solutions that can help
mitigate these grid related costs.
AI is a promising tool in many cases to both reduce and postpone grid-related investments and
maintenance costs.
15. DISCUSSION
Uncertainties and lack of understanding regarding the value creation of AI solutions are important
factors delaying or even preventing adoption of these technologies.
Although important results have been achieved using AI, further deployment of AI solutions is hindered
by a number of other factors and specificities of the power sector.
For example, risks of cyber-attacks and privacy issues raise important questions about the deployment of
these technologies.
Although arguably a very good option to integrate VRE solutions is to incorporate a diverse set of
flexibility options, depending on the system context different flexibility options will lead to different
VRE integration levels.
Nevertheless, AI solutions have been already successfully deployed all along the power sector’s value
chain and in many cases they are shown to be more efficient compared to previously used solutions,
helping reduce integration costs of VRE sources hence increasing the economic viability and adoption
potential of these energy sources.
AI based optimization can help decrease long-term integration costs of VRE in specific ways through a
number of measures and use-cases.
16. CONCLUSION
The paper is looking at the value creation potential of AI methods to support the integration of VRE into
the power sector via improving their economic and business viability.
Integration costs of VRE are not part of the generation costs, but cause additional expenses in different
parts of the power system.
Decomposing the integration costs provides a good opportunity to understand cost mitigation measures
and to identify value creation potential.
Overall, despite such challenges, the literature indicates that AI provides promising possibilities to
support VRE by reducing their integration costs in different ways. Future research on AI in the energy
sector may further improve these reductions.