基于相似日和CNN-LSTM的短期负荷预测

Short-Term Load Forecasting Based on Grey Relational Analysis and CNN-LSTM

为充分发掘历史信息,解决气象数据不足影响预测精度的问题,采用灰色关联分析(GRA)选取天气相似日和CNN-LSTM混合神经网络的方法来预测电力负荷。利用GRA计算每日各气象因素与日总负荷的灰色关联度,再计算各日与典型日的相同气象因素之间的欧氏距离,将各气象因素的欧氏距离分别乘以对应因素的关联度,并将同一天的结果累加,得到一个综合得分。选取待预测日之前分数最低的5天作为相似日,将相似日各时刻的负荷数据输入CNN-LSTM网络中,预测出待预测日的负荷,通过与其他模型对比,验证了该方法的有效性。

This paper used grey relational analysis(GRA) to select days with similar weather conditions and explore more historical information.In addition,it employed the CNN-LSTM hybrid neural network method to predict power load and solve the problem of insufficient meteorological data affecting prediction accuracy.This research used GRA to calculate the grey relational grade between daily meteorological factors and overall load.In addition,it computed the Euclidean distance of the same meteorological factors between each day and the typical day.The Euclidean distance of each meteorological factor was multiplied by the relevancy of corresponding factors.The accumulation of the calculation results in the same day could obtain an overall score.This study took five days with the lowest score before predicted days as the similar days.It inputted load data into the CNN-LSTM network to forecast the load of prediction days.Compared with other models,the effectiveness of this method is verified.