考虑环境因素的电动汽车充电站实时负荷预测模型

Real-time Load Prediction Model of Electric Vehicle Charging Station Considering Environmental Factors

摘要:

为了减少电动汽车大规模集成到电网造成的不利影响，提出了一种能够实现充电站充电负荷精准预测的方法。该方法利用LightGBM（light gradient boosting machine）与XGBoost（eXtreme gradient boosting）模型构建线下?线上组合模型。考虑充电负荷、时间、温度、天气等历史数据，利用LightGBM模型初步建立充电负荷线下预测模型；基于XGBoost模型，以线下预测模型输出负荷和实际负荷的误差为优化目标，实时变化的交通流量为协变量，建立线上预测模型，并对初步预测结果进行误差修正。某市实际充电站预测结果表明，相比于随机森林（RF）、LightGBM模型、XGBoost模型、多层感知机（MLP）以及LightGBM?RF组合模型，该组合模型具有更高的预测精度，同时可以准确预测不同充电站的实时充电负荷。

Abstract:

To mitigate the adverse effects of large-scale integration of electric vehicles into the grid， a method for the precise prediction of charging station load is proposed in this paper. The method employs a combination of LightGBM and XGBoost to construct offline-online ensemble models. Historical data including charging load， time， temperature， and weather are considered. Firstly， a charging load offline prediction model is established using LightGBM. Based on the XGBoost model， with the error between offline prediction model output load and actual load as the optimization target， and the real-time varying traffic flow as a covariate， an online prediction model is developed， and the error correction is performed on preliminary prediction results. Predictions from actual charging stations in a certain city indicate that compared to random forest （RF）， LightGBM， XGBoost， multilayer perceptron （MLP）， and LightGBM-RF ensemble models， the ensemble model demonstrates higher prediction accuracy while accurately forecasting real-time charging loads for different charging stations.