德政 张 / University of Science and Technology Beijing
永红 谢 / University of Science and Technology Beijing
It is not common for power users arrearage to be found in time and accurately in China.This phenomenon leads to shortage of arrearage data and process of predicting arrearage can not be further developed. Stratified sampling has become a general method for solving this problem.In this paper we explore a new methods of predicting arrearage that uses the WGAN-GP(Improved Training of Wasserstein GANs) to simulate the default data and generate data using to predict arrearage with DBN. Power data (256 indicators), like the image pixel (16*16), is used as input to the wgan-gp to train the generator. For the first time, this paper proposes to change the prediction of arrears to predict the user's arrears interval, and designs a series of relevant indicators to help the experiment.The experiment proves that the prediction accuracy can be improved with analogue data generated by WGAN-GP.