Leak location of pipelines based on transient model and PSO-SVM

Abstract

An improved and integrated approach of support vector machine and particle swarm optimization theory (PSO-SVM) is first used to detect the leak location of pipelines and overcome the problem of multiple leaks. The calibration and predictive ability of improved PSO-SVM is investigated and compared with that of other common method, back-propagation neural network (BPNN). Two conditions are evaluated. One with a leak involves a set of 20 samples, while another with two leaks has 127 samples. Both internal and external validations are performed to validate the performance of the resulting models. The results show that, for the two conditions, the values calculated by improved PSO-SVM are in good agreement with those simulated by transient model, and the performances of improved PSO-SVM models are superior to those of BPNN. This paper provides a new and effective method to inspect the multiple leak locations, and also reveals that improved PSO-SVM can be used as a powerful tool for studying the leak of pipeline.