Robert A. F. Currie, Teddy Ward, James L. Carney, Greg Mandelman, Margot C. Everett, Aram Shumavon, Nathan Phelps, Lindsay Griffin, Stephan Roundtree
Many countries are enhancing the planning and interconnection process to accelerate the interconnection of solar, wind, and other technologies to the distribution grid. The electrification of transportation, however, is going to have a much larger impact on utility planning and operations, essentially transforming utilities into providers of foundational mobility-related services. The speed of electric vehicle (EV) adoption is increasing and is an order of magnitude faster than the pace at which utilities build new distribution and transmission infrastructure. EV adoption scenarios must be sufficiently accurate, granular, and specific to identify critical grid investment needs. Identifying probable EV adoption and usage patterns and then modeling their impacts on the power grid is a complex process that will be fundamental to decarbonizing the grid.
It is generally held that data on the grid are considered critical energy/electric infrastructure information and are to be protected from a national security perspective. However, hosting capacity maps and associated datasets are increasingly being made available in many markets. These resources usually do not provide detailed information about the grid at a component level but can help guide the developer community toward locations that are a good fit for solar, storage, and EV developments, among other types of distributed energy resources (DERs). In addition, publicly available satellite and street view imagery contains very large amounts of useful information that could be harnessed to support modeling the energy transition.
Adoption propensity modeling is concerned with determining who is likely to adopt a DER, the size of the adopted DER, the subsequent energy behaviors, when that adoption will most likely occur, and what technologies it will involve. These processes need to be quick and scalable; too often, lengthy studies are required to establish a common basis for discussion among electric utilities, DER developers, and customers.
In this article, the authors use public data to explore potential EV adoption scenarios and their associated grid impacts, prior to exploring the amount of solar and storage capacity that would be required to meet the new demand profile in the area. The authors also consider the various stakeholders impacted by electrification, exploring how regulation and policy are evolving to ensure an equitable energy transition.
Data and analytics are both critical in planning for the electrification of transportation. The collection, processing, and presentation of structured and unstructured data from multiple domains to a range of predictive analytic modules are required to identify the relevant behaviors and scenarios.
In recent years, open data initiatives have become increasingly prevalent across many sectors, including in the electric sector. OpenStreetMap is an example of a public source of a range of data on power networks and generators. Furthermore, utilities across the world are publishing maps of their infrastructure to enable project developers to evaluate the likelihood of interconnecting a new generation or storage project to the grid. An example of this is the hosting capacity maps for New York State utilities. Another leading example of this is the requirement for distribution utilities in the United Kingdom to share their power system models publicly using the common information model. These resources can be harnessed in a number of ways and provide an important reference point for any analysis concerned with the impacts of the electrification of transportation on the power grid.
To understand the location and size of solar installations, the authors utilize satellite imagery and computer vision algorithms. The authors collected overhead satellite imagery for 8,479 known photovoltaic (PV) systems and 6,732 rooftops that were known to not have a PV system. A “two-level” model is then employed: a single-label classification model and an object detection model. The single-label model is much faster, much cheaper, and much easier to use for creating training data. Single label refers to a model where the trained data have exactly one label, opposed to multiple, in this case, “existing PV” or “no PV.” The object detection model gives the precise location of solar PV systems and potential sizing information. Both models were trained on a range of imagery and provided an initial quick assessment of the installed solar in an area. Figure 1 shows an example of PV systems identified in this way, for an unspecified location in California. Each of the identified PV systems is identified as belonging to a specific parcel or land, given an estimated kW rating and production profile (using weather data), and associated with the distribution feeder (orange line) supplying the parcel.
Many countries make information available (e.g., census data) about the population. In the United States, census tract information reveals the demographics of the population, typical commuting modes, how people heat their homes, and income levels. Transportation analysis zones (TAZs) provide additional information regarding the typical means of transportation within a zone (typically comprising multiple census tracts).
When it comes to the electrification of transportation, top-down scenarios exist that forecast expected adoption rates at a geographic level, including counties, census tracts, or even utility service territories. Many utilities are introducing “EV-friendly” rate designs, which can play a role in the affordability of EVs.
When it comes to the electrification of everything (i.e., heating, cooling, and transportation), the natural and built environment must be considered within the modeling and projection of technology adoption and its impact on the power system. Modeling the rural environment entails the consideration of many factors, including wetlands, flood prevention zones, wildfire risk areas, endangered species habitats, environmental protection zones, and weather (typical meteorological year or specific event). These factors must be considered when exploring the development of electric infrastructure.
To model the built environment, it is necessary to understand the size, footprint, and type of each building; the heating, ventilation, and air conditioning systems used; and to which piece of grid infrastructure each building is connected. Much can be learned from census and TAZ information, but a means of disaggregating this information and performing address-specific analysis is needed to capture the kinds of localized and clustered adoption patterns that can drive grid infrastructure enhancements.
When it comes to EV adoption-propensity modeling, understanding the built environment as it relates to the grid and TAZs is crucial. In addition, it is critical to automate the identification of suitable locations for EV fleet adoption and to optimize the deployment strategy, including bus depots, delivery truck depots, industrial facilities, and the like. A significant limitation of top-down forecasting approaches is where the space for DER adoption is assumed to be universally available at the same rate across the grid. As with the mapping of rooftop solar discussed above, the problem of locating existing fleets can also be approached as a computer vision problem. For example, the authors collected overhead satellite imagery for 1,864 known fleets and 588 groups of vehicles that were known to not be fleets. As illustrated in Figures 2 and 3, the images were labeled as either light duty (LD) (e.g., taxis and pickup trucks), medium duty (MD) (e.g., U-Haul and school buses), or heavy duty (HD) (e.g., semitrucks and cement mixers). The authors then used these data to train a model using Google Cloud’s Vertex AI platform.
This approach supports the quick identification of existing potential fleet candidates that can then be incorporated into further analysis, helping the planner to consider the kinds of technology adoption and behavior that can be expected in the area in question.
To identify how EV adoption will progress over time and location (i.e., all parcels), a detailed understanding of the economic and social dynamics of adoption propensity is required. Also, to state the obvious, EV load is mobile.
Historically, power system planning has been focused on extreme scenarios that capture peaks and troughs on the system, both for a fully intact grid and during contingencies. Harnessing public data sources and augmenting these with utility data or predictive models (trained with utility or customer data) can provide a very effective means of performing address-specific modeling that can be quickly aggregated up to the feeder and substation level. This type of analysis does not replace detailed engineering work but rather provides additional information for where deeper analysis is required, while also supporting quick and useful econometric modeling scenarios.
To demonstrate the kinds of insights that can come from the gathering of publicly available data and the allocation of these data to publicly available grid data, the authors selected an undisclosed distribution system location in New York State for study. The host utility has stated that across the territory the headroom for a new EV load is around 300 MW. What is not clear is where EVs are likely to appear on the system and how this corresponds to the capacity available at a local level.
The goal of this public data case study is to perform a range of EV adoption scenarios and to analyze the impacts on the distribution system. This case study will allow us to generate insights into the grid impacts of each scenario and the likely capacity of solar and storage required to mitigate any grid stress under each scenario.
The authors undertook a seven-step process to perform this analysis, as described in the following sections.
In step 1, the authors produced nine EV adoption forecasts across 2025, 2030, and 2035 as shown in Table 1. Each adoption forecast was analyzed according to three EV charging use cases, namely:
Table 1. The total percentage of EV ownership by year, for the low, medium, and high scenarios.
The low forecast assumes that EV sales have peaked at 5.33% market share by 2035 (based on data from the Alliance for Automotive Innovation) and that EV penetration will only increase by that amount for the estimated 943,000 vehicles sold per year in the state of New York (according to the New York Department of Motor Vehicles), for the duration of the study (2025–2035). The medium forecast shows what happens if the governor’s goals that 35% of new vehicle sales be zero-emission vehicles (ZEVs) in 2026, 68% by 2030, and 100% by 2035 are achieved, collectively resulting in ∼65% of the vehicles in New York being ZEVs by 2035. Finally, the high forecast shows what happens if the state goal of 1,000,000 ZEVs by 2025 is achieved and the compound growth rate required to reach that milestone is extrapolated, resulting in 100% of the vehicles in New York being ZEVs by 2035.
Each forecast contains important characteristics about the adopted vehicles. The EV duty indicates whether the vehicle is LD or MD/HD. Within the duties, vehicle counts are further distinguished by body type. The study employed simplified body types to reduce unnecessary analytical complexity while retaining all the critical features necessary to generate a robust load forecast. LD vehicles were separated into either large cars or SUVs, while MD/HD vehicles were divided into class 3–6 vehicles for MD and class 7–8 for HD vehicles. All body types were assumed to have EV powertrains, propelled entirely by a battery and thus consuming significantly more energy than plug-in hybrid EVs.
Step 2 involved determining the specific vehicle characteristic inputs to assign to each vehicle duty/body type combination. These inputs fall into two categories:
The two key inputs for a vehicle battery are its capacity, which indicates the amount of energy the battery can store in kilowatthours (kWh), and the battery range, which indicates the estimated distance the vehicle is expected to travel starting at a fully charged state. By dividing the battery capacity (kWh) by its range (e.g., miles), the battery’s efficiency (kWh/miles) can be estimated. This value indicates how much energy the vehicle will require to travel a given distance. This study adapted its vehicle battery capacity and range assumption from the California Energy Commission (CEC), published in Appendix B of its 2021 Assembly Bill 2127 Electric Vehicle Charging Infrastructure Assessment (Revised Staff Report) (CEC AB 2127 report).
Once battery capacity and range inputs have been determined, the amount of miles a vehicle is expected to travel (i.e., its annual VMT) is required for calculating the amount of energy a vehicle will require over the year. The VMT could be explored for variations throughout the year; however, for this study, the authors assumed a reduced VMT at the weekend. The study used the following sources for LD and MD/HD VMT:
The CTPP data were used to understand the unique commute patterns of every TAZ that was in the investigated area. The CTPP flows record the estimated number of commutes to and from every TAZ and can be used to determine the total number of VMTs by commuters in the study area.
Step 3 involved forecasting the amount and type of EV supply equipment required to support each scenario’s number and type of EVs. The study followed the CEC’s Assembly Bill 2127 report’s methodology of using a vehicle-to-charger ratio for determining the number and type of EV chargers that are required to support a given vehicle population. Charger types were limited to single-family homes and workplace chargers.
Step 4 involved allocating the charging of these forecasted EVs to specific feeders. For MD/HD, vehicles were located and assigned using computer vision, whereas for LD, vehicles were allocated via heuristic. For MD/HD, we used the model for identifying fleets described in the section “Understanding Land Use” to locate premises that the model identified as containing existing nonelectric fleets of vehicles, with confidence greater than our selected confidence threshold of 99.9%. This enabled us to reduce the 180,613 known premises in the service territory to 586 locations (shown in Figure 4) that were most likely to have existing nonelectric fleets of vehicles.
This approach allowed us to develop forecasts of charge shapes that were specific to the premise-level makeup of the individual circuits, rather than assuming equal adoptions across all circuits. The scenarios are then applied to provide a range of forecasts of how many of these fleets will convert to EVs. For instance, as shown in Figure 5, the locations shown with green dots can then be represented with hourly charging load forecasts over the course of the year.
We approached the problem of locating future LD EV adopters with a heuristic model. This model utilized the factors presented in Table 2.
Table 2. The factors utilized in heuristic model for LD EV adoption.
This approach resulted in the heat map of adoptions illustrated in Figure 6. Each EV is a two-pixel radius point, where the opacity increases from 0 to 1 based on the relative likelihood of a premise being used to charge an EV. Note that workplace charging scenarios concentrate all the load on very few locations, whereas residential charging scenarios spread the load across the whole territory (albeit not evenly). This simplified approach can and should be refreshed with new adoption data over time that capture changes in adoption behavior. These predictions could also be validated by comparing with known building permits for at-home chargers.
Step 5 involves developing vehicle charging load shapes. As described in step 2 and step 3, the amount of energy delivered is calculated using a given vehicle’s battery efficiency (kWh/mile) and its annual VMT and the demand required of the distribution system to provide this energy is determined by the type of charger and its capacity level. The hourly variations associated with the “plug-in” and “plug-out” charging patterns were derived using travel and commute data sourced from the census. CTPP table B302104 gives the average and standard deviation of the time that all vehicles leave home (usually, but not always, in the morning), and CTPP table B102217C gives the amount of time that these vehicles take to arrive at their destinations. By combining these two data sources, we can estimate the time that commuters arrive at work and the time that they return home (usually in the evening).
Step 6 involved looking at overloads on the different circuits, after allocating EVs. A simple thermal assessment based on location and utility reported asset capacity was performed. More complex analysis of power flows and voltages would be feasible with more detailed power system model data. The charge shapes from step 5 were compared with the reported headroom on each circuit, according to the host utility. An example of the thermal violations is shown in Figure 7.
The “charge at work” scenario in 2025 shows some broadly similar results in terms of the feeders that exceed 100% of rated capacity across all three adoption levels, suggesting that in the short term, there could be some priority investments in grid reinforcement in those areas.
The high scenario in 2035 (Figure 8 and Table 3) shows some broadly similar results in terms of the feeders that exceed 100% under different charging strategies. The “charge at work” strategy results in the most significant overloading of distribution feeders, with “charge at home” and “mixed” strategies providing similar results. The mixed scenario assumed that 50% of all charging was done at home, and 50% was done at work. As was the case with the example given above for “charge at work” in 2025, it is clear that there are consistently overloaded parts of the distribution system.
Table 3. The number of circuits exceeding the thermal limit, by year, charge strategy, and scenario.
As was discussed earlier, the host distribution utility has published the available capacity for new EV connections. In Table 4 we show the additional MW capacity that is needed across all ∼170 circuits to support the low, medium, and high scenarios across all charging strategies. The “charge at work” scenario has the largest requirement for additional MW capacity in all the scenarios, with nearly 200 MW of capacity required on top of the ∼300 MW the host distribution utility has said is available for EV load.
Table 4. The MW exceedance of thermal limits, by year, charge strategy, and scenario.
Step 7 involved estimating the adoption of solar PV and battery storage required to mitigate this new load and address the thermal capacity overloads identified in the previous step. This modeling could include more detailed optimization of the placement, sizing, and operation of solar PV and battery storage; however, for the purposes of demonstrating “quick” insights that can support initial planning for electrification at a high level, a simplified approach was taken. Based on analysis of the distribution system constraints identified in step 6, the capacity of solar PV and battery storage to mitigate the peak distribution system overloads was calculated. This capacity was calculated by determining the MW of local PV production required to remove any daytime overloads and the MW of battery storage capacity that would need to be available at the time of the overload in the evening (assuming a simple power to energy ratio of 1:4 for battery storage, e.g., 1 MW/4 MWh). This analysis provides quick insight into the volume of DERs such as solar PV and battery storage that could be required to be deployed and managed to avoid overloads on the distribution system.
As shown in Table 5, the amount of solar PV required increases with each of the adoption scenarios, with the most being required for the “charge at work” scenario and the high scenario of 100% adoption by 2035. In this scenario, 530 MW of solar PV is needed; this is an impractical number and suggests a need for more in-depth modeling of transmission and distribution infrastructure to meet the needs of electrification.
Table 5. The PV capacity required (MW), by year, EV charge strategy, and scenario.
The simplified approach to modeling solar PV (Table 5) and battery storage (Table 6) gives an indication of the power needs of the distribution area. Supporting the high scenario will require significant additional capacity, and a cost–benefit analysis is required to evaluate the best mix of solar PV, batteries, and grid upgrades to support this distribution system.
Table 6. The battery storage power required (MW), by year, EV charge strategy, and scenario.
The power rating of solar PV and batteries needed varies by scenario and EV charging strategy but remains significant in 2030 and 2035 under the medium and high EV adoption scenarios. This analysis also presumes that the solar PV and batteries would be exclusively available for the management of the peak load due to EV charging. This issue raises several questions around policy developments to enable coordinated and integrated decision-making concerning transmission and distribution upgrades and the local DER potential to enable electrification in an area such as this. It is clear from going through the steps laid out in this article that the demand for EVs in the area being studied will have a significant impact on the planning and operation of not only the distribution system but also the transmission system as well as local DERs.
This article has demonstrated that while using publicly available data, it is feasible to create scenarios of EV adoption and to identify the grid impacts. This kind of “quick” analysis can help bring insight into the issues and locations that require further and deeper analysis. However, this kind of analysis can very quickly support the utility in understanding the stakeholder communities impacted by the electrification of transportation in a particular area, including the following:
In this article, with the example of New York, the authors have shown that publicly available data can be used to create a range of insights about the impacts on the distribution grid of the electrification of transportation. A compelling need exists to increase the speed at which these insights can be generated and used, to support the identification of the need for deeper analysis, more complex planning, and consideration of and engagement with the stakeholder cohort for a particular area.
In addition, a number of avenues are available that need to be explored in planning that could help identify the right strategy for any area looking to electrify:
“Electrification impacts study: Part 1,” California Public Utilities Commission, Kevala Inc., San Francisco, CA, USA, 2023. Accessed: Aug. 3, 2023. [Online] . Available: https://docs.cpuc.ca.gov/PublishedDocs/Efile/G000/M508/K423/508423247.PDF
Joint Utilities of New York, New York, NY, USA. [Online] . Available: https://jointutilitiesofny.org/
“The common information model (CIM) regulatory approach and the long term development statement,” Office of Gas and Electricity Markets, London, U.K., 2022. Accessed: Aug. 3, 2023. [Online] . Available: https://www.ofgem.gov.uk/sites/default/files/2022-01/The%20Common%20Information%20Model%20%28CIM%29%20regulatory%20approach%20and%20the%20Long%20Term%20Development%20Statement.pdf
“Assembly bill 2127 electric vehicle charging infrastructure assessment: Analyzing charging needs to support zero-emission vehicles in 2030,” California Energy Commission, Sacramento, CA, USA, CEC-600-2021-001, 2021.
“Disadvantaged communities interactive map,” New York State Energy Research and Development Authority, New York, NY, USA. Accessed: Aug. 3, 2023. [Online] . Available: https://www.nyserda.ny.gov/ny/Disadvantaged-Communities
Robert A. F. Currie, Teddy Ward, James L. Carney, Greg Mandelman, Margot C. Everett, and Aram Shumavon are with Kevala, Inc., San Francisco, CA 94133 USA.
Nathan Phelps, Lindsay Griffin, and Stephan Roundtree are with Vote Solar, Oakland, CA 94612 USA.
Digital Object Identifier 10.1109/MPE.2023.3308237
Date of current version: 19 October 2023
1540-7977/23©2023IEEE