Hybrid Machine Learning Forecasting for Online MPC of Work Place Electric Vehicle Charging
Graham McClone, Avik Ghosh, Adil Khurram, B. Washom, Jan Kleissl
Abstract
This work proposes a novel EV forecasting technique that predicts each EV’s arrival time (AT), energy demand (ED) and plug duration (PD) over the course of a calendar day using a hybrid machine learning (ML) forecast. The ML forecasts as well as persistence forecasts are then input in a model predictive control (MPC) algorithm that minimizes the electricity costs incurred by the charging provider. The MPC with the hybrid ML forecast reduced peak loads and monthly electricity costs over a base case scenario that determined costs for uncontrolled L2 charging: Reductions in weekday mean peak load during a 30 day summer time case study were 46.7% and 2.8% from the base case to ML MPC and persistence to ML MPC, respectively. Reductions in utility costs during the summer case study were 20.9% and 0.1% from base case to ML MPC and persistence to ML MPC respectively. Results are similar for a 30 day winter case study.