电动车充电需求预测XGBoost决定系数达0.980胜Transformer：《科学报告》用55座充电站15304条记录比较五种模型
《科学报告》11月13日刊出一项电动车充电需求预测研究。研究构建了可解释的多模态预测方法，融合充电会话数据、天气信息、电网价格、电网运行信息、站点占用与交通信号状态。
数据来自55座充电站在2019年1月至2020年7月间的15304条充电会话记录，研究据此构建了包含13680个小时的多模态数据面板，并比较了线性回归、随机森林、XGBoost、长短期记忆网络与作者提出的基于注意力的Transformer网络共五种架构。
结果显示，XGBoost取得最低均方根误差与最高决定系数，均方根误差为5.63千瓦时、R²为0.980；随机森林的平均绝对误差与平均绝对百分比误差略低，分别为1.67千瓦时与7.39%，XGBoost对应数值为1.72千瓦时与8.36%。
两种树模型均明显优于Transformer网络与LSTM网络，后两者均方根误差分别为27.98千瓦时与30.02千瓦时，R²分别为0.513与0.440。
对最佳模型的SHAP分析显示，近期充电需求、变压器负载与一天中的时段是最重要的预测变量。作者结论称，在当前数据规模下，树模型配合SHAP在单站聚合充电需求预测上兼具准确性与可解释性。论文作者为Md Mushfiq Us Salehin，数字对象标识符为10.1038/s41598-026-78819-1。
（来源：《科学报告》）
XGBoost Reaches R² of 0.980 for EV Charging Demand, Ahead of Transformer, in Scientific Reports Study of 55 Stations
Scientific Reports published a study on electric vehicle charging demand forecasting on November 13, 2026. The study developed an explainable multi-modal forecasting approach fusing charging session data, weather information, grid prices, grid operating information, station occupancy and traffic signal status. Using 15,304 charging session records from 55 charging stations between January 2019 and July 2020, the author constructed a 13,680-hour multi-modal data panel and evaluated five forecasting architectures: linear regression, Random Forest, XGBoost, a Long Short-Term Memory network and a proposed attention-based Transformer network. XGBoost achieved the lowest RMSE and highest R², at RMSE of 5.63 kWh and R² of 0.980, while Random Forest achieved a marginally lower MAE and MAPE, at 1.67 kWh and 7.39 percent, versus XGBoost's 1.72 kWh and 8.36 percent. Both tree-based models substantially outperformed the Transformer network, at RMSE of 27.98 kWh and R² of 0.513, and the LSTM network, at RMSE of 30.02 kWh and R² of 0.440. SHAP analysis of the best model identified recent past charging demand, transformer loading and the hour of the day as the most significant predictors. The paper, by Md Mushfiq Us Salehin, carries DOI 10.1038/s41598-026-78819-1.
(Source: Scientific)
来源：《科学报告》
发布：GSM·美国（发布链接）
图片来源：图：一处电动车充电站与正在充电的车辆。Julian Herzog (Website)，CC BY 4.0，经 Wikimedia Commons。 / Photo: Julian Herzog (Website), CC BY 4.0, via Wikimedia Commons.
GSM Global · 本篇英文版 English edition of this story
