End-to-End Smart EV Charging Framework: Demand Forecasting and Profit Maximization With Causal Information Enhancement

Abstract

Smart Charging of Electric Vehicles (EV) is a method of optimizing the EV charging schedule. As a result, it can provide more profit for EV charging stations. This profit usually comes from reducing the peak demand charge cost while still satisfying the EV user demand. Most charging stations let EV users input their demand. However, users can intentionally or unintentionally poorly estimate their demand, leading to lower profit for EV charging stations. In this paper, we propose an end-to-end framework of Smart Charging that aims to maximize the profit of charging stations while satisfying EV user demand. Our framework consists of two main modules. First, the demand forecasting module focuses on predicting the EV user's energy demand and session duration using various machine learning techniques, e.g., XGBoost, Random Forest (RF), and TabNet. Second, the EV charging schedule optimization based on model predictive control (MPC) has been employed to optimize the charging schedule. The optimization module has been further improved by using the feedback from causal information and behavior of EV charging profiles called Constant Current Constant Voltage (CC-CV). The experiment was conducted on simulation with real EV charging data. The results show that our framework outperforms the baseline with higher profits of $616.28 a month, a 27.18% improvement. In addition, our forecasting module can avoid user biases from user inputs and predicts more accurately-the winning model is XGBoost with a symmetric mean absolute percentage error (SMAPE) of 10.72% and 11.85% for session duration and energy demand, respectively.