DMM-WGAN: An Industrial Process Data Augmentation Approach

How to apply effective data augmentation methods to supplement datasets in harsh industrial environments is an important problem in complex industrial process modeling. In response to this problem, this paper proposes a new industrial process data augmentation method, DMM-WGAN, based on WGAN. Firstly, a Deep Threshold Mixing Feature Extraction Module (DMM) is proposed in the generator, which adopts a dual-channel fusion strategy. one channel extracts deep features of industrial data, while the other channel extracts global features of industrial data to enhance the feature extraction ability of the generator. Then, the DMM module and the Wasserstein Generative Adversarial Network are combined to establish the DMM-WGAN generation model. Finally, the proposed model is optimized and extensively experimented on a thermal power plant dataset, and the results are evaluated based on MSE, RMSE, MAE, and R2. The results show that the proposed DMM-WGAN generation model is superior to traditional VAE, GAN and WGAN generation models.