A Novel Deep Learning Representation for Industrial Control System Data

Feature extraction plays an important role in constructing artificial intelligence (AI) models of industrial control systems (ICSs). Three challenges this field are learning effective representation from high-dimensional features, data heterogeneity, and noise due to the diversity dimensions, formats sensors, controllers actuators. Hence, a novel unsupervised autoencoder model is proposed for ICS paper. Although traditional methods only capture linear correlations our deep (DIRL) based on convolutional neural network can mine high-order thus solving problem heterogeneous data. In addition, denoising introduced noisy DIRL. Training allows better mitigate sensor problem. way, representative features learned by DIRL could help evaluate safety state ICSs more effectively. We tested with absolute relative accuracy experiments two large-scale datasets. Compared other popular methods, showed advantages four common indicators AI algorithms: accuracy, precision, recall, F1-score. This study contributes analysis data, which promotes stable operation ICSs.