Artificial Intelligence is unleashing a wave of transformation in the energy industry, revolutionizing processes and enabling unprecedented advancements. In efficiency, accuracy. And agility.
Prominent energy companies worldwide have swiftly embraced AI, recognizing its potential to optimize planning, facility management, environmental impact, energy storage, and distribution. AI applications implemented in the energy industry with documented case studies include:
- Optimizing Energy Systems: The Grid
- Forecasting Energy Supply and Demand
- Generative AI/Large Language Models and Natural Language
- Improving Scheduling and Flexible Demand
- Predictive Maintenance and Fault Detection
- Market Redesign
- Managing Energy Systems Data
- Industry risks (e.g., energy theft)
- The same “back office” automation as all other companies
Optimizing energy systems - the grid
Power grids are getting much attention lately, as massive failures in Texas, for example, caused misery for people and businesses from February 10-27, 2021, during a deep freeze winter storm. As of January 2, 2022, the total loss of life was reported to be 246. Grids are physical networks to transport electricity. Managing the grid is understandably complex and becoming more so. The irregular input from solar, wind, and other variable electricity generators requires power plants, storage, or other sources to compensate for unexpected events (weather in particular).
AI techniques for balancing power generation with consumption in real time, such as unit commitment and optimal power flow, address this challenge by minimizing costs and efficiently scheduling generating units:
Unit commitment minimizes the total cost of power generation by defining adequate scheduling of the generating units. For an academic paper on Unit Commitment in power systems, see A Stochastic Unit Commitment in Power Systems with High Penetration of Smart Grid Technologies.
Optimal power flow (OPF) finds a steady state that minimizes the cost of generation, satisfying constraints, and meeting demand. For an. academic paper on OPF, see Optimal Power Solutions.
Both techniques are NP-hard, a measure of complexity defined by the NIST as problems “that are intrinsically harder than those that can be solved by a non-deterministic Turing Machine in polynomial time.”
Given the complex nature of optimization problems, and the flow and scale of data, previous technologies, such as rules-based systems and decision trees, are hindered by their difficulty in adapting to new situations. The capacity of AI to solve optimization problems at scale opens new opportunity for grid operators to not only improve grid performance, but enables incorporation of new sources and uses of power.
Forecasting energy supply and demand
Accurate energy supply and demand forecasting is critical for efficient operations and planning. AI, particularly Machine Learning (ML), has proven instrumental in this domain. By leveraging historical data, physical models, images, and video data, ML algorithms can create short- to medium-term forecasts for variables like solar and wind power generation. AI models can optimize forecasts to align with system goals, such as minimizing costs or reducing greenhouse gas emissions.
Accurate demand forecasting allows energy companies to balance supply and demand more efficiently. AI models that incorporate factors like weather, time of day, location-level demand, and events have been shown to produce demand forecasts superior to traditional methods. Companies like Engie, National Grid, and NextEra Energy are using AI to enhance forecasting and gain visibility into future demand across their networks.
The optimization of electric power grids starts with some basic principles of physics and engineering, posing several challenges. To insert enough power into the grid to equal the amount consumed at every moment, the grid topology's physical constraints must be accurately cataloged. Optimization models balance these constraints using, among other techniques:
AI adapts very well to these kinds of problems. Legacy optimization models are slow and cumbersome to modify and test. Yet, power system optimization must operate much more frequently and under greater uncertainty from unexpected events, both because of this variability and on account of climate change adaptation–related needs.
Grid constraints are cited as a major barrier to renewables. The solution is investments in enhancing and reinforcing grid infrastructure, especially to mitigate against "congestion" and make it more capable of handling higher intermittent renewables. Some proposed AI ideas include methods to enhance and accelerate legacy optimization techniques or to replace them altogether. Security solutions are prime avenues for AI and include formulating secure power system optimization as an attacker–defender game and attempting to find a Nash equilibrium of this game using Reinforcement Learning. However, most RL techniques do not provide provable guarantees, requiring they work in concert with other traditional models,
These methods span various types of supervised ML, fuzzy logic, and hybrid physical models and take different approaches to quantifying (or not quantifying) uncertainty. At a more granular level, some work has attempted to understand specific demand categories, for instance, by clustering households or disaggregating electricity signals using game theory, optimization, regression, and/or online learning. Since variable generation and electricity demand both fluctuate, they must be forecast ahead of time to inform real-time electricity scheduling and longer-term system planning.
Better short-term forecasts allow system operators to reduce their reliance on polluting standby plants and proactively manage increasing amounts of variable sources. Better long-term forecasts can help system operators (and investors) determine where and when variable plants should be built. While many system operators today use basic forecasting techniques, forecasts must become increasingly accurate, span multiple horizons in time and space, and better quantify uncertainty to support these use cases. AI can help on all these fronts (emphasis mine).
AI algorithms of the future will need to incorporate domain-specific insights. For instance, since weather fundamentally drives both variable generation and electricity demand, ML algorithms forecasting these quantities should draw from innovations in climate modeling, weather forecasting, and hybrid physics-plus-ML AI models should also directly optimize for system goals. For instance, using a deep neural network to produce demand forecasts that optimize for electricity scheduling costs rather than forecast accuracy could be extended to produce forecasts that minimize GHG emissions.
In settings where power system control engineers decide power generation, AI's generative capabilities shine. Using large language models and automated visualization techniques, AI empowers engineers to gain deeper insights from forecasts. This newfound understanding aids in optimizing scheduling and enhancing the performance of low-carbon generators.
Improving scheduling and flexible demand - scheduling and dispatch
Determining power generation levels is a complex and time-sensitive task. AI offers promising solutions to address these challenges. ML algorithms can approximate or simplify optimization problems, identify redundant constraints, and learn from power system control engineers. Dynamic scheduling and reinforcement learning techniques further contribute to real-time grid balancing, accounting for fluctuations in electricity production from storage, variable generators, and elastic demand.
Predictive maintenance and fault detection
AI's ability to swiftly detect and rectify faults in energy systems delivers numerous benefits. By promptly addressing issues, AI-powered systems enhance equipment longevity, reduce waste, and bolster overall system robustness. ML techniques have proven effective in fault detection and forecasting, covering diverse energy systems such as natural gas pipelines, solar arrays, and nuclear power plants. Anomaly detection, supervised learning, clustering, and denoising methods aid in flagging irregularities, ensuring proactive maintenance and optimized system performance.
Transitioning to renewable energy sources necessitates innovative market designs. AI, in conjunction with traditional economic and game theoretic approaches, facilitates the development of such markets. Reinforcement learning techniques enable setting dynamic prices and achieving equilibria in energy trading markets. Additionally, AI models can incentivize participation in demand response programs and optimize power prices, promoting efficient energy utilization. Managing Energy Systems Data AI supports data-driven analysis by assisting in data cleaning and preprocessing. Techniques such as clustering and ML algorithms automatically match records from different energy systems datasets. In smart grid applications, AI-driven approaches ensure sustainable data collection and transmission, minimizing the environmental impact.
By harnessing the power of AI, energy companies can navigate the complexities of grid optimization, enhance forecasting accuracy, streamline scheduling, ensure proactive maintenance, provide for a stable and sustainable future, and optimize market dynamics. However, most organizations must develop good data architecture and discipline to support Safety-Critical Systems (SCS). Nor can they overlook ethical and societal issues. For example, an optimization and scheduling model for retrofitting and upgrading buildings likely to exceed their specifications could discriminate against marginalized neighborhoods. Since many of the applications of AI in the energy business are highly technical, it will be difficult to consider the social context. This is a problem that good processes from end-to-end can mitigate.