基于LSTM+Attention模型的典型配电台区短期负荷预测方法  

Short Term Load Forecasting Method for Typical Distribution StationsBased on LSTM+Attention Model

作　　者：黄朝凯 林洪浩 王柯成 陈晓瑜 卢海明 HUANG Chaokai;LIN Honghao;WANG Kecheng;CHEN Xiaoyu;LU Haiming(Shantou Power Supply Bureau of Guangdong Power Gird Co.,Ltd.,Shantou 515041,China)

机构地区：[1]广东电网有限责任公司汕头供电局,广东汕头515041

出　　处：《微型电脑应用》2024年第8期88-91,共4页Microcomputer Applications

基　　金：南方电网公司科技项目(030500KK52200002(GDKJXM20200706))。

摘　　要：为解决配电区短期负荷难预测,导致负荷调度不均匀问题,研究基于长短期记忆网络+注意力机制(LSTM+Attention)模型的典型配电台区短期负荷预测方法。考虑节假日、正常作息以及气温等不同变化因素导致的短期负荷变化,确定区域划分单元中各负载与影响因素间关系,集群分析负载抽样数据,根据不同成分特征向量的贡献率,统计负荷变化影响因素数量,利用长短期记忆网络充分存储历史信息,凭借注意力机制模型确定输入向量和网络中隐含层的实际状态值以及地区负载的主要参数,计算归一化后电力负载的实际概率密度,得出短期负荷预测结果。实验结果表明,所提方法响应速度快,预测准确度高,预测效果和拟合优度理想。In order to solve the problem of uneven load scheduling caused by the difficulty of short-term load forecasting in distribution areas,a short-term load forecasting method for typical distribution areas based on LSTM+Attention model(short long-term memory network+Attention mechanism model)is studied.It considers the short-term load changes caused by different change factors such as holidays,normal work and rest and temperature,determines the relationship between each load and influencing factors in the regional division unit,analyzes the load sampling data,counts the number of influencing factors of load changes according to the contribution rate of different component eigenvectors,and uses the long short-term memory network to fully store historical information.Based on the Attention mechanism model,the input vector,the actual state value of the hidden layer in the network and the main parameters of the regional load are determined,and the actual probability density of the normalized power load is calculated to obtain the short term load forecasting results.The experimental results show that the proposed method has fast response speed,high prediction accuracy,and ideal prediction effect and goodness of fit.

关 键 词：长短期记忆网络 Attention模型 短期负荷 负荷预测 配用电信息 

分 类 号：TM715[电气工程—电力系统及自动化]