Introduction
Three ways artificial intelligence (AI) is bringing value to electric utilities include building energy modeling (BEM), geographic information systems (GIS), and structural equation modeling (SEM). Of course, there are other ways too, and they do not align one-to-one with challenges in the sector such as safety, reliability, maximizing cost-effectiveness, minimizing environmental impacts, integrating intermittent renewable and distributed energy resources (DER), servicing the emerging electric vehicle (EV) market, enabling population growth, and being typecast as the first casualty in a heist film. Nevertheless, these three things are making an impact in a lot of areas and will surely continue to add value for years to come.
1. Building Energy Modeling (BEM)
Building energy modeling (BEM) uses AI-based simulation software to perform detailed analyses of a building’s energy use and energy-using systems. For example, the term “eighty-seven sixty analysis” is commonly used to describe an analysis that includes every hour of a complete calendar year (365*24 = 8760). Peak demand models can be extrapolated from 8760 analyses through application programming interfaces (APIs), and several versions of that functionality should be available for standard software users soon. Similarly, the time-scale resolution of 8760 analyses will advance from hourly to minute by minute (8760 * 60 = 525,600) within the tools, and then finer and finer resolution beyond that as is already performed by boutique electrical power flow simulation software. BEM tools are instrumental for developing building and appliance efficiency standards, sizing on-site DER, neighborhood-scale urban planning, microgrid community planning, and more. BEMs can have thousands of parameters for parts, materials, geometry, physics, economics, ambient weather conditions, etc. Residential building models are more precise than commercial building models, and there are significant opportunities for improvements in quantifying uncertainties associated with the technological, weather, and human behavioral factors that affect energy usage in buildings. BEM results are also used as inputs to analyses of utilities' distribution circuits. And since both of these types of quantitative studies require at least some basic amount of geographic information, it is not a coincidence that GIS is emerging as the premier IT integration platform.
2. Geographic Information Systems (GIS)
Geographic information systems are complex software platforms used to integrate business, engineering, and environmental systems over wide-ranging geographies. To say that GIS is map-making software would be an understatement. This AI-based software industry is growing, and new interdisciplinary GIS departments are creating advanced capabilities every day. For example, higher levels of geographic resolution are being enabled in pre-existing processes, such as energy demand forecasts; predictive risk models for component capacities, resource adequacy, and outages; mobile communications between corporate, field, and customers; transaction recordkeeping; and so on. Going forward, GIS will be instrumental in solving power quality and reliability problems associated with high levels of intermittent DER and EVs, including optimizing trading in energy imbalance markets from distribution-level aggregated activities. Utilities and other entities in different sectors can learn a lot from each other as they continue to digitize their infrastructure data, implement new sensor technology, consolidate legacy processes, and create other new ones synergistically with the blossoming of GIS as an AI platform.
3. Structural Equation Modeling (SEM)
Structural equation modeling is the process of defining an algorithm with mathematical equations and computational parameters in the form of software code that a machine can execute to deduce quantitative descriptions of uncertain patterns based on a sample of data. In other words, to do SEM is to do multi-layered regression analyses. The most basic form of regression analysis, that we are all familiar with, is fitting a straight line through a scatter plot of data in a manner that provides the best fit.  For example, in large geographic regions with high air conditioner (AC) penetration, electricity demand appears to increase linearly as a function of air temperature from about 85—100 °F (30—40 °C) or so. But there is nothing linear about AC thermostat settings, AC on/off duty-cycling, the additional workload that compressor motors do to compensate for less effective radiator performance under higher ambient air temperatures, etc. That linear-looking section of the graph is really the middle-section of a slanted S-curve. I explained this concept in reproducible detail in my 2017 publication in the Journal of Applied Energy. The model fit parameters were based on freshly published data from the US Department of Energy, the local county assessor, appliance surveys, weather stations, and utilities in the US Southwest. Not only did this approach to answering the research question—what would happen in a record-breaking heatwave—yield the first sub-linear trending results in a Climate Change study published in an academic journal, it also yielded several inferences about latent variables in the balancing authority areas. Going forward, companies with large amounts of operations data should be using SEM to improve forecasts and optimize intelligent control systems wherever there are uncertainties that we can put a sizable dollar amount on.
Conclusion and Recommendations
BEM, GIS, and SEM are three examples of different types of AI systems that offer different kinds of value for electric utilities. However, as with any technology, a tool has no purpose until its user gives it one. In the same way that these AI systems will be used to create our next generation of dynamic clean energy grids, they will also be used in advanced cyber-warfare. For example, the 2028 Summer Olympic Games in Los Angeles is an obvious target for a Stuxnet-like cyber-attack. And since California has pioneered so much research and development in the public sector, anyone with an internet connection and the right knowledge can access address-level operations data and calculate multiple ways to disable the State's power grids. Going forward, these kinds of detailed data about our critical infrastructure systems should not be public record. But, since Pandora’s box has already been opened in California, and since implementation timelines are typically 7-10 years, local utilities should take defensive action against such cyber-attacks now. In those regards, as economic activity returns to pre-COVID-19 levels, it is imperative that we invest heavily in energy efficiency, and multi-layered redundancies between central and distributed power systems, to mitigate these risks. What other AI systems would you add to the list? And what else do you think we should be doing now to pioneer clean energy grids?