Understanding the characteristics of user load distribution is an important part of smart grid construction. Non-Intrusive Load Monitoring (NILM) is widely recognized by power system workers for its advantages of convenience, efficiency and low cost. This paper proposes a NILM method based on long-term and short-term memory networks. By collecting the current waveform at the user''s power inlet and performing data processing, the user''s load characteristic data is obtained. Use principal component analysis to reduce the number of load features and improve operational efficiency. Use the long-short-term memory network model that is good at processing continuous data, the model is evaluated on the divided verification set and test set, in order to obtain the optimal parameter model. The prediction experiment results show that the non-intrusive load monitoring method designed in this paper can accurately identify household appliances including low-power appliances.
