Research on fault diagnosis based on RBF NN optimized by an improved QPSO algorithm

For QPSO (Quantum-behaved Particle Swarm Optimization) algorithm´s disadvantages of premature convergence and easily getting into local extremum, an improved QPSO algorithm called co-evolutionary QPSO algorithm with two populations is presented in this paper. Particles are updated by adopting QPSO algorithm inside populations and by using annexing or cooperation operator between populations. The annexing strategy makes the population with worse performance accept the other population´s optimal information with certain probability; And the cooperation strategy makes the two populations exchange optimal information with each other. Moreover, one population introduces Cauchy mutation when the two populations trap into the same optimal value. Then RBF NN (Radial Basis Function Neural Network) is trained by the improved QPSO algorithm and it is applied to fault diagnose of diesel engine valve. The simulation results showed that the improved QPSO-RBF algorithm enhanced accuracy and speeded up convergence rate of fault diagnosis.