This is the sixth in my series of articles focused on the Modernization of the Electric Grid. In this chapter, I discuss the application of artificial intelligence and machine learning (AI/ML) technology in our transmission and distribution system.
OK. First things first. Let me be clear that I am NOT using any AI technology in the creation of this blog article!
It seems that everywhere we look in the news and other media these days, we hear about another interesting or amusing application of some new AI platform. It’s a fascinating technology, and the speed of evolution is only beginning to be understood. In fact, several technology executives just recently issued an appeal for a 6-month “pause” in the development of new AI tools. (Wall Street Journal, March 30, 2023)
Naturally, my thoughts turn to its existing and, more importantly, potential applications in the planning and operation of the electric grid. There are many, and indeed the success of modernizing the grid to handle the increased demands and complexity of transportation electrification and renewables integration might actually require the use of AI/ML technologies. More to come on that.
In this article, I will first discuss some basic definitions. Then I will provide some examples of existing uses of AI in the industry. Finally, I will delve into some potential applications where AI is likely to become a critical tool for the planning and operation of the modern grid.
Some Quick Definitions First!
Let us first establish some clarification on the definitions of AI and ML since these terms have been around for quite a while and have been used fairly broadly in society. A quick check into dictionary.com provides an adequate definition of artificial intelligence: the capacity of a computer, robot, or other programmed mechanical device to perform operations and tasks analogous to learning and decision-making in humans, as speech recognition or question answering. Machine learning can be considered a subfield of AI where computers process and evaluate data beyond programmed algorithms, learning and recognizing patterns from examples.
Historical Use of AI at Electric Utilities
Regional Load Forecasting And Weather
In addition to the number of popular existing uses of AI, like speech recognition and autonomous driving, the electric utility industry has used AI technology for several purposes for some time. Regional load forecasting for capacity planning has long used a number of relatively simple AI/ML techniques like artificial neural networks (ANNs), “fuzzy” expert systems, and deep learning. These techniques have also incorporated logical models with regression analysis and probabilistic methods to find associations between customer load and environmental variables like temperature and humidity for each hour of the day, day of the week, and month of the year. Weather forecasts are then fed into the model to predict system loads for the next hour or the next few days. Longer-term weather forecasts, coupled with predictions of customer load growth, can extend the modeled forecast, with a much larger margin of error, of course.
Energy Price Forecasting
Closely related to load forecasting is energy price forecasting, which has used similar models and methods. The load forecast is actually a critical input to this process, along with commitments of generator capacity availability and risks of the unplanned shutdown of any units. Energy market prices have become extremely volatile at times, especially under extreme weather conditions or grid contingencies. In addition, the introduction of significant solar and wind capacity has also complicated the price setting, actually causing market prices to go negative under certain conditions.
Predicting Interruptions And Post Storm Assessment
A more recent AI-enabled use case for electric utilities is in the prediction of customer interruptions due to pending extreme weather. Most electric utilities have extensive detailed history of customer outage events which can be combined with detailed historic weather conditions (e.g., wind speeds, precipitation, storm speeds, and tracks) to create predictive models. These models are continually refined as history is captured. Predictions are far from perfect still (just like the weather!), but providing any planning tool to help utilities more efficiently reserve and pre-position line and tree crews, as well as important replacement stock for damaged facilities, can be a huge help for efficient restoration.
On the back end of the storm, utilities are also beginning to use AI technology for applications like video and still image recognition. In order to improve the speed and efficiency of post-storm emergency system patrols, some utilities are using this technology to rapidly assess where facilities are damaged and to determine as quickly as possible what resources will be needed for repair efforts. Similar technology is also beginning to perform routine inspections and even to assess risk of wildfire caused by transmission and distribution lines in extreme winds.
Huge Potential for AI in the Electric Utility Industry
There are many more exciting possibilities for near- and long-term future potential applications of AI in the electric T&D industry. In fact, as mentioned above, AI technology will likely be required for safe and effective operation of the grid for at least a few reasons. The following are a few examples that come to my mind.
DER Production Forecasting
As I have described in previous blog articles, the advent of renewable energy sources from solar and wind has added great complexity to the planning and operation of the T&D system. One impacted process in particular, and discussed just a few paragraphs earlier, is load forecasting. In addition to traditional customer load forecasting, there is now a fast-growing need to predict power generation from solar and wind to calculate the net system load (for a neighborhood, a distribution circuit, a substation, or even a large transmission area). Prediction needs can range from minutes or hours ahead (think cloud movement for solar generation) to days ahead (weather forecasts for sun and wind) to years ahead (prediction of customer adoption of rooftop solar and EV purchases). Now that we are accumulating vast amounts of historical data from meters and sensors across the grid and more detailed and localized weather history, we can apply AI technology and methods to predict the future more quickly and accurately. These predictions will be key inputs to the critical IT systems we will need to manage the grid at all levels.
Digital Twin Accuracy
In my previous blog article, I discussed many aspects of the distribution digital twin, including how critical an accurate distribution model is for the same critical systems mentioned above. I also discussed the challenges of improving (and maintaining!) the completeness and accuracy of existing data sources that make up the entire digital twin. Scouring these large databases for gaps and errors is a massive challenge, which will have to be a continuous maintenance effort going forward. Applying automation to this process is essential, and AI will likely prove to be necessary to improve the determination of the most critical issues, find the source where correction is needed, and avoid potential tsunamis of “false negatives” or other noise that can overwhelm the correction effort.
Customer Engagement and Distributed Resource Optimization
Customer meter data from newer Advanced Metering Infrastructure (AMI) systems (now covering about half of the Country) is providing extremely granular time series information about customer energy use, along with other valuable metrics like voltage, power factor, and power quality data like momentary outages and voltage sags or spikes. This data is already providing (in some cases with AI-enabled methods) significant value to utilities and customers by indicating potential problems before they cause an outage and by improving the customer connectivity details for the digital twin model.
AI technology will also provide interesting and very useful ways of virtually “clustering” customers into useful groups. For example, customers can be grouped by those who likely have rooftop solar, EVs, or electric heat and hot water or just appear to have very similar load profiles. This type of clustering will become very important as customer-distributed energy resources (DERs) are increasingly used to support the grid.
Grid support can happen with traditional demand response (think Nest thermostats) to reduce grid peak load, or through intelligent EV charging (to limit grid impacts), or through virtual power plants (VPPs) where large groups of customers can be aggregated to provide a variety of functions. VPPs can use small amounts of power export from many customer batteries, for example, to provide short-term, large-scale generation capacity to the regional grid. Those same batteries could also be used to import power if needed rather than curtailing large-scale wind or solar generation. Being able to target customer groups more efficiently and accurately forecasting the response of aggregated VPPs, will truly unleash the benefits of DERs to the grid.
Another area where AI will provide critical support is in the field of cybersecurity for the grid and its underlying information technology/operation technology (IT/OT) systems. Parsons has long had significant expertise in cybersecurity, particularly with large commercial and government clients and in the defense industry. The recent acquisition of IPKeys Power Partners, including the IPKeys Cyber Partners unit, has significantly expanded this capability into the utilities industry, supporting grid modernization of electric, water, and gas systems and other critical infrastructure security.
According to Loney Crist (SVP, IPKeys Cyber Security Software Development), “AI technology will enable improved detection of malicious activity involving grid devices and networks, as well as in the underlying IT/OT systems.” Because the grid is constantly in a state of change and reconfiguration, with new customers and numerous devices frequently being introduced, and existing devices being removed or relocated, there is a continuous need for monitoring and threat detection. Grid devices are supplied from a variety of manufacturers, with a variety of software/firmware standards and varying capabilities for threat detection. Utility companies are challenged with keeping device firmware updated, and grid operators risk being overwhelmed with alarms (real or false) without some type of AI-enabled screening of potential problems and summarization of related alarms. AI can also provide benefits in flagging suspicious remote user activity on corporate networks, and in assessing the potential risk of internet URLs being accessed through the networks.
And That Was Just A Sample…
The potential opportunities for AI in the electric utility sector are significant and go well beyond the examples I have discussed here. As technology rapidly evolves, engineers and operators of the grid will be continually faced with decisions on whether or how to apply it to critical processes. In all likelihood, it will eventually become apparent that AI tools and methods will indeed be critical for many aspects of operating the modern electric grid. Exciting times!