For the research of non-intrusive power load detection technology, a load identification method based on deep learning LSTM network model is proposed. In the method, in order to avoid the interference of signals such as volt-age and current, a Gaussian window-based adaptive weighted-average variation point optimization algorithm is proposed to detect the load event, and the harmonic components are extracted as load feature tags, which are used as input to establish the mapping relationship between information in the LSTM model. And the offline training of the model and the online identification of the information are implemented in order to achieve a reliable identification of the operating status of the electrical equipment. Furthermore, this method solves the “gradient disappearance” phenomenon that occurs in the training process of the RNN network model when the sample data is large, and effectively improves the recognition accuracy. The experimental data shows that this method can accurately identify the status of the electrical equipment.