Abstract: Aiming at the problem of poor performance of urban gas pipeline fault diagnosis, a pipeline fault diagnosis method based on improved particle swarm optimization optimized deep belief network (IPSO-DBN) is proposed. This method first modifies the inertia weight ω, acceleration factor C1 and C2 in the particle swarm optimization algorithm (PSO) to obtain an improved particle swarm optimization algorithm (IPSO), and uses two benchmark functions to compare and test the network performance of PSO and IPSO to prove the superiority of the selected improvement method. Secondly, use IPSO to optimize the initial weights of the deep belief network (DBN), establish a suitable DBN network, and use the experimental data under four different gas pipeline conditions for training and prediction of the IPSO-DBN network. Finally, the fault diagnosis accuracy obtained from the experiment is compared and analyzed with BP, DBN, PSO-DBN methods. Experimental results show that for the fault classification and identification of gas pipelines under different working conditions, the average test set diagnosis accuracy of the IPSO-DBN method is as high as 94.5%, and the diagnosis effect is better than the traditional BP, DBN and PSODBN methods. 

Key words: gas pipeline, fault diagnosis, particle swarm optimization, deep belief network