How Utilities Are Using AI to Autonomously Manage the Energy Grid

When power generation was centralized, distributing electricity across the grid was relatively straightforward for utilities. In most energy markets, demand followed predictable patterns, which allowed grid operators to pull whatever levers were necessary to onboard sufficient power. Some forward-thinking utilities even installed on-site battery storage in anticipation of the slow ramp-up speeds that are often associated with fossil fuel power generation. 

Today, however, the energy landscape is very different. 

Power generation and storage are no longer centralized thanks to the rapid rise of privately owned distributed energy resources (DERs) like: 

• Solar photovoltaic (PV) panels 
• Wind turbine technology 
• Battery energy storage systems 
• Electric vehicle (EV) charging stations 

These green investments are critical for moving society off fossil fuel. However, rapid decarbonization continues to pose significant challenges for those tasked with managing the electricity grid. 

Aside from intermittency hurdles that solar and wind power generation bring, there are even more fundamental challenges that grid operators must overcome to help transition the world to a truly carbon-neutral future.

When Simply Pulling Levers Is No Longer an Option

By definition, most distributed energy resources are outside the direct control of grid operators. That these edge devices are able to autonomously and independently generate, store, or consume power makes it extremely difficult for utilities to properly balance energy supply and demand – leading to shortages, surges, device failures, and less reliable delivery. 

Worse still, the number of DERs continues to grow exponentially as more homeowners and businesses invest in renewable power. In 2019, for example, the Solar Energy Industries Association (SEIA) reported that the US had roughly 70 GW of installed PV capacity. Despite a global pandemic and economic recession, that number had mushroomed to 130 GW by 2022. 

Against this backdrop, how can any utility operator know when to onboard more power or relay excess electricity into storage devices? Equally important, how can grid operators remotely optimize privately owned edge devices – especially when their internal circuitry was never designed for the types of power surges one might expect from a poorly balanced grid? 

Even with perfect insights, there will always be a lag if human actors are pulling the lever. There is simply too much real-time data to collect, analyze, and act on. This is particularly true since technologies like solar and wind are highly dependent on weather conditions that fluctuate every second of every day. 

Faced with the above challenges, one investor-owned utility in the American Southeast decided to adopt an innovative approach to these increasingly common problems. 

Outsourcing Power Management to a Higher Power

Like a growing number of utilities, this power provider had set ambitious carbon targets, with the goal of replacing all fossil fuel generation with renewables by 2050. Having already installed 1,200 MW of solar capacity, the utility had successfully reduced its CO2 emissions by 50% over the previous 2 decades – with a corresponding 95% reduction in NOx, SOx, and other greenhouse gasses over that same timeframe. 

This is clearly the right direction from an environmental standpoint. 

However, the sheer amount of resources devoted to greening the grid was eating into the utility’s profitability. Worse still, edge devices continued to malfunction under the stress of volatile power delivery across the electricity grid. 

Faced with these challenges, the utility decided to turn its largest liability – “Big Data” – into a powerful asset in their quest toward net-zero energy goals. More specifically, the operator started using a distributed energy resource management system (DERMS) powered by artificial intelligence (AI) for one of their solar and storage power plants. 

The utility made this decision due to AI’s unrivaled capacity for:

1. Data Collection and Analysis

Unlike humans, artificial intelligence can easily collect and parse many terabytes of historic and real-time data. In addition to grid-wide supply and demand, this analysis also included weather and climate information – plus the production, storage, and consumption levels of every grid-connected edge device equipped with sensors to transmit information back to the utility.

2. Predictive Modeling

Artificial intelligence can make computational decisions much faster than humans can. AI is able to identify hidden patterns in large data sets, which allows it to make far better predictions about where the supply, demand, and price of energy will be seconds or days into the future. When coupled with iterative machine learning, these predictions are matched against real-world results to ensure the AI’s forecast modeling becomes even more accurate with time. 

During early trial runs with its AI-powered DERMS solution, for example, the utility achieved an accuracy rate of 3.5% compared to actual solar PV production.

Within 6 weeks of refining its own predictions, however, that gap had shrunk to just 0.5%.

3. Autonomous Decision-Making

AI has the ability to autonomously act on real-time data and its predictive modeling much faster than humans can (when pulling levers). Prior to using AI-powered DERMS technology, for example, the utility had manually set conservative charging and discharging schedules for its onsite battery systems in anticipation of future demand. With artificial intelligence, however, these schedules were managed automatically – based on when grid conditions were optimal. 

The end result is cheaper, cleaner, and more reliable electricity for the entire grid. 

Equally important, this autonomous control also helps to prevent unnecessary device degradation. For example, the utility was able to reduce solar inverter temperature fluctuations and overheating by nearly 33%. In addition to extending inverter lifetimes, AI also helped to reduce the utility’s spinning reserve margins.

Can the World Go Green without Artificial Intelligence?

It’s important to note that the above example was highly localized, with the grid operator only using AI-powered DERMS technology to manage the energy assets under its direct control. However, this early project clearly illustrates the technology’s potential. In fact, adding more DERs actually improves predictive modeling and autonomous decision-making since there are more patterns to identify among all the historic and real-time data collected. 

The levers are still there. But there are literally millions of them being pushed and pulled throughout the day as artificial intelligence and iterative machine learning work in tandem to maintain the optimal balance of clean and affordable energy – regardless of grid, weather, or market conditions.