Global Parallel Technology Based On Radial Basis Function Neural Network

At present,deep learning technology is increasingly being replaced in various fields of various industries.On the contrary,the growth of deep learning technology and the rapid development of data volume,serialized neural network models,and increasingly difficult to meet current needs.Radial Basis Function Neural Network(RBFNN)is a three-layer feedforward neural network with simple structure and universal approximation.It is widely used in pattern recognition and prediction.However,in its actual application,selecting the appropriate RBFNN size is a time-consuming process.The process of manually selecting the network size requires a lot of experiments,and the results may lack versatility.In view of the above two problems,this paper combines RBFNN with particle swarm algorithm,and proposes,and on this basis,combines it with a health-based particle swarm optimization and optimization algorithm,and proposes a method based on RBFNN parallelization.Forecasting model.The work here is mainly as follows:Based on the in-depth study of the RBF neural network algorithm,this paper first combines the parameter settings of the RBF neural network with the particle swarm optimization algorithm,and adds a strategy for the nonlinear automatic adjustment of the inertia weight to the particle swarm optimization algorithm.An RBF neural network that can adaptively change its structure.Then,after researching a variety of parallelization strategies,a HHAPSO algorithm is proposed by improving and optimizing the PSO algorithm based on health.In this algorithm,the particle state is divided into two states according to the health degree,and different search strategies are used for particles of different health states,thereby enhancing the algorithm's search ability and enriching the particle distribution,which overcomes the previous difficulty.A local optimal solution occurs,the algorithmic and local search capabilities are insufficient.Here,a HHAPSO-RBFNN prediction model based on parallelization is proposed by combining the HHAPSO algorithm with an RBF neural network that can adaptively change the network structure size.Finally,in the multi-routine environment of the Spark platform,the HHAPSO-RBFNN prediction model proposed in this paper is tested by function approximation verification,Mackey-Glass time series prediction experiment and a traffic flow prediction experiment.The comparison of data validates the effectiveness and efficiency of this model.