ISSA优化Attention双向LSTM的短期电力负荷预测  

Short-term Power Load Prediction of Bidirectional LSTM with ISSA Optimization Attention Mechanism

作　　者：王金玉[1] 金宏哲 王海生 张忠伟[1] WANG Jinyu;JIN Hongzhe;WANG Haisheng;ZHANG Zhongwei(School of Electrical Engineering&Information,Northeast Petroleum University,Daqing 163000,China;Qingxin Oilfield Development Co.,Ltd,Daqing 163000,China)

机构地区：[1]东北石油大学电气信息工程学院,大庆163000 [2]庆新油田开发有限责任公司,大庆163000

出　　处：《电力系统及其自动化学报》2022年第5期111-117,共7页Proceedings of the CSU-EPSA

基　　金：国家自然科学基金资助项目(61873058)。

摘　　要：针对短期电力负荷数据的复杂性和多样性,提出一种含Attention的双向LSTM预测方法,简称Bi-LSTM-AT。该方法将电力负荷历史数据作为输入且考虑温度、湿度和日期类型因素的影响。通过建模学习构建网络模型,挖掘网络特征内部变化规律,通过映射加权和学习参数矩阵赋予Bi-LSTM-AT网络隐含状态相应的权重。同时,针对该模型超参数选择困难的问题,提出利用改进麻雀算法实现该模型超参数的优化选择,使得全年最后两天预测值的MAPE为0.42%、RMSE为0.29%和MAE为0.21%,验证了模型线性回归拟合能力的准确性和稳定性。In view of the complexity and diversity in short-term power load data,an Attention-containing bidirectional long short-term memory(LSTM)prediction method is proposed,which is referred to as Bi-LSTM-AT for short.This method adopts the historical power load data as its input and takes into account the influences of factors such as temper⁃ature,humidity and date type.Through modeling learning,a network model is constructed,and the internal variation rules for network features are mined.By mapping weighting and the learning parameter matrix,the corresponding weight value of hidden state of Bi-LSTM-AT network is given.At the same time,aimed at the problem of difficulty in se⁃lecting the hyperparameters of the model,an improved sparrow search algorithm is proposed to optimize the selection of hyperparameters.The results of an example show that the mean absolute percentage error,root mean squared error and mean absolute error for the last two days in one year are 0.42%,0.29%and 0.21%,respectively,which verifies the ac⁃curacy and stability of the linear regression fitting capability of the proposed model.

关 键 词：电力负荷 预测 长短期记忆 注意力机制 改进麻雀搜索算法优化 

分 类 号：TM714[电气工程—电力系统及自动化]