Predicting Excess Energy and Estimating Users for Vehicle-to-Grid (V2G) Services Using Machine Learning

Vehicle-to-grid (V2G) technology provide significant benefits to both Electric Vehicle (EV) owners and electric utilities. The utilization of V2G technology offers incentives to EV owners, by shedding excess energy back to the grid. This paper addresses two problems: (1) predicting excess energy, and (2) estimating user availability for V2G services. First, by using the input features such as charger connect time, delivered kWh and requested kWh, the prediction of excess kilowatt-hour (kWh) are investigated for the period from 2018-2021. For this problem, Machine Learning (ML) models such as Support Vector Regression (SVR), Long Short Term Memory (LSTM), Gradient Boosting Regression (GBR), and Random Forest (RF) are deployed. Missing values in the historical data are imputed by using interpolation technique with time-based order of data points. After this step, the number of users available for V2G services are examined by considering 15% and 30% of excess kWh from the charging stations. The analysis is conducted by using ML classification models such as Decision Tree (DT) and K-Nearest Neighbor (KNN). According to the preliminary results, LSTM model performs better with Mean Absolute Percentage Error (MAPE) of 3.12. RF as second best with lowest 3.59. In the context of classification models, DT performed better compared to KNN, with highest 89% and 84% accuracy, respectively.