基于自注意力的电力数据能耗预测算法Dual-channel LSTM
A Self-Attention-based power data energy consumption prediction algorithm Dual-channel LSTM
杨文清 1陈怀新 1王召 1朱佳1
作者信息
- 1. 国网电力科学研究院,南京 210044
- 折叠
摘要
电力能源需求随不可预测事件波动性加剧,为实现准确预测建筑用电能耗,提出一种电力数据能耗预测算法Dual-channel LSTM.算法框架主要分为3个模块,首先通过T-LSTM模块对时序存在缺失值的电力数据预处理,使其时序排列规则,随后输入双通道注意力特征模块,该模块引入自注意力机制来引导权重分配,最后由LSTM和FCN结合的预测模块输出结果.在AEP和IHEPC公开电力数据集上评估了该算法,Dual-channel的损失函数MAE/MSE分别仅有0.12/0.06和0.06/0.04,通过消融实验发现,引入T-LSTM对算法性能起决定性影响,引入自注意力机制对算法起辅助性作用,所提方法提高了预测精度和鲁棒性.
Abstract
Electricity energy demand increases volatility with unpredictable events.In order to achieve ac-curate prediction of building energy consumption,this paper proposes a Dual-channel LSTM for power data energy consumption prediction algorithm.The algorithm framework is mainly divided into three modules.Firstly,the power data with missing values in the time series is preprocessed by the T-LSTM module,so that the time series can be arranged regularly.The Dual-channel attention feature module is then input,which introduces a self-attention mechanism to guide the weight assignment.Finally,the prediction module combined with LSTM and FCN outputs the result.The algorithm is evaluated on the AEP and IHEPC public power datasets.The loss function MAE/MSE of Dual-channel is only 0.12/0.06 and 0.06/0.04 respec-tively.Through ablation experiments,it is found that the introduction of T-LSTM has a decisive impact on the performance of the algorithm,and the introduction of self-attention mechanism plays an auxiliary role in the algorithm.The proposed method improves the prediction accuracy and robustness.
关键词
电力能源/能耗预测/注意力/LSTM/时序数据Key words
power supply/energy consumption prediction/attention/LSTM/time series data引用本文复制引用
基金项目
国家电网电力科学研究院科技项目(52460821-0187)
出版年
2024
NETL
NSTL
万方数据