基于注意力机制的LSTNet日前电价预测  

LSTNet Day-ahead Electricity Price Prediction Based on Attention Mechanism

作　　者：李璐[1,2] 阚小瑞 毕贵红 范玉瑞[3] 朱泽良 周旭龙 LI Lu;KAN Xiaorui;BI Guihong;FAN Yurui;ZHU Zeliang;ZHOU Xulong(Faculty of Electric Power Engineering,Kunming University of Science and Technology,Kunming 650500,China;College of Engineering,Design and Physical Sciences,Brunel University of London,London UB83PH,UK;Department of Civil and Environmental Engineering,Brunel university of London,London UB83PH,UK)

机构地区：[1]昆明理工大学电力工程学院,云南昆明650500 [2]伦敦布鲁内尔大学工程、设计和物理科学学院,英国伦敦UB83PH [3]伦敦布鲁内尔大学土木与环境工程学院,英国伦敦UB83PH

出　　处：《电力科学与工程》2025年第4期1-10,共10页Electric Power Science and Engineering

基　　金：国家重点研发计划资助项目(2022YFB2700011)。

摘　　要：为了提高日前电价预测精度,提出了一种基于注意力机制的长期和短期时间序列网络日前电价预测模型。首先,通过相关性分析筛选出对日前电价预测影响较大的因素;然后,利用卷积神经网络初步提取电价数据和各个因素之间的局部依赖关系;进一步,运用循环神经网络和循环跳跃神经网络挖掘出当前数据与前后时刻数据之间的联系,再通过注意力机制进行权重自适应分配后,仿真非线性部分的预测值。采用自回归模型对线性部分的电价数据进行提取。最后,将线性和非线性部分的预测值进行融合,得到最终预测结果。经仿真验证,所提模型有效提高了日前电价预测的精度。In order to improve the prediction accuracy of day-ahead electricity price,a prediction model of day-ahead electricity price based on long-term and short-term time-series networks is proposed.Firstly,the factors which have great influence on the prediction of electricity price are selected by correlation analysis.Then,the convolutional neural network is used to extract the local dependence between the electricity price data and various factors.Furthermore,cyclic neural network and cyclic jump neural network are used to dig out the relationship between the current data and the data before and after the time,and the predicted value of the nonlinear part is simulated after the weight adaptive allocation through the attention mechanism.An autoregressive model is used to extract the electricity price data of the linear part.Finally,the predicted values of the linear and nonlinear parts are fused to obtain the final prediction results.Simulation results show that the proposed model can effectively improve the prediction accuracy of day-ahead electricity price.

关 键 词：注意力机制 电价预测 卷积神经网络 长期和短期时间序列网络 自回归模型 

分 类 号：TM73[电气工程—电力系统及自动化] TM-9