电力系统及其自动化学报2025，Vol.37Issue(9)：71-81,11.DOI:10.19635/j.cnki.csu-epsa.001603
基于改进双向时序卷积网络的非侵入式负荷分解模型
Non-intrusive Load Disaggregation Model Based on Improved Bidirectional Temporal Convolutional Network
摘要
Abstract
Aimed at the problems in the existing non-intrusive load technology such as the insufficient disaggregation ac-curacy for low-power and multi-state appliances and the low generalization performance of models,a load disaggrega-tion model integrating a multi-scale channel-enhanced attention mechanism and an improved bidirectional temporal con-volutional network is proposed in this paper.This model combines various convolutional and residual networks,and it overcomes the limitations of traditional convolutional neural networks including difficulty in capturing the global infor-mation and handling the time-series data,as well as gradient explosion when the network depth increases.The bidirec-tional structure enables current state inference from the historical data while utilizing transient fluctuations in the future to rectify the current state,thereby mitigating misjudgments caused by state transition delays or transient noise.Simulta-neously,the multi-scale channel-enhanced attention mechanism adaptively extracts temporal features across varying granularities through parallel multi-scale pooling operations,which is also combined with dynamic channel interaction modules to reinforce weight allocation for critical features.Experimental results show that the proposed model achieves low disaggregation errors for low-power and multi-state appliances in the Reference Energy Disaggregation Data(REDD)dataset,as well as a strong generalization capability.关键词
非侵入式负荷分解/双向时序卷积/残差网络/注意力机制/多尺度池化/深度学习Key words
non-intrusive load disaggregation(NILD)/bi-directional temporal convolution/residual network/attention mechanism/multi-scale pooling/deep learning分类
信息技术与安全科学引用本文复制引用
张彼德,钟子怡,陈豪,马俊梅,李天倩..基于改进双向时序卷积网络的非侵入式负荷分解模型[J].电力系统及其自动化学报,2025,37(9):71-81,11.基金项目
四川省科技计划资助项目(2023YFG0191). （2023YFG0191）
