非侵入式负荷分解对于节能减排、负荷调峰、智能用能等方面均具有重要的现实意义。针对目前非侵入式负荷分解方法在低频采样条件下（1 Hz及以下）分解准确率较低的问题，提出了一种基于卷积神经网络(convolution netural network, CNN)与长短期记忆网络(long short-term memory, LSTM)相结合的seq2seq的非侵入式负荷分解算法(seq2seq based on CNN and LSTM，seq2seqBCL)。该深度学习模型将功率时间序列作为网络的输入，通过CNN做特征提取。考虑到电力数据的时序性，增加了LSTM层进行电器识别，相比于NILMTK中seq2seq模型降低了网络层数，简化了网络结构。在REDD数据集上对算法性能进行了评估，所提出的算法提升了整个网络系统的性能，与FHMM、CO和传统seq2seq算法相比，负荷分解准确率有明显提升。
英文摘要:
Non-intrusive load decomposition has important practical significance for energy conservation and emission reduction, load peak shaving, and intelligent energy use. Aiming at the problem of low decomposition accuracy of current non-intrusive load decomposition methods under low frequency sampling conditions (1 Hz and below), a seq2seq non-intrusive load decomposition algorithm (seq2seq Based on CNN and LSTM, seq2seqBCL) based on the combination of convolutional neural network (CNN) and long short-term memory network (LSTM) is proposed in this paper. This deep learning model uses power time series as input to the network, and uses CNN for feature extraction. Considering the time sequence of power data, the LSTM layer is added to identify electrical appliances. Compared with the seq2seq model in NILMTK, the number of network layers is reduced and the network structure is simplified. The algorithm performance is evaluated on the REDD data set. The proposed algorithm improves the performance of the entire network system. Compared with FHMM, CO and traditional seq2seq algorithms, the accuracy of load decomposition is significantly improved.
