引入新特征的短期电力负荷预测

为了提高短期负荷预测的精度，综合分析了气象、日期等因素，并计算各特征与被预测负荷之间的相关系数，根据各特征与负荷之间的相关性，提出了一种将预测日前几天的负荷作为新特征进行负荷预测的方法。为了验证算法的普适性，采用支持向量回归、随机森林和梯度提升决策树3种机器学习算法，在2016-2018年我国北方某地的真实电力负荷和欧洲智能技术网络（EUNITE）竞赛负荷预测样本数据两个数据集上进行验证，并将预测结果与采用传统特征的算法进行了对比。预测结果显示，相较于传统方法，采用新特征后的短期负荷预测具有更高的预测精度。

Short-Term Power Load Forecasting with the Introduction of New Features

In order to improve the accuracy of short-term load forecasting,the weather,date and other factors are comprehensively analyzed,and the correlation coefficient between each feature and the predicted load is calculated.According to the correlation between each feature and load,a method of load forecasting is proposed with the load of the days before the forecast as a new feature.In order to verify the universality of the algorithm,three machine learning algorithms,support vector regression,random forest and gradient boosting decision tree,are used to verify on two data sets of the real power load of a certain place in northern China from 2016 to 2018 and the sample data of European Intelligent Technology Network (EUNITE) competition load forecast,and the prediction results are compared with the algorithm using traditional features.The forecast results show that compared with traditional methods short-term load forecasting with new features has higher forecast accuracy.