粒子群优化的神经网络在齿轮故障诊断中的应用

Application to Neural Network of particle swarm optimization in gear fault diagnosis

针对BP神经网络容易陷入局部极小及收敛速度慢的问题,本文利用粒子群优化算法代替BP算法中的梯度下降法训练神经网络的权重和阈值,有效地改善了BP网络诊断性能;利用训练后的神经网络对齿轮进行了故障诊断,并比较了基于粒子群优化算法与BP算法的诊断结果,通过仿真实验表明:无论是在诊断速度上还是在诊断精度上,PSO-BP神经网络诊断性能都比单独的运用神经网络有很大提高。

According to the problem that back propagation（BP） neural network algorithm might easily fall into local minimum and converge slowly,this paper used particle swarm optimization（PSO） algorithm to instead of gradient descent method and train the weights and thresholds of BP network,improved the diagnostic performance of BP neural network Effectively.The neural network trained by PSO was applied to gear fault diagnosis.The diagnostic results between PSO and BP algorithm were compared.Simulating experiments showed that the diagnosis performance of PSO-BP network was better than that of using individual BP network in both aspects of speed and accuracy of gear fault diagnosis.