Study on a multi-feature and multi-model fusion methodology for gas pressure regulator station fault diagnosis

Aiming to address the problems of low accuracy, such as false alarms and missed faults during the operation of medium- and low-pressure regulator stations, this paper proposes a multi-feature and multi-fusion fault diagnosis method for use in pressure regulator stations in gas pipeline systems. The methodology first decomposes non-linear outlet pressure signals using complementary ensemble empirical mode decomposition (CEEMD) and then fuses them with auxiliary parameters. Then, kernel principal component analysis (KPCA) is used to downscale the feature matrices. Subsequently, a genetic algorithm-support vector machine (GA-SVM) and crested porcupine optimiser-back-propagation (CPO-BP) perform preliminary fault classification. Finally, the initial diagnostic results are fused by Dempster-Shafer (D-S) evidence theory to obtain the optimised final diagnostic results. The experimental results show that the diagnostic accuracy of the CPO-BP can reach 99%, with further performance improvements observed after D-S fusion.