A charging start time prediction method for electric vehicles based on improved SVM and RF

In the context of the rapid popularization of electric vehicles and the growing advancement of V2G technology, the prediction of charging demand has become critical research directions. To tackle the challenge of forecasting the charging start time of EV, this paper proposes two improved algorithms, namely WMK-SVM and CW-RF, and further develops an EV charging start time prediction model based on the two methods. Specifically, three sequential steps are implemented in this research. First, the pending charging state is defined based on vehicle charging and driving behaviors. Second, feature engineering is carried out to extract discriminative features, where partial discrete features are transformed into frequency-based indicators through the statistical analysis of historical data to enhance the accurate recognition capability of the model. Third, on the basis of historical EV operation datasets, a pending charging state prediction module is constructed using the WMK-SVM algorithm, while a charging state prediction module is established via the CW-RF algorithm. Based on full-year operational data from multiple vehicles, the optimal parameters of the model are determined, and an in-depth analysis of its necessity, reliability and stability is carried out. By comparing with other machine learning methods, the proposed model highlights its superiority by reducing the prediction errors by approximately 20 min across the board, successfully verifying its capability to accurately predict the start time of vehicle charging. This helps the grid side predict the number of vehicles connected to the grid during the target time period based on the prediction results, thereby preventing excessive operational pressure on the power grid caused by a large number of connected vehicles during peak power consumption periods through measures such as electricity price adjustment. Ultimately, this strategy can improve the operational efficiency of the grid and be effectively applied to actual vehicle operation scenarios.