A Novel Event Detection Model for Water Distribution Systems Based on Data-Driven Estimation and Support Vector Machine Classification

Abstract

In this study, a novel event detection model based on data-driven estimation and support vector machine (SVM) classification was developed and assessed. The developed model takes advantage of the data-driven model - namely artificial neural networks (ANNs) - to predict the complicated behavior of water quality parameters without relevant physical and chemical knowledge. In addition, SVM presents high classification performance when dealing with high-dimensional data and has a better generalization ability than ANNs so that SVM can complement ANN predictions. Key parameters of SVM were optimized by genetic algorithm. After calculation of ANN prediction error and outlier classification by SVM, the event probability was estimated by Bayesian sequence analysis. The performance of the proposed model was evaluated using data from a real water distribution system with randomly simulated events. The results illustrated that the proposed model exhibited a great detection ability compared with two models with analogous structures, a pure SVM classification model and a conventional ANN-threshold classification model, demonstrating the superiority of the hybrid data-driven – SVM classification model.