Machine Learning Models for Behavioral Adaptation and Demand Forecasting of EV Users in Smart Grids

Abstract

Electric vehicle (EV) rapid increase of the number of EVs is shifting power grid systems today creating both opportunities and challenges for smart grid operations. Unlike traditional electrical loads where demand is static (usually predictable based upon seasons), EV charging load and therefore demand is a moving target that varies based on the EV user's behavior; their mobility patterns, EV charging prices, and whether conditions where they will charge their EVs. Accurate forecasting of EV demand and understanding how EV user's behavior changes with changing pricing is essential for maintaining both the reliability of our grid and our ability to optimize available energy resources, as well as creating the means by which renewable sources of energy can be successfully integrated into the grid. This paper provides a complete analytical framework that leverages machine learning approaches to conduct an analysis of EV user's behavior and EV charging demand forecasting related to the smart grid. This framework performs a data-driven analysis of how user responses to both the pricing of electricity (through Time-of-Use Tariffs) and the electric infrastructure itself will result in behavioral changes. Short- and medium-term load forecasting are accomplished using historical charging profiles of EV users, their mobility data, weather related datasets, the price of electricity based on a time-of-use pricing schedule, and other factors allowing for better predictive accuracy through the use of those combined sources of data. Simulation results demonstrate significant improvement in demand forecasting and peak load management with the incorporation of user behavior in forecasting models. The contributions of this work demonstrate the importance that intelligent analytics will have in creating stable economic, and user-driven energy resource management systems for the evolution of future smart cities.