一种改进的遗传优化策略在电机故障诊断中的应用

One Kind of Improvement Heredity Optimization Strategy in Motor Failure Diagnosis Application

依据神经网络和遗传算法的特点,本文提出了利用遗传算法(GA)优化神经网络,将二者有机的结合起来,建立故障诊断的优化模型(GA—BP)网络,在遗传算法中放弃传统的轮盘赌,采用一种叫锦标赛的选择策略并改变迁移策略来进行遗传算法,优化BP网络的初始权值和阈值。以各种原始资料和现场实录资料作为训练样本,首先进行遗传算法的运行,得到了优化的权值和阈值,作为BP网络的的初始权值和阈值,接下来通过BP网络训练样本,实现BP网络学习的目的,建立样本(作为输入变量)与实际故障类型(作为目标变量)之间的潜在联系。最后用测试样本对GA—BP网络进行测试,检验表明用改变选择策略并改变迁移策略的遗传算法来优化BP网络的诊断正确率明显得高于未进行优化BP网络,不仅能发挥神经网络的泛化映射能力而且诊断速度也有提高,有较强的学习能力。

Based on neural network and genetic algorithm is proposed using genetic algorithms (GA) optimization of neural networks the two combine organic set up the optimization model for fault diagnosis (GA-BP) network,the genetic algorithm to abandon the traditional The roulette,using a strategy called tournament selection and change migration strategy for genetic algorithms,optimization of BP network's initial weights and thresholds.A variety of original data and on-site recording information as a training sample,the first genetic algorithm has been optimized weights and thresholds,as the BP network's initial weights and thresholds,followed by the adoption of BP network training samples achieve the purpose of studying BP network,set up samples (as input variables) and the actual fault type (as the target variables) between the potential contact.Finally use of the test samples of GA-BP network testing,inspection showed that the change in strategy selection and strategy to change the migration of genetic algorithm to optimize the BP network diagnostic accuracy was significantly higher than non-optimized BP network can not only play a pan-neural network mapping of the ability and speed of diagnosis has improved,has a strong ability to learn.