基于WPD和双重注意力机制TCN的短期电价预测

Short-term electricity price forecasting based on wavelet packet decomposition and TCN with double attention mechanism

针对电价的高频、非平稳性且受多种因素影响、时间卷积网络(TCN)在实际应用中忽略各输入特征的关联性以及在处理历史信息上表现较差的问题,本文提出了一种基于小波包分解和双重注意力机制TCN的短期电价预测方法。首先利用小波包分解对电价序列进行分解重构,去除高频部分并进行重构;然后使用引入双重注意力机制的TCN模型进行电价预测。为挖掘电价与其影响因素的关联性,引入特征注意力机制实时计算各影响因素特征量的权重,突出关键输入特征;同时,为挖掘当前时刻电价与历史时刻信息的关联性,引入时序注意力机制提取历史关键时刻点信息,提高关键时刻点预测的精确度;最后以澳大利亚新南威尔士州电力市场实时数据为例进行预测分析,对比其他几种电价预测方法,验证了本文所提方法的有效性。

To address the problems that electricity prices are high-frequency,non-stationary and influenced by many factors,and that temporal convolutional networks(TCN)ignore the correlation of each input feature and perform poorly in handling historical information in practical applications,this paper proposes a short-term electricity price prediction method based on wavelet packet decomposition and dual-attention mechanism TCNs.Firstly,the wavelet packet decomposition is used to decompose and reconstruct the electricity price series to remove the high-frequency parts and reconstruct them;then the TCN model with the introduction of the dual-attention mechanism is used for electricity price prediction.To explore the correlation between electricity price and its influencing factors,the feature attention mechanism is introduced to calculate the weights of each influencing factor feature quantity in real time to highlight the key input features;meanwhile,to explore the correlation between electricity price at current moment and historical moment information,the temporal attention mechanism is introduced to extract the historical key moment point information to improve the accuracy of key moment point pre-diction.The prediction analysis is based on the real-time data of the New South Wales electricity market in Australia,and the effectiveness of the proposed method is verified by comparing several other electricity price prediction methods.