改进PSO-BP网络在故障诊断中的应用

Application of BP Network Based on Improved PSO in Fault Diagnosis

针对油田抽油机井故障诊断方法较落后的问题,提出一种基于改进PSO-BP网络的故障诊断系统。神经网络权值的训练采用改进的PSO算法,克服了BP学习算法收敛速度慢、易陷入局部极值的缺点。将该网络用于抽油机井的故障诊断,并与传统BP模型的故障诊断结果进行比较。结果表明:基于改进PSO-BP的故障诊断方法正确率达96%以上,可以在更短的时间内、用更少的迭代次数达到精度要求,为设备检修提供了可靠的依据。

A BP neural network fault diagnosis system was proposed based on a kind of improved particle swarm optimization （PSO） algorithm by aiming at the problem of pump-jack fault diagnosis method backward. The improved PSO algorithm was used to train the BP neural network weight values. The shortcomings of slow convergence and trapped easily in the local extreme values by BP learning algorithm could be overcome. The system was used to diagnose the faults of pump-jacks in oil field and compared with traditional BP model. The results show the correctness of improved fault diagnosis method based on PSO-BP is arrived at above 96%, and accuracy requirement can be met during shorter time and using fewer repeating degree. It provides reliable basis for examining and repairing of equipment.