基于小波变换的Bi-LSTM-TCN短期电价预测  

Wavelet transform-based Bi-LSTM-TCN method for short-term electricity price forcasting

作　　者：竺筱晶 薛睿萌 ZHU Xiaojing;XUE Ruimeng(School of Mathematics and Physics,Shanghai University of Electric Power,Shanghai 201306,China)

机构地区：[1]上海电力大学数理学院,上海201306

出　　处：《电工电能新技术》2023年第12期60-68,共9页Advanced Technology of Electrical Engineering and Energy

基　　金：国家自然科学基金项目(12271342、12172210)。

摘　　要：随着可再生能源大规模并入电网,电价预测变得越来越困难。为更准确预测含新能源电力市场中的电价,本文提出一种基于离散小波变换(DWT)的双向长短期记忆网络(Bi-LSTM)和时间卷积网络(TCN)的短期电价预测模型。首先利用DWT提取数据的时频图,并对重构后的子序列进行相关性分析;然后对于受不同因素影响的子序列建立不同的模型分别进行预测,最后叠加预测结果得到最终预测值。并在北欧丹麦DK1电力市场数据集上进行实验,该方法的均方根误差(RMSE)和平均绝对误差(MAE)分别为3.081%和2.588%,与一些基准模型和现有预测模型相比,该方法的预测精度更高。Electricity price forecasting is becoming increasingly difficult as renewable energy is integrated into the grid on a large scale.In order to more accurately predict the electricity price in the new energy power market,a short-term electricity price forecasting model based on discrete wavelet transform(DWT),bi-directional long shortterm memory network(Bi-LSTM)and temporal convolution network(TCN)was proposed.Firstly,the time-frequency diagram of the data was extracted by DWT,and the correlation analysis was performed on the reconstructed subsequences.Then,different models were established to predict the subsequences affected by different factors.Finally,the predicted results were superimposed to obtain the final predicted value.Experiments are performed on the dataset of the Nordic Danish DK1 power market,the root mean square error(RMSE)and mean absolute error(MAE)of this method are 3.081%and 2.588%respectively.Compared with some benchmark models and existing prediction models,the method has higher prediction accuracy.

关 键 词：短期电价预测 风力发电量 离散小波变换 双向长短期记忆网络 时间卷积网络 

分 类 号：TM715[电气工程—电力系统及自动化]