Aiming at the current problems of high dimensionality,complex features,and difficulty in effective analysis of electric power data,a hybrid model of household electricity usage feature recognition based on time-frequency characteristics extraction is proposed.Firstly,to reduce the impact of the information redundancy and dimensional explosion or noise interference caused by high-dimensional data,time-frequency characteristics of users'household electricity usage are extracted based on wavelet transform and random forest(RF)algorithm.Secondly,the features are substituted into the classification prediction model for training,and the weights in the back propagation neural network(BPNN)are determined by dynamically changing the inertia factor in the particle swarm optimization(PSO)algorithm,to improve the performance of network training.The simulation stage is based on the data provided by an electric power company to classify and analyze the characteristics of electricity users'household electricity usage.The analysis results show that after the time-frequency characteristics extraction,the proposed model has an average accuracy of 0.796 8,which is 5.56%and 8.87%higher than the time-domain feature and frequency-domain feature extraction methods,respectively.In addition,the training performance of the proposed model is improved by 4.75 times and 2.58 times compared with the traditional PSO algorithm and the unoptimized model,respectively.The proposed model provides a reference for the characterization analysis of household electricity usage of power customers.
维普
万方数据