基于XGBoost算法的短期电力负荷预报

Short-Term Electricity Load Forecasting Based on the XGBoost Algorithm

摘要: 短期电力负荷的精准预报是电力能源管理系统(EMS)合理安排生产调度计划、实现节能、经济运行的前提条件和重要保障。本文针对电力负荷波动特征具有不同时间尺度的周期相似性，根据前一天同一时刻、前一周同一时刻负荷和最近24小时平均负荷历史数据、融合气象数据以及工作日、节假日时间事件信息，采用梯度提升算法建立多信息融合的短期电力负荷极限梯度提升(XGBoost)模型，提前一天预测24点电力负荷变化趋势。通过某地区电网日负荷24点曲线预报的实验结果，表明所构建的电力负荷XGBoost预报模型相比随机森林、贝叶斯和KNN方法在计算速度和预测精度方面具有优势。

Abstract: High accurate forecasting of short-term power load is important to make reasonable production plan and scheduling task and achieve energy saving and economic operation of electric power management system (EMS). In this paper, the periodic similarity of power load fluctuation char-acteristics with different time scales is given. According to the historical data of the same time load of the previous day, the same time load of previous week and average load of the last 24 hours, meteorological data and the information about the working day and holiday event, a short-term power load extreme gradient boosting (XGBoost) model with multi-information fusion is built by using the gradient boosting algorithm. It can predict the trend of 24 o’clock power load ahead one day. The results, through a regional State Grid 24 o’clock curve forecast, show that the built XGBoost forecasting model has advantages over random forest, Bayesian and KNN methods in terms of speed and prediction accuracy.