Application Of Improved Particle Swarm Optimization Algorithm In Neural Network

Purpose—In the current research,particle swarm optimization algorithm has some shortcomings,because it is easy to fall into the local optimal solution,resulting in low convergence accuracy and difficult convergence.Particularly,the application of PSO to other algorithms,especially in the training of neural networks,having a very unsatisfactory effect.It is particularly important to explore the sufficient and necessary conditions for the global convergence of PSO.Design/methodology/approach—Taking the variance of group adaptive degree and the minimum number of stagnation as the judgment conditions of particle swarm optimization convergence,APSO optimization algorithm is proposed.The results are verified by comparing five standard test functions with standard particle swarm optimization algorithm.APSO optimization algorithm is applied to the training of three-layer neural network.Compared with traditional BP algorithm and standard particle swarm optimization neural network,the APSO algorithm is better effective.Findings—Firstly,the convergence region of standard particle swarm optimization algorithm is deduced from mathematical theory.Due to the limitations of traditional particle swarm optimization,it is easy to converge locally,resulting in low convergence accuracy and difficulty in convergence.Therefore,an APSO optimization algorithm is proposed.The convergence conditions of PSO are the variance of group adaptation degree and the minimum number of stagnation.Experiments verify the efficiency of APSO algorithm.Finally,APSO optimization algorithm is combined with neural network.Compared with traditional back propagation algorithm and standard particle swarm optimization neural network algorithm,the effectiveness of APSO optimization algorithm is verified by UCI data set.Practical implications—Through the research and extension of the particle swarm optimization algorithm,the performance of global convergence is improved,and its training in neural network has achieved good results.Originality/value—The APSO optimization algorithm is proposed,and the APSO optimization algorithm is applied to the training of neural networks.