Reliability prediction-based improved dynamic weight particle swarm optimization and back propagation neural network in engineering systems

Abstract

Aiming at the problem of low accuracy of reliability prediction, a back propagation neural network (BPNN) model is developed. In the process of reliability prediction, a dynamic weight particle swarm optimization-based sine map (SDWPSO) method including a novel inertial weight update strategy is developed. This new strategy introduced a linear decreasing parameter in the sine-map, which enables particles to perform a fine search at a very low speed in the later stage of the search and greatly improves the convergence speed of the algorithm. Furthermore, a hybrid model named SDWPSO-BPNN is created to improve the reliability prediction accuracy in engineering systems. The proposed SDWPSO approach is compared with four algorithms using fourteen benchmark functions to verify the effectiveness. The experimental results indicate that SDWPSO has a better search ability than the other algorithms. Then, the hybrid SDWPSO-BPNN is applied to predict the reliability of turbocharger and industrial robot systems, respectively. The obtained results manifest that the SDWPSO-BPNN is more powerful than that of SVM and ANN methods for reliability prediction in engineering.