Research on residential non intrusive load identification algorithm based on equipment operation state detection and energy regression synchronous evaluation
The non-intrusive load monitoring technology can efficiently and low-cost obtain the sub item power of users and support a variety of services. The neural network based on the energy regression of sub item electrical appliances provides an important support for the load identification technology. In this paper, aiming at the noise identification pollution at the equipment shutdown during the energy regression of the neural network and the limitations of the evaluation of equipment operation status based on the energy threshold method, a hard parameter sharing multi task learning model based on the energy regression and status classification is proposed. According to the sensitivity difference between energy regression and status classification to the global and regional information of the input sequence, a time convolution network based on multiscale receptive field is proposed. The experimental results show that the proposed DNN model has improved the disaggregation performance, and reduced the MAE by 50% compared with the traditional network on small power fluctuation devices such as washing machines and dishwashers.