Abstract: Aiming at the problems of poor classification accuracy of electrical appliances caused by the limited scope of application and complex model of traditional power data analysis methods,a power data analysis model based on multi-layer stacked Long Short-Term Memory (LSTM) network is proposed.Firstly,the features of power data are extracted from the spectrum diagram,Mel Frequency Cepstrum Coefficient (MFCC) and Mel spectrum diagram of power data,and then applied to the deep learning model to improve the performance of classification tasks,so as to improve the over fitting problem.Secondly,a multi-layer stacked LSTM model is established to effectively improve the classification and regression ability of the model.Finally,an improved soft coding and multi-scale training method is proposed to prevent the peak probability distribution and improve the generalization ability of the model.In the experimental stage,the proposed model is verified by taking the household power data set as an example.The simulation results show that the soft coding and multi-scale training of the proposed model have a certain effect on accelerating the training effect,and the final classification accuracy reaches 89.85 %.

Key words: Power system, Data analysis, Deep learning, Code, Feature extraction