Volume 20: Sustainable Energy Solutions for a Post-COVID Recovery towards a Better Future: Part III

A GAN-CNN Based Model for Short-Term Load Forecasting Xiangya Bu, Qiuwei Wu, Jian Chen, Jinyong Dong

Abstract

This paper proposes a power system load forecasting method based on generative adversarial network and convolutional neural network (GAN-CNN), and applies it to the short-term load prediction. In this model, the generation layer and the discrimination layer form a maximum-minimum game and finally reach a Nash equilibrium. The data feature extraction method is integrated with the CNN convolution operation and variational mode decomposition (VMD) technology, which improves the quality of sample generation and reduces the prediction error. This paper provides a new and effective method for selecting similar days and forming the input matrix of the model. Finally, the real data were used to demonstrate the superior performance of the proposed method.

Keywords convolutional neural network, generative adversarial network, input matrix, load forecasting