Classification Method of Unbalanced Power Consumption Data Based on Prior Knowledge and Deep Boltzmann Machine Sampling

The existing customer labeling system in the smart grid construction process is not perfect. In the classification management of electricity data for massive users, there is a problem of small sample data and unbalanced distribution of labels. This paper proposes a classification method of unbalanced electricity data based on prior knowledge and deep Boltzmann machine(DBM) sampling. Firstly, the characteristics of load curve are extracted, the sampling principle is established, and the prior knowledge and DBM are used to sample the load curve. Then, the sample data are trained through the extreme learning machine(ELM) network. Finally, the Irish users’ electricity data are used as the data source. Contrastive experimental analysis results of original non-sampling, random oversampling and synthetic minority oversampling technique(SMOTE) show that the proposed method can better classify the unbalanced electricity data sets, realize the analysis of the user’s electricity usage behavior, and effectively support the peak shifting and peak avoidance at user side.