基于注意力机制的CNN-LSTM短期负荷预测
Short-term Load Prediction by CNN-LSTM Based on Attention Mechanism
王晓兰 1张惟东 1王惠中1
作者信息
摘要
精准的电力负荷预测是电力系统运行的关键前提.论文为提高短期电力负荷的预测精度,采用一种新的预测方法.通过Spearman相关系数对四季影响因素与负荷序列进行相关性分析;结合CNN易处理高维数据且可以更好挖掘到负荷序列的隐含特性,注意力机制可以对输入的影响因素进一步分配权重的优势,以相关性分析出的四季主要因素为基础进行仿真实验.实验结果表明,特征选择输入的CNN-LSTM-Attention组合网络对比全特征输入的CNN-LSTM-Attention网络、全特征输入的CNN-LSTM网络,在不同季节上的日负荷预测精度均有进一步提高.
Abstract
Accurate power load forecasting is a key prerequisite for power system operation.In this paper,a new forecasting method is proposed to improve the accuracy of short-term power load forecasting.The correlation analysis between the influencing factors of four seasons and the load sequence is carried out by Spearman correlation coefficient.Combining the advantages that CNN is easy to handle high-dimensional data and can better tap into the implicit characteristics of the load sequence,and the attention mechanism can further assign weights to the input influencing factors,simulation experiments are carried out based on the main fac-tors of four seasons from the correlation analysis.The experimental results show that the combined CNN-LSTM-Attention network with feature-selective input has further improved the daily load prediction accuracy in different seasons compared with the CNN-LSTM-Attention network with full feature input and the CNN-LSTM network with full feature input.
关键词
负荷预测/相关性分析/CNN/LSTM/注意力机制Key words
load prediction/correlation analysis/CNN/LSTM/attention mechanism引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据