In view of the problems of insufficient accuracy and low computational efficiency of the current non-intrusive load monitoring（NILM）technology in processing complex electrical loads and long time series, a TCN-1DEMA-MTL model is proposed. The model utilizes the temporal convolutional network （TCN）module to extract the contextual features of the data and realize multi-level feature fusion; then, the one-dimensional exponential moving average（1DEMA） attention mechanism is introduced to dynamically weight and optimize the feature extraction effect; and the multi task learning（MTL） framework is adopted to integrate the load decomposition and state recognition tasks to further improve the prediction accuracy. In the experimental stage, the proposed model is compared with the existing bidirectional LSTM, S2P model and LDwA model, and all performance indicators show significant advantages. The experimental results verify the effectiveness and practicality of the scheme, which provides a powerful reference for the development of residential power load monitoring.
