With the rapid development of smart grid, the problem of nonintrusive load monitoring (NILM) and disaggregation has been widely concerned in order to effectively improve the utilization of power, rationally plan the power resources, and establish the nationwide smart power and load monitoring system. To improve the performance of nonintrusive load disaggregation, a nonintrusive load disaggregation method based on coupled neural network is presented. Firstly, the dataset is normalized and preprocessed. Secondly, a hybrid deep learning model is constructed, which combines convolution neural network (CNN) with bidirectional gated recurrent unit (BiGRU). The spatial and temporal characteristics of data are fully explored, and attention mechanism is added to focus on important information and eliminate redundant features. Finally, the domestic selftest data set is used for the experiment, and the coupled neural network is evaluated with different evaluation indexes, and compared with other commonly used disaggregation models. The experimental results show that the mean absolute error and sum of absolute error of the proposed method are reduced compared with other disaggregation methods, the mean absolute error is reduced by 35.9% and the sum of absolute error is reduced by 39.9% on average.