Study of the Fault Diagnosis Method of Control Systems Based on MCCSAPSO-SVM

10.3966/199115992017102805004

Ruicheng Guo；Pu Yang；Xu Pan；Jianwei Liu

fault diagnosis ； MCCSAPSO ； MKF ； partial binary tree SVM

電腦學刊

28卷5期（2017 / 10 / 01）

39 - 50

英文

In order to improve the accuracy of actuator fault diagnosis of control system, a new method based on Multi-swarm cooperative chaos simulated annealing particle swarm optimization-support vector machine (MCCSAPSO-SVM) is proposed in this paper. Firstly, the noise reduction and feature extraction for the output signal are taken by the joint noise reduction and improved empirical mode decomposition (EMD) method. Secondly, structure parameters of SVM are optimized by MCCSAPSO, which not only can effectively avoid the premature convergence of particle swarm, but also can overcome the misjudgment problem caused by single particle information exchange. This algorithm can accelerate the convergence velocity and improve the accuracy of traditional PSO. Thirdly, the use of the mixture kernel function (MKF) can guarantee the good generalization and learning ability of SVM. Finally, a partial binary tree SVM is constructed by using the training data. This structure transforms a complex multi-classification problem into a number of two classification problems, which reduces the computational complexity and improves the real-time performance of the diagnosis. The experimental results of quad-rotor semi-physical simulation platform verify the feasibility and effectiveness of the proposed method.