基于seq2seq和Attention机制的居民用户非侵入式负荷分解

Nonintrusive Load Monitoring based on Sequence-to-sequence Model With Attention Mechanism

非侵入式负荷分解(nonintrusiveloadmonitoring,NILM)是大数据分析在智能配电系统中为终端用户提供的重要应用之一,能够提升对负荷的认知水平,显著提升需求侧响应的潜力。长期以来,传统的NILM算法存在误判率高,功率分解值准确度低等问题。为此,采用深度学习的框架,提出一种基于序列到序列和Attention机制的NILM模型。该模型首先将输入的有功功率时间序列通过词嵌入映射到高维向量,并利用基于长短时记忆模型的编码器进行信息提取;然后通过引入Attention机制的解码器,从提取的信息中选取与当前时刻相关度最高的信息,用于解码并最终得到负荷分解结果。提出的深度学习网络模型能够显著提升对信息的提取与利用能力。基于REFITPowerData数据集的测试结果验证了方法的有效性。

Nonintrusive load monitoring(NILM) is one of the key applications of big data analytics in smart power distribution systems for end-use customers. A successful implementation of nonintrusive load monitoring can improve the knowledge of load, and has great potential in increasing demand side response. Traditional nonintrusive load monitoring algorithms have long suffered from the problems of high misjudgment rate and low accuracy of disaggregated power value. To address these problems, the deep learning framework was adopted. Specifically, a nonintrusive load monitoring model based on sequence-to-sequence(seq2seq) model with attention mechanism was proposed. The model first embeds the input active power time sequence into a high dimensional vector, extracts information with a long short term memory(LSTM)-based encoder, and then selects the most relevant information to decode and reaches the final disaggregation results with a decoder wrapped by attention mechanism. Compared with existing models, the proposed deep learning network structure increases model’s ability to extract and utilize information dramatically. The proposed model was tested on the REFITPowerData dataset, and compared with the state-of-the-art model.