基于MSA-LSTM的短期电力负荷预测模型  

Short-Term Power Load Forecasting Model Based on MSA-LSTM

作　　者：冯勇[1] 张校铭 FENG Yong;ZHANG Xiao-ming(Faculty of Information,Liaoning University,Shenyang 110036,China;Financial Assets Department,State Grid Liaoning Electric Power Co.,Ltd.Liaoyang Power Supply Company,Liaoyang 111000,China)

机构地区：[1]辽宁大学信息学部,辽宁沈阳110036 [2]国家电网辽宁省电力有限公司辽阳供电公司财务资产部,辽宁辽阳111000

出　　处：《辽宁大学学报(自然科学版)》2024年第4期360-367,共8页Journal of Liaoning University:Natural Sciences Edition

基　　金：辽宁省属本科高校基本科研业务费专项资金资助项目(LJKLJ202414)。

摘　　要：短期电力负荷预测是对未来较短时间内的电力负荷进行预测的过程.当前短期电力负荷预测存在不确定性强、负荷变化快、计算成本高等问题.针对以上问题,本文通过融合多头自注意力(Multi-head self-attention,MSA)机制与长短期记忆(Long short-term memory,LSTM)网络,提出了一种新型的MSA-LSTM模型用以进行短期电力负荷预测.该模型旨在处理电力负荷数据的时间依赖性和复杂性,增加MSA结构作为LSTM网络结构的输入模块,增强LSTM网络的长期记忆能力和对关键时间序列特征的捕捉能力.对目标数据集的实验验证表明MSA-LSTM模型在预测精度和稳定性方面均优于传统LSTM模型和双向长短期记忆(Bidirectional long short-term memory,BiLSTM)模型.利用第九届电工杯电力负荷数据和气象数据的数据集对本文所提出的模型进行十折交叉验证,相比LSTM模型和BiLSTM模型,MSA-LSTM模型的平均均方误差(mean square error,MSE)分别减少4.285%和2.672%,误差的标准差分别减少6.575%和3.406%.研究结果表明,该模型在电力系统负荷预测中具有较高的应用价值,对优化电力系统运营和决策支持具有重要意义.Short-term power load forecasting refers to the process of predicting the power load over a relatively short period of time in the future.The current short-term power load forecasting faces problems such as strong uncertainty,rapid load changes,and high computational costs.In response to the above issues,this article proposes a novel MSA-LSTM model for short-term power load prediction by integrating multi-head self attention(MSA)mechanism and long short-term memory(LSTM)network.This model aims to handle the time dependence and complexity of power load data,add MSA structure as the input module of LSTM network structure,enhance the long-term memory ability of LSTM network and the ability to capture key time series features.Through experimental verification on the target dataset,the MSA-LSTM model outperforms traditional LSTM models and BiLSTM models in terms of prediction accuracy and stability.Using the dataset of the 9th Electric Power Cup power load data and meteorological data,a ten fold cross validation was conducted on the model proposed in this paper.Compared with the LSTM model and the bidirectional long short-term memory(BiLSTM)model,the average mean square error(MSE)of the MSA-LSTM model was reduced by 4.285%and 2.672%,respectively;The standard deviation of errors decreased by 6.575%and 3.406%,respectively.The research results indicate that the model has high application value in power system load forecasting and is of great significance for optimizing power system operation and decision support.

关 键 词：短期电力负荷预测 多头自注意力机制 LSTM 

分 类 号：TP302.1[自动化与计算机技术—计算机系统结构]