基于时间模糊化长短时记忆的非侵入式负荷分解方法

Non-intrusive Load Decomposition Method Based on Time-fuzzified Long Shortterm Memory

基于深度学习的负荷分解方法忽略了设备状态的关联性,导致应用过程中会出现功率误判现象。针对上述问题,提出一种基于时间模糊化长短时记忆(TFLSTM)的非侵入式负荷分解方法。首先,根据合适的时间区域分割数据集,利用长短时记忆(LSTM)建立时刻关联性。然后,通过编码和解码去除非目标设备信息,并根据用户在不同时间区域的用电习惯来确定模糊策略,最终完成负荷分解。基于公开数据集,将文中方法与经典方法及前沿方法进行对比,实验结果表明TFLSTM能够有效降低功率误差,功率分解准确率可提高4%~15%。

The deep learning based load decomposition methods ignore the relevance of equipment states,which will lead to the phenomenon of power misjudgment in the application process.In view of the above problems,a non-intrusive load decomposition method based on time-fuzzified long short-term memory(TFLSTM)is proposed.First,the dataset is divided according to the appropriate time zone,and the long short-term memory(LSTM)is used to establish the correlation of time.Then,the information of non-target equipment is removed through encoding and decoding,and the fuzzy strategy is determined according to user's power consumption habits in different time zones,and finally the load decomposition is completed.Based on the public datasets,the method in the paper is compared with the classical methods and the state-of-art method.The experimental results show that TFLSTM can effectively reduce the power error and improve the accuracy of power decomposition by 4%~15%.