Feature Subset Selection-based Fault Diagnoses for Automobile Engine

According to the stability of the membership degree and low identification rate in fuzzy inference system, this dissertation proposes the application of adaptive neural network-based fuzzy inference system to engine error diagnosis. To reduce the impact of excessive parameters on classification accuracy and cost, it also raises an asynchronous parallel particle swarm optimization method applied to the selection of feature subset. The method use uniform mutation operator, balancing effectively the particles´ ability to search globally and to develop. The asynchronous parallel particle swarm optimization algorithm(AP-PSO) that is used to select the feature subset is a potential feature subset that carries the characteristic of firstly initializing every particle as the selection question. Then, the method adopts improved asynchronous parallel particle swarm algorithm to conduct optimal searching based on the particle swarm that include several feature subsets and evaluate the classification ability (adaptive value) of the feature subset selected by way of Support Vector Machine. Finally, the optimal feature subset is got. Through verification of the build diagnosis model with data of engine tests, it has been found that the recognition accuracy attain to 98.72%, training error falling to 0.004423.The experiment indicates that the recognition rate of ANFIS system is significantly better than independent neural network reasoning system, fuzzy inference system.