Ruminations on the Stochastic Modeling of Electric Vehicle Charging Behavior

In the last decade, the proliferation of wind and solar generation has made the power sector cleaner and has been a significant step towards decarbonization. However, progress on the electricity grid alone won’t single-handedly address climate change. This will require decarbonizing other greenhouse gas-emitting sectors of the economy like transportation, which in 2016 surpassed power generation as the largest source of emissions in the US.

With a fully renewable electricity grid becoming increasingly attainable, electrifying cars, trucks, and other motorized vehicles is the straightest route to a cleaner transportation system. Thankfully, the electric vehicles (EV) industry has also been growing rapidly. However, as highlighted on an episode of The Interchange Podcast in October 2020, electrifying transportation is creating new challenges for utilities as they ensure the influx of charging demand becomes a flexible resource rather than a grid stressor. 

How can utilities help EVs reach their full potential as a distributed asset? 

It starts with simulating charging behavior as accurately as possible and using these simulations to experiment with various policies and incentives. However, modeling these scenarios is challenging due to the complexity of EV drivers and the limitations of existing datasets. 

As a mode of transportation, EVs differ from other distributed energy resources (DERs), such as water heaters, that are geographically constrained and exhibit more predictable behavior. EVs are DERs on wheels, capable of interacting with the grid at multiple locations and subject to the random and unique schedules of individual drivers.

Additionally, existing data on EV charging behavior is limited in availability and scope. Although the penetration of EVs has increased in the last decade, grids where EVs make up a significant share of electricity demand are still scarce. The data that is currently available is also unlikely to accurately reflect future high-penetration scenarios because early adopters have generated the majority of data and may not exhibit the same behavior as mainstream users. 

Most modeling methods, especially those reliant on vast amounts of data to train, are not sophisticated enough to simulate the range of behaviors that EV users exhibit and exceed the limitations of available datasets. What would an analytics tool look like that would be able to properly plan for these future scenarios while compensating for these limitations? 

At Kevala, we aggregate public and private data to help distribution system operators make smarter infrastructure decisions. To properly model EV-related impacts on distribution networks, our data science team has been researching promising approaches that allow utilities to simulate the effects of high EV penetration while providing the functionality to experiment with different rate structures and charging locations. 

Kevala’s data science team has outlined an agent-based methodology that models each EV user individually. This model structure enables powerful and configurable simulations of distribution grids while accounting for the complexity of EV driver behavior even when data is limited. We are excited for the opportunity to implement this algorithm and build the dynamic EV simulation tool outlined in this accompanying whitepaper. For more details, see our full outline of the approach here