A fusion data preprocessing method and its application in complex industrial power consumption prediction

Abstract

Data-driven prediction methods often encounter problems in industrial applications due to noise, data redundancy, and insufficient labeled data. Preprocessing the data is required to improve prediction accuracy. At present, most data preprocessing methods are independent of each other and the correlation between methods is relatively small, resulting in repeated redundancy and reduced efficiency. To address this problem, this work proposes a fusion data preprocessing method based on a stacked denoising auto-encoders network (SDAE) and generative adversarial networks (GANs). This method reconstructs the noise reduction link in the SDAE network, improves the noise reduction effect of the model and extracts the main features of the data; the GANs is used to avoid overfitting caused by insufficient training data. Finally, the proposed method is applied to power consumption prediction in the metallurgical industry. Experiments show that the proposed method effectively reduces the model calculation time and improves the overall prediction accuracy.