The rapid adoption of electric vehicles (EVs) presents big opportunities and challenges for utilities. On the one hand, EV charging is expected to significantly increase electricity sales and support peak shaving, voltage management, and other aspects of grid operations. On the other hand, these loads—if not managed—can increase a utility’s peak demand and overload transformers, power lines, and other grid assets, compromising grid reliability and leading to costly grid upgrades and maintenance needs.
The good news is that tools are available today for utilities to address these challenges. With active managed charging programs, utilities can use smart, controllable chargers, EV telematics, and software to monitor and control EV charging—in ways that respect EV owners’ needs while reducing peak demand, regulating voltage, and deferring grid upgrades. However, since less than half of EV owners are typically enrolled in utility programs, a majority of EVs remain invisible to the utility, creating a headache for utility planners and operators who are responsible for managing EV impacts.
In this article, I will explain how utilities can use machine learning algorithms to find invisible EVs and then use that awareness as part of a holistic approach to managing EVs and other distributed energy resources (DER).
Classification and Disaggregation Algorithms
Classification algorithms use machine learning to evaluate smart meter usage data and identify EV charging loads. Meter data with 15-minute granularity works quite well, and is often available to utilities that have deployed smart meters. Humans train these algorithms by feeding them historical data on real EV charging events. The algorithms then learn to identify the large load jumps and drops typical of EV charging.
By evaluating data from all residential smart meters in a utility’s service territory, the algorithms can estimate the total number of home charging events and the subset of those events occurring during the grid’s peak demand. This analysis, when coupled with Geographic Information System (GIS) data on the meters’ location, can associate loads with upline grid equipment—including specific transformers.
After classification algorithms identify charging events and locations, utilities can use another type of machine learning algorithm that disaggregates and quantifies each EV charging load from the overall home load. These algorithms use regression analysis to estimate the amount of power consumed. The results help quantify the opportunity for load shedding or shifting if the EV is enrolled in a demand management program.
While classification and disaggregation algorithms can be quite useful, they are not perfect. They are more effective at identifying and quantifying loads from Level 2 charging equipment (typically about 7 kilowatts). Smaller Level 1 charging loads (1-2 kilowatts) may look like loads from other home appliances and remain hidden from view. The algorithms may also mistake Level 2 charging for other residential loads with similar profiles—such as heating systems that work by charging storage tanks.
The algorithms may be less effective for charging loads at commercial facilities, such as office complexes and malls. These utility customers generally have many other large loads that make it more difficult to identify and quantify EV charging usage. A utility may be able to find public charging stations in its service territory using existing databases: DOE EERE or PlugShare.
My company, Camus Energy, is not alone in helping utilities use machine learning algorithms to improve EV awareness. Companies like Bidgely, Uplight, and Oracle also have these capabilities. Some utilities are building their own algorithms in-house.
A utility may be able to supplement algorithm analyses with publicly available data on EV registrations. The organization Atlas Public Policy has compiled anonymized EV registration data from various state agencies. Utilities can use this information to estimate the number of EVs in their service territories. The registration data has limitations. Data from a relatively small subset of states is available, and it’s organized either by county or zip code, making it difficult to associate EVs with specific, local grid areas.
Deriving Value from EV Charging Awareness
Utility operations, planning and programs teams can use the outputs of these algorithms for numerous valuable applications. Planners responsible for building out grid capacity can better forecast future EV load growth and identify "hotspots"—local concentrations of EV charging—where infrastructure upgrades may be needed.
With a better understanding of EV loads and grid impacts, utility program staff can design effective active managed charging programs to mitigate adverse EV-related grid impacts and maximize grid benefits. They can also notify identified EV owners about EV programs and encourage them to enroll. Operations teams can design control protocols to address grid needs while ensuring EVs are charged when owners need them.
In addition, identifying and quantifying EV loads can help utilities promote social equity. For example, program staff may discover that EV adoption in the utility’s service territory is concentrated in affluent communities. Based on this insight, they may launch an EV program that supports adoption among lower income groups.
Put simply, understanding where and when EVs are charging can help utilities better manage their grid – today and into the future.
Optimizing EV Charging with Grid Orchestration
Managing EV impacts is quickly becoming an important responsibility for utilities, but EV adoption isn't the only change happening at the grid edge. Forward-looking utilities are developing holistic strategies to orchestrate EV charging alongside other DERs such as rooftop solar, battery storage, and flexible loads. Doing so relies on what we call a DER orchestration platform—a single interface for monitoring local grid conditions and controlling DERs.
Camus Energy has been working with utilities to integrate EV-related capabilities into its DER orchestration platform. For instance, using Camus’ software, a utility can view a map of its grid showing the locations of known EV chargers participating in its EV programs along with EV charging loads identified by classification algorithms and quantified by disaggregation algorithms. The platform can aggregate the loads in specific parts of the grid, such as a feeder or a set of homes served by a particular transformer. Based on the aggregate load forecasts and transformer ratings, the platform can implement charging protocols that protect transformers from overloading and reduce peak demand.
At Camus, we're building tools to help utilities prepare for and benefit from the growth of EVs. From mitigating local equipment impacts to helping reduce system-wide peak supply costs, we strive to help utilities engage with EVs in a smart and coordinated way.
With that in mind, we're excited to host a conversation with leaders from General Motors, Duquesne Light Company, and Connexus Energy on the topic of “How Utilities Can Ensure The Grid Is Ready For EVs.” This webinar will explore the challenges and opportunities related to EVs and how multiple parties—from OEMs to utilities to software partners—can work together for a smooth transition.
The live webinar will be held on May 16, 2023 at 12pm CST, and a recording will be available immediately afterwards. Register for free today.