Integrating Batteries into the Grid

Utility companies across the world have begun replacing coal- and gas-fueled power plants with large batteries that store solar and wind energy. In the United States, California and Texas are leaders in deploying this technology, with states including New York developing a nascent capacity for grid-scale storage.

This significant advancement brings new challenges. With traditional grids, a utility could easily adjust its generators to meet consumer demand. Managing a grid that relies on batteries requires a more strategic approach. Increasingly, grid managers will make decisions (or oversee algorithms that make decisions) based on the type of predictive models that my colleagues and I are developing.

We’re all accustomed to thinking of energy infrastructure in terms of power plants, electrical lines, and other physical aspects of the grid, but AI systems are quickly becoming an indispensable part of the energy system. With smart research and investment, it is possible to keep electricity cheap and reliable while drastically reducing the amount of fossil fuels we burn.

The Challenge of Managing Grid-Scale Batteries

In theory, these batteries should be charged when renewable sources are producing more energy than consumers need, and they should send that extra energy onto the grid when demand exceeds supply. 

In reality, it’s not so easy. To ensure that power is always available, grid operators have to predict the production and consumption of energy hours or even days in advance. They use algorithms to analyze large and diverse datasets — including weather data, historical consumption data, and market prices — to make these predictions. 

The systems that make these forecasts are rapidly becoming an essential piece of the electrical infrastructure. In California, where battery capacity now accounts for nearly 30% of the state’s power capacity, decisions about when to charge and discharge batteries have become critical to maintaining grid reliability.

The promise – and complexity – of integrating ai

These large batteries and the electrical grids they serve are usually owned by different companies. These companies interact by continually setting and updating the price at which they’re willing to buy and sell energy. 

Using an approach called predictive control, the battery owners and grid operators use historical data and real-time inputs to forecast future conditions. These systems, which are very similar to the algorithms used in automated stock trading, ultimately inform decisions about when to charge and discharge the batteries.

In a well-designed market, this application of AI will ensure that stored energy is available when it’s most needed while offering investors the incentives to fund new storage products and extend the reach of renewable energy.

The use of AI in energy markets raises questions about fairness and transparency. As more decisions are driven by algorithms, there is a risk that the complexities of AI could obscure the decision-making process, making it difficult for regulators and consumers (and even the battery owners and grid operators themselves) to understand how energy prices are set or why certain decisions are made. 

Furthermore, while fossil fuel prices are usually public, prices for electricity from grid-scale batteries usually aren’t. This opacity could lead to perceptions of unfairness, particularly if AI-driven decisions result in price spikes that leave consumers with sky-high utility bills and increased risk of outages.

Charting the path forward

Addressing these challenges requires a better understanding of the dynamics, incentives, and technical systems that underlie these new markets. By investing in this research, governments and private companies can better position themselves to foresee potential issues and quickly address problems as they arise.

My research group and I are working on several lines of research to address these issues related to integrating batteries into power systems. With support from the National Science Foundation and Columbia University’s Data Science Institute, we are doing the theoretical heavy-lifting to develop new algorithms that analyze and facilitate storage operation in the power system. The U.S. Department of Energy is funding a more practical project, in association with Lawrence Berkeley National Lab, aimed at developing frameworks to improve power system operation in California. 

We are also working on an industry-sponsored project using AI to improve the market offerings of a battery company in Texas. Along with colleagues at Columbia Engineering, members of my group are analyzing the life cycle of in-home batteries and scoping out the use of sodium-sulfur batteries for long-term storage.

Read original story here: https://www.engineering.columbia.edu/lever/integrating-batteries-grid