摘要
准确的电力负荷预测有助于电网安全稳定地运行.针对传统单算法单模型存在负荷预测精度低,误差波动大的弊端,提出一种多算法多模型融合的短期电力负荷预测方法.首先利用自助法对数据集进行多次采样.然后使用梯度提升回归树、最小二乘支持向量机和XGBoost算法对采样数据集进行学习,得到多个单预测模型.最后将预测点前一段时间的实际负荷值和对应时间点第一次学习到模型的预测值构成新的训练集,并利用XGBoost算法对此训练集进行在线二次学习,得到最终预测结果.利用上述方法对某市短期电力负荷进行预测,全年小时负荷预测结果的平均绝对百分误差(MAPE)低于单模型、多算法单模型和单算法多模型.结果表明,所提方法具有更高的预测精度.
Accurate power load forecasting of grid can make the power grid safe and stable operation. Directed at the problem that the traditional single-model has low prediction accuracy and large fluctuations in error,a short - term load forecasting method based on multi - algorithm & multi - model ensemble is proposed. First,the bootstrapping method was used to sample the data sets several times. Then,the sampled data set was learned by gradient boosting regressor tree,least squares support vector machine and XGBoost algorithm,and multiple single prediction models were obtained. Finally,the new training set was formed by the actual load value of the prediction point and the model load prediction value was obtained at the corresponding time point for the first time. The XGBoost algorithm was used to learn the training set online for two times,and the ensemble model was obtained. The model was used to predict the short - term power load of a city. The mean absolute percentile error of hourly forecasting results is lower than that of single - model,multi - algorithm & single - model and single - algorithm & multi - algorithm. The results show that the proposed method has higher prediction accuracy.
作者
任利强
张立民
王海鹏
郭强
REN Li-qiang;ZHANG Li-min;WANG Hai-peng;GUO Qiang(Institute of Information Fusion,Naval Aviation University,Yantai Shandong 264001,China)
出处
《计算机仿真》
北大核心
2019年第10期103-108,共6页
Computer Simulation
基金
国家自然科学基金重大研究计划资助项目(91538201)
泰山学者工程专项经费资助项目(Ts201511020)
关键词
电力负荷预测
自助法采样
多算法多模型融合
极限梯度提升算法
Power load forecasting
Bootstrap sampling
Multi-algorithm & multi-model ensemble
Extreme gradient boosting( XGBoot) algorithm
