基于改进粒子群优化BP网络的发动机故障诊断方法

Improved BP-neural Network of the Particle Swarm Optimization in the Research on Engine Fault Diagnosis

使用BP神经网络进行故障诊断过程中,随着输入变量的增加会造成"维数"灾难,导致训练效率不高,而且易陷入局部极小的问题。基于粗糙集的约简是常用的降低"维数"的方法,但约简是NP问题,随着信息量增多计算量会随之剧增;本文采用基于属性重要度的启发式值约简算法进行属性约简,建立了一种模糊信息知识发现方法结合粒子群优化BP网络的故障诊断方法。通过实验表明此方法不仅能有效获取规则,降低网络的输入维数,还能有效避免陷入局部极小,从而提高故障诊断的效率。

In the process of using BP-neural network in fault diagnosis,there will be ＂dimension tragedy＂ as the input variable increases,which causes the lower training effective.Besides,traditional BP algorithm tends to fall in local optimization.The reduction based on the rough set（RS） is the conventional ＂reduce dimension＂ method,but it is NP-hard problem,whose computing will gradually augment as the information increases.Therefore,a heuristic algorithm was used for attribute reduction based on the importance of attribute value to reduce attribute,a fault diagnosis approach was formed combining the fuzzy information system knowledge method with BP-neural network of the particle swarm optimization（PSO） algorithm to diagnose the fault of engine.The experiments show that comparing with the conventional method,it can not only require fault diagnosis rule,but also reduce net input dimensions effectively,avoid falling in local optimization and increase the efficiency of fault diagnosis.