AI, ML Add Resilience, Safety to Aging Power Grid

Artificial intelligence (AI) is assisting utility providers with strengthening North America’s grid since it can predict what existing assets will fail and how to best schedule repairs. Increasing the safety and security of the grid is key to protecting it from wildfire hazards. As the grid becomes more prone to wildfire risk, AI will become increasingly useful over the next few decades. 

Argonne National Laboratory (ANL) is at the forefront of the effort to employ AI and machine learning (ML) to benefit its utility provider partners. 

 

AI and ML can aid with maintenance and prevent outages. Image used courtesy of Wikimedia Commons

 

Reducing Outages and Maintenance Costs With AI

There is a great deal of room to improve the grid, according to Feng Qiu, principal computational scientist and section leader at Argonne National Laboratory. 

“The age and condition of the utility grids is rated C- by the American Society of Civil Engineers. Many of the hydro generators, transformers, and other devices were deployed half a century ago. (They have already come) to the end of their lifetime. What makes it worse is that, with deep renewable integration and new energy components being integrated into the grid, emerging grid operational patterns continue to push our aging transmission grid to its limits,” Qiu told EEPower.

ANL’s work with AI advises utility providers on how they can reduce maintenance costs, maximize machine life utilization, and optimize spare part inventory. ANL conducts this work by leveraging condition-monitoring information and developing stochastic process models and other machine learning (ML) models to approximate the degradation process. So far, ANL has analyzed data concerning solar inverters and batteries. It has also analyzed information regarding hydro generator assets, for which the data cover half of the U.S.’s hydro generation capacity. 

Qiu noted another impact of climate change is more frequent extreme weather events, which can lead to electricity outages. 

 

Change in the number of days and area with Keetch-Byram Drought Index (KBDI) > 600 from the historical period to the late 21st century, using Weather Research and Forecasting (WRF) model CCSM4. Image used courtesy of Brown, Wang, and Feng

 

“Most existing forecast methods largely rely on grid physical models and simulations. (They) cannot comprehend the increasingly complex and large-scale weather patterns. To foresee the potential impact on (the) power grid in a high resolution, specifically the outage progress and evolution during the extreme weather event, we leveraged ANL’s electricity customer outage database,” Qiu said. 

ANL then developed new AI/ML models independent of grid physical models. These capture the spatial-temporal features of extreme weather. The models can deliver forecasting of the outage process with good accuracy. 

 

How AI Predicts Electricity Outages

ANL starts the process by combining statistic learning with neural networks. The team captures sophisticated impacts from weather variables to model the random outage process. Given a weather forecast, the prediction can deliver a time series of outage numbers in 15-minute time resolution and county-wise geographic resolution. The prediction can also give the worst scenario, like maximum wind speeds, temperatures, and precipitation amounts, according to Alinson Xavier, computational scientist for Argonne National Laboratory.

One technology ANL uses is MIPLearn, an open-source framework for the next generation of power system optimization tools. MIPLearn relies on a combination of Mixed-Integer Linear Programming (MIP) and ML to improve the decision-making process. MIPLearn utilizes AI to identify patterns in previously solved optimization problems. 

 

Severe weather incidents by year. Image used courtesy of National Oceanic and Atmospheric Administration

 

MIPlearn then employs these patterns to generate hints to enhance the performance of conventional optimization solvers like CPLEX. The approach can produce speedups of four to ten times in power systems applications without reductions in the solution quality. 

Pinpointing power outages is a better way to handle risk than shutting off power to too many customers. One example comes from California, where utilities avoid wildfire risk by engaging in a safety operation called a Public Safety Power Shutoff (PSPS). In a PSPS, a utility provider will temporarily shut off certain substations, transmission lines, or transformers for a few days. The actions result in significant disruptions to a high number of customers. 

A better alternative to handling the risk is to develop an enhanced power dispatch model. Developing the model requires rerouting the power flows to reduce loading on transmission lines within high wildfire-risk areas. ANL has also developed an improved transmission planning expansion model to select a transmission path avoiding high wildfire-risk areas. 

 

Funding and Partnering for Argonne’s AI Projects

ANL is currently turning some research results into usable tools for the industry and research community. For example, given extreme weather forecasting, ANL developed an outage forecasting tool using spatial-temporal models to predict the electricity customer outage process. ANL is now developing a web app for use by multiple stakeholders, including grid operators and the U.S. Federal Emergency Management Agency. 

Different ANL projects have distinct budgets. The Advanced Grid Modeling Program of the Office of Electricity under the U.S. Department of Energy funds the electricity customer outage forecasting work. It has allotted the program $700,000 for two years of work. The initial $50,000 will come from ANL’s Laboratory Directed Research and Development program for the AI-enhanced optimization project. Then, the Advanced Grid Modeling Program will provide $800,000 in funding for two years of work. 

 

Future of Grid AI Improvements

Going forward, more grid improvements could significantly benefit utility providers and the public. Not all solutions will work for every utility grid. Still, many utility providers would benefit by finding a way to manage a large integration of renewable energy and massive distributed energy resources. This requires a suite of solutions, including load forecasting, behind-the-meter prediction, grid-forming, technologies, smart charging, and Distributed Energy Resource controls. 

As utility providers use AI and ML more in asset health management, they can benefit from embracing the changes brought by climate change and decarbonization. The steps include resilience enhancement and preparing for extreme weather events. 

The wide array of efforts involving AI and ML require different resources, including new technologies, policy changes, investment, and incentive programs. This makes it difficult to put a dollar amount on total costs. 

ANL currently partners on projects involving AI and ML with Commonwealth Edison (ComEd), a utility provider. ANL’s other partners include the Hydropower Research Institute, a collaborative designed for hydropower industry leaders, which provides condition-monitoring data for hydro generators on the hydro prognostics projects, and SunEnergy1 and Ideal Energy, both solar power companies. CB Solar, an installer of solar energy systems, provides inverter operational data for solar prognostics projects. Independent System Operator (ISO)-New England analyzes the electric vehicle integration into their network to study impacts on energy justice. Midcontinent Independent System Operator (MISO) provides domain expertise and advice on the AI-enhanced optimization project. Southern California Edison provides domain expertise and supervision on ANL’s wildfire project.