XGBoost在超短期负荷预测中的应用  

Application of XGBoost in ultra-short load forecasting

作　　者：杨修德 王金梅 张丽娜 YANG Xiu-de;WANG Jin-mei;ZHANG Li-na(College of Physics and Electrical and Electronic Engineering,Ningxia University,Yinchuan 750000,China)

机构地区：[1]宁夏大学物理与电子电气工程学院,宁夏银川750000

出　　处：《电气传动自动化》2017年第4期21-25,共5页Electric Drive Automation

基　　金：国家自然科学基金项目(NO.51167015)

摘　　要：随着智能电网、电力负荷数据呈指数级增长,传统模型在处理海量负荷预测数据时显现出疲态、无法达到高效的问题。提出采用正则项限制模型复杂度、进行二阶泰勒展开、可并行计算特征分裂增益的XGBoost模型来解决这类问题。选用第九届电工数学建模A题负荷预测竞赛的某地区真实电荷数据作为样本,对2015-01-10日每隔15分钟的电网负荷进行预测,并利用平均绝对误差(MAE)、均方误差(MSE)、均方根误差(RMSE)对XGBoost与现有的GBRT及RF分别预测的96个时刻的数据进行分析。结果表明XGBoost对电网超短期负荷预测具有更好的拟合度和更高的精准度,可运用到实际负荷预测当中。With the rapid development of the intelligent power grid,the load data increases exponentially,and the weakness of the traditional forecasting model with low prediction efficiency and low accuracy appears in massive data processing.An XGBoost model is proposed to solve the problem by using the regular term to limit the complexity of the model,and to perform the second order Taylor expansion and to calculate the characteristic splitting gain in parallel.The actual charge data of a certain area of the load forecasting competition of the ninth electrician is used as a sample to predict the power load for every 15 minutes in 2015-01-10,and the average absolute error(MAE),Mean square error(MSE)and Root mean square error(RMSE) are used to analyze the data of 96 time points predicted respectively by the XGBoost,the GBRT and the RF softwares.The experimental results show that the XGBoost software has better fit and higher accuracy for short term load forecasting of the power grid,and can be applied to actual load forecasting.

关 键 词：XGBoost 梯度增强回归树 随机森林算法 大数据 超短期负荷预测 

分 类 号：TM71[电气工程—电力系统及自动化]