An Attack Prediction and Recognition Method for Water Treatment System Based on One-dimensional Convolutional Neural Network and Cumulative Sum

An attack prediction and recognition method for water treatment system based on one-dimensional convolutional neural network (1D-CNN) and cumulative sum (CUSUM) is proposed in this paper. Firstly, a 1D-CNN is employed as a time series predictor to establish an independent prediction model for each attack point, followed by training and learning. Subsequently, the deviation between predicted values and actual values is compared with a threshold value using CUSUM statistics to identify anomalies within the water treatment system. Experimental evaluations conducted on the Safe Water Treatment (SWaT) Test Bench demonstrate that 1D-CNN exhibits superior predictive performance compared to other models. Furthermore, CUSUM successfully detects and identifies all listed attacks based on statistical bias.