Current Access Level “I” – ID Only: CUID holders and approved guests only. Building Access: Normal building operating hours with exceptions. Read more about the campus status level system and campus access information. See the latest updates to the community regarding campus planning.
Explore Our Work
Our work is committed to independent and nonpartisan research that meets the high standards of academic integrity and quality at Columbia University.
CGEP is pleased to announce a new AI & Energy series—part of our Energy Explained blog. In the first entry, the authors write about AI's potential impacts on the...
The energy transition is transforming how we power our world – clean energy systems are becoming more interconnected, automated, and reliant on digital infrastructure. But with this transformation...
Women in Energy at the Center on Global Energy Policy at Columbia SIPA is pleased to host Anne-Sophie Corbeau.
Event
• CGEP Large Conference Room
1255 Amsterdam Ave Floor 1, New York, NY 10027
About Us
We are the premier hub and policy institution for global energy thought leadership. Energy impacts every element of our lives, and our trusted fact-based research informs the decisions that affect all of us.
CGEP is pleased to announce a new AI & Energy series—part of our Energy Explained blog. In the first entry, the authors write about AI’s potential impacts on the power sector in the years ahead. Future blog posts will discuss AI and greenhouse gas emissions, road transport, nuclear power, and other topics. This post is based on research in the Artificial Intelligence for Climate Change Mitigation Roadmap (Second Edition), recently released by David Sandalow and a team of 24 coauthors.
AI’s growing power demand has received enormous attention in recent months. In many places, the lack of power supplies is an important constraint on the growth of data centers to train and run AI models.
AI’s potential to transform the power sector has received less attention. That potential is significant. AI can help accelerate the growth of renewables, improve transmission and distribution, deploy virtual power plants, revolutionize energy storage and much more. Yet a number of barriers and risks must be addressed. This blog post highlights several ways AI could transform the power sector and recommends five steps to help realize that potential.
AI in the Power Sector – Four Examples
AI has applications throughout the power sector. The examples below illustrate the range of potential applications.
First, AI can help with planning, permitting and operation of renewable power projects. Machine learning algorithms can recommend the optimal size and location of solar and wind power projects, performing complex calculations on topics such as weather patterns and grid constraints.[i] Large language models (LLMs) can accelerate permitting processes, extracting text from past permit applications to help applicants and summarizing pending applications to help permitting authorities.[ii] AI algorithms can predict solar radiation and wind speeds more accurately than traditional methods, allowing for better scheduling and dispatch of renewable energy.[iii]
Second, AI can improve the transmission and distribution of electric power. AI can help with transmission expansion planning, determining the best location and capacity of new transmission lines.[iv] AI can be especially helpful in optimal power flow (OPF) analysis—evaluating the most efficient and reliable flow of electricity through a transmission network.[v] AI algorithms are essential for dynamic line rating — a method of determining the maximum capacity of transmission lines based on current weather and line conditions that can increase the capacity of transmission lines by at least 30%.[vi]
Third, AI plays an especially important role in virtual power plants (VPPs) — networks of decentralized, distributed energy resources including end-use devices.[vii] Many VPPs combine AI-driven demand predictions and the ability to manipulate the power demand of end-use devices. VPPs help integrate renewable power into electric grids, limiting the need for expensive peaker plants that supply power during high demand periods and cutting greenhouse gas emissions.
Finally, AI can improve – and potentially revolutionize – energy storage. AI can help integrate energy storage into power grids, predicting when renewable power will be curtailed and supporting energy storage scheduling more broadly.[viii] AI can help turn electric vehicles into grid assets, supporting vehicle-to-grid (V2G) programs.[ix] AI has the potential to dramatically accelerate the pace of innovation in battery chemistry and other energy storage technologies, using neural networks and other AI techniques to identify innovative materials for energy storage.[x]
Barriers
However several barriers limit the adoption of AI in the power sector.
First, poor data quality limits the AI use in the power sector. In the United States, for example, utilities, independent system operators (ISOs) and regional transmission organizations (RTOs) make data available in slightly different ways—across different time horizons, in different formats and with different frequencies—making it difficult or impossible to do analysis across all relevant players in the power system.
Second, the lack of AI-training in the workforce is a significant barrier. AI’s application in grid infrastructure requires a workforce that is knowledgeable on both the electric grid and AI.
Third, poor market design hinders adoption of AI in the power sector. When utility revenues are based on regulated rates of return on capital assets, reward utilities may lack incentives to invest in AI-driven solutions. Fragmented markets and inconsistent regulations across regions can complicate the deployment of AI, limiting its potential to optimize energy systems, reduce emissions and enhance grid reliability.
Risks
Deploying AI in the power sector creates a number of serious risks, including those related to bias, invasions of privacy, safety and security.
First, AI can lead to biased outcomes when training data do not accurately represent real-world conditions. Data sets from one region could work poorly in another region due to differences in weather conditions, topographies or other factors. An AI model trained on power system data without adequate information on poor communities could recommend infrastructure investments that fail to adequately serve those communities.
Second, use of AI in the power sector could result in privacy breaches. AI systems require large amounts of data to function well. Data collection on topics such as energy consumption patterns and customer payment histories may be important for some AI applications but creates a risk of unauthorized access, identity theft and related problems.
Third, catastrophic failures could result if an AI system recommends or makes an incorrect decision due to a flaw in its algorithm or an unforeseen situation. Such failures could include equipment damage, power outages or worse. Rigorous testing, continuous monitoring and robust fail-safe mechanisms are crucial to ensure the safety of AI-operated energy systems.
Fourth,AI systems are susceptible to cyberattacks, including attacks where malicious actors manipulate the AI’s input data to cause harmful outputs. Such attacks could lead to incorrect decisions that could disrupt power supply, damage infrastructure or even facilitate further attacks on the grid. Robust cybersecurity measures, regular updates and stringent access controls are essential to protect AI systems from such threats.
Recommendations
AI has the potential to transform many parts of the power system, if barriers can be overcome and risks can be addressed. To help realize that potential, the authors recommend the following steps:
National governments, electricity regulators and utilities should work together to develop and enforce data standards for all aspects of grid operations. Regional governing bodies, such as the US ISOs and RTOs, should prioritize standardization of data to enable cross-regional analysis. These data should be available in industry standard formats in free and publicly available portals for use in AI modeling and research.
Utilities and electricity regulators should launch programs for training workers in the power sector to assess and use AI-driven technologies.
Electricity regulators should create clear regulatory frameworks to support using AI in energy management. These frameworks should include rates that provide cost recovery for AI-related investments, such as smart meters, sensors and open-source grid management software. The frameworks should address risks related to data privacy, safety and cybersecurity.
Utilities, regulatory agencies and academic experts should work together to develop AI-driven AC-OPF (alternating current-optimal power flow) models and permitting reforms. These models should be used to reduce delays in the interconnection process and accelerate deployment of new renewable generation sources to the grid.
National governments should encourage and fund collaborative R&D projects between academic institutions, industry and utilities focused on AI and related applications for renewable power, energy efficiency and emissions reduction, including AI-driven forecasting tools and grid management systems.
With these steps and others, AI could help deliver a cleaner, cheaper and more reliable power systems in the years ahead.
Part Two of this series will look further at the energy and GHG emissions challenges associated with AI.
[i] E. Engel & N. Engel. A Review on Machine Learning Applications for Solar Plants. Sensors-Basel 22 (2022). https://doi.org/10.1016/j.nexus.2021.10001110.3390/s22239060.; R. Ahmed et al. A review and evaluation of the state-of-the-art in PV solar power forecasting: Techniques and optimization. Renew Sust Energ Rev 124 (2020). https://doi.org/10.1016/j.rser.2020.109792; L. Ekonomou et al. Estimation of wind turbines optimal number and produced power in a wind farm using an artificial neural network model. Simul Model Pract Th 21, 21-25 (2012). https://doi.org/10.1016/j.simpat.2011.09.009; S. A. Renganathan et al. Data-driven wind turbine wake modeling via probabilistic machine learning. Neural Comput Appl 34, 6171-6186 (2022). https://doi.org/10.1007/s00521-021-06799-6.
[iii] Quentin Paletta et al. Advances in solar forecasting: Computer vision with deep learning. Advances in Applied Energy 11, 100150 (2023). https://doi.org/10.1016/j.adapen.2023.100150; S. X. Lv & L. Wang. Deep learning combined wind speed forecasting with hybrid time series decomposition and multi-objective parameter optimization. Appl Energ 311 (2022). https://doi.org/10.1016/j.apenergy.2022.118674; G. Etxegarai et al. An analysis of different deep learning neural networks for intra-hour solar irradiation forecasting to compute solar photovoltaic generators’ energy production. Energy Sustain Dev 68, 1-17 (2022). https://doi.org/10.1016/j.esd.2022.02.002.
[v] Pascal Van Hentenryck. Machine Learning for Optimal Power Flows. Emerging Optimization Methods and Modeling Techniques with Applications, 62-82 (2021). https://doi.org/10.1287/educ.2021.0234; F. Hasan, A. Kargarian & A. Mohammadi. “A Survey on Applications of Machine Learning for Optimal Power Flow” in 2020 IEEE Texas Power and Energy Conference (TPEC), College Station, Texas. 1-6, https://doi.org/10.1109/TPEC48276.2020.9042547,(2020).
[viii] P. Moutis, S. Skarvelis-Kazakos & M. Brucoli. Decision tree aided planning and energy balancing of planned community microgrids. Appl Energ 161, 197-205 (2016). https://doi.org/10.1016/j.apenergy.2015.10.002; K. S. Zhou, K. L. Zhou & S. L. Yang. Reinforcement learning-based scheduling strategy for energy storage in microgrid. J Energy Storage 51 (2022). https://doi.org/10.1016/j.est.2022.104379.
[ix] C. Scott, M. Ahsan & A. Albarbar. Machine Learning Based Vehicle to Grid Strategy for Improving the Energy Performance of Public Buildings. Sustainability-Basel 13 (2021). https://doi.org/10.3390/su13074003;
Kenya and South Africa have recently started moving toward an open access regime in their electricity sectors, while the US and India have been on this path for over two decades.
The rapid expansion of artificial intelligence (AI), especially Large Language Models (LLMs) such as GPT-3 and Gemini on which the now well-known ChatGPT AI and Gemina assistant systems...
Recent congressional legislation has the potential to accelerate the construction of new electricity transmission projects that cross existing electric grid boundaries. These interregional transmission projects provide a host...
ICEF develops roadmaps on how key innovative technologies can contribute to a transition to clean energy. Roadmaps consider industrial, academic and governmental perspectives to identify a realistic, fact-based pathway and meaningfully inform the work of all stakeholders.
Paul M. Dabbar testified on October 19, 2023 before the U.S. House of Representatives’ Energy & Commerce Subcommittee on Energy, Climate, and Grid Security.
Testimonies & Speeches
by Paul M. Dabbar• October 19, 2023
