基于遗传优化的RBF-BP神经网络电液伺服阀故障诊断算法研究

Fault Diagnosis Algorithm Research of Hydraulic Servo Valve Based on Genetic Optimization of RBF-BP Neural Network

电液伺服阀是液压伺服系统的核心元件,因此是故障诊断的重点对象,其故障原因经常呈现出非线性和不确定性等复杂状态.单一的BP网络是全局逼近神经网络,学习速度很慢,容易陷入局部极小,易产生震荡等不足,RBF网络是局部逼近神经网络,训练速度快,在训练时不会发生震荡,也不会陷入局部极小.基于它们各自的优缺点,通过将RBF神经网络和BP神经网络有效地结合在一起,取长补短,建立一个由RBF子网和一个BP子网两部分串联构成的双隐藏层RBFBP组合神经网络.该网络既具有BP网络较好的泛化性能,又具备RBF网络较快的逼近速度.用遗传算法优化该神经网络的初始权值和阈值.该网络同时具有RBF网络和BP网络的优点,适用于复杂非线性系统的故障检测.

The servo valve as a core component of the hydraulic servo system is the focus of fault diagnosis,and the cause of the malfunction often exhibits complex state of uncertainty and nonlinear. Single BP network is a global approximation of neural networks with some shortcomings,such as learning speed is very slow,easy to fall into local minima,and easy to produce shock. RBF network is a local approximation neural network with training speed,and it will not produce shock or trap in local minima during training. Based on their respective advantages and disadvantages,by using RBF network and BP network effectively together,this paper complements each other,and builds double combinations of hidden layers RBFBP neural network consisting of two parts series. The network not only has better generalization performance of BP network,but also had a faster approach speed of RBF network. Genetic algorithm is used to optimize the initial weights and thresholds of the neural network. The network also has advantages of RBF network and BP network for complex nonlinear system fault detection.