GA-BP算法在船舶主机故障诊断中的应用

GA-BP Algorithm Application in Fault Diagnosis of the Marine Main Engine

针对船舶主机故障具有诊断对象多、多因素耦合造成诊断准确率低等问题,提出了用遗传算法优化BP神经网络的故障诊断方法,利用GA算法对BP神经网络的初始权值、阈值在较大范围内搜索寻值,同时采用反向传播算法在较小范围内进行微调,优化网络结构和参数,加快目标最优值的求解,最后结合一般BP神经网络方法进行分析比较。实验结果表明,优化初始权值和阈值后的测试样本的误差由0.996 43减少到0.097 333,训练样本的误差由1.464 1减少到0.080 657;经GA优化后的BP神经网络模型对主机故障类型的诊断的准确率为100%,实现对船舶故障诊断的高效判别。

Aiming to solve the problems of multiple diagnosis objects,poor diagnosis accuracy because of multifactor coupling,this paper provides a new fault diagnosis method,namely,to optimize BP neural network by means of GA（ Genetic Algorithm）. Firstly,it searches the initial weights and thresholds in a more extensive range. At the same time,it adopts the BP algorithm to achieve the fine adjustment in an intensive range,the optimization of the network configuration and the optimal solution of the target. Finally,it analyzes and compares the results based on traditional BP neural network. The experiment results indicate that the error of test sample whose initial weights and thresholds have been optimized decreases from 0. 996 43 to 0. 097 333,the error of the training sample has been reduced from 1. 464 1 to 0. 080 657,and the BP neural network model optimized by GA helps to achieve 100% accuracy of the diagnosis on the main engine＇s fault,which eventually contribute to the efficient fault diagnosis on any ships.