An Integrated Algorithm for Short Term Charging Load Prediction of Electric Vehicles Based on a More Complete Feature Set

Abstract

The large-scale development of electric vehicles has made accurate short-term charging load prediction increasingly important for ensuring the safe operation of the power grid. To address issues of poor generalization ability and overfitting in single models, this paper proposes an integrated stacking prediction algorithm that combines three models: category boost (CatBoost), light gradient boosting machine (LGBM), and ridge regression (RR), using a stacking integration framework. The Cat–LGBM–RR model uses an internal stacking mechanism, where the RR model calculates the final prediction results after the CatBoost and LGBM models generate the necessary metadata. The effectiveness of the proposed model is demonstrated using load data from a new energy charging pile organization in a province of China. This paper’s contributions include: (1) proposing a stacking integration-based prediction algorithm; (2) providing a more thorough feature construction approach; (3) comparing and verifying the performance using enterprise real data sets and a variety of reference models. Numerical examples show that the mape of the Cat–LGBM–RR model was 4.52%. Compared with other models, it has precision advantage.