Multi-Task Gaussian Process Learning for Energy Forecasting in IoT-Enabled Electric Vehicle Charging Infrastructure

Abstract

Modern electric vehicle (EV) charging infrastructure represents an application of an Internet of Things (IoT) framework in power systems where communication networks are the basis to transmit, monitor, and control a network of EV charging stations. This paper takes advantage of the high-resolution EV charging data made available by IoT-enabled charging stations for developing models that would forecast the energy demand of EV charging stations. More superficially, this paper proposes a Multi-Task Learning (MTL) method that enhances the estimation and forecast accuracy by fusing data from multiple charging stations via a Gaussian Process (GP) model in order to extract common features of the charging data. The proposed GP model learns a shared covariance function over all data sources, and transfers the meaningful knowledge of one charging station to other stations, which helps improve the forecasting accuracy. Data from multiple EVs charging stations in Utah is used in a case study to implement and validate the proposed forecasting model. The implementation results show that the proposed method significantly outperforms the state-of-the-art forecasting methods.