A Study on Sampling of Electric Vehicle Charging Data Based on K-Means-GMM-MLP Hybrid Model

After large-scale electric vehicles are connected to the power distribution network, the disorderly charging behavior of users with significant uncertainty seriously affects the power quality and stable operation of the local distribution grids.Due to the short running time and the lack of collection devices, it is difficult to obtain a large amount of historical data required for electric vehicles to access the distribution grids to optimize the dispatching operation, so the generation of large-scale electric vehicle charging data based on existing small samples has become an important research direction. In this paper, we propose a data sampling method based on the K-Means-GMM-MLP mixed model. Firstly, the historical charging data was transformed, and the starting charging time data was sampled by K-Means clustering and Gaussian Mixture Model (GMM), and then the potential relationship between the starting SOC, the end SOC, the total battery capacity and the starting charging time in the historical EV charging data was learned through the Multilayer Perceptron (MLP) to obtain the MLP model, and finally, the sampled starting charging time data was used to guide the trained MLP model to obtain the remaining EV charging information. This method can generate a large amount of data through small-sample single-dimensional data sampling and the learning and training of its potential relationship with multi-dimensional data. Experimental results show that the model has good convergence and can obtain a large number of sampling data that conform to the actual distribution law and retain the latent logical relationship between each dimension.