A novel DVAPSO-LSTSVM classifier in compressed sensing domain for intelligent pipeline leakage diagnosis

Abstract

Leakage diagnosis is of great significance for maintaining the normal operation of pipelines. However, there is a large amount of redundant data in leakage data collection, resulting in increased detection time. In addition, the leakage detection model lacks the necessary optimization, which limits diagnostic accuracy. To overcome these two drawbacks, this paper proposes a novel leakage diagnosis model in compressed sensing (CS) domain. Firstly, the leakage signal is converted into the CS domain to reduce redundant data, and extract features in the CS domain to form a feature dataset. Then, a particle swarm optimization algorithm with decreased and variable amplitude strategy (DVAPSO) is proposed, which enhances the ability of particle optimization and jumping out of local optimum. The diagnosis model LSTSVM is optimized using DVAPSO algorithm to decrease the risk of trapping into local optimum in the training process. Finally, the leakage feature dataset of CS domain is sent to the DVAPSO-LSTSVM diagnosis model to complete the leakage detection. The experimental results show that pipeline data in CS domain improves detection velocity by 71.7% when the compression ratio is 50%. Moreover, the DVAPSO-LSTSVM has higher leakage identification accuracy compared to other diagnosis models, with a detection accuracy of 95.2%.