Based On MapReduce Parallel Framework In The Improvement Research And Application Of Neural Network

With the increasing application of neural networks and the increasing amount of data processing,the research on improving the speed of neural networks has attracted much attention of scholars at home and abroad.This thesis is to improve the problem of low convergence and large time cost in training samples and aims to strengthen the convergence of BP network,RBF network and CNN network,the research direction to improve network speed,from the two aspects of using MapReduce(MR)parallel programming model framework and improving network itself.At the same time,the improved neural network is applied to image recognition.The main contents of this paper are as follows:(1)Based on the adaptive learning method,an improved BP network is introduced by introducing the momentum influence factor that can be adaptively iterated with the network.At the same time,the MR framework is used to parallelize the BP network,and good acceleration results are obtained.The optimized BP network is applied to the actual problem of traffic accident prediction.Compared with the original BP network,the influence of the factor on the network convergence is analyzed.(2)In this paper,the RBF neural network is improved by using the firefly algorithm,in order to achieve the requirements of the implicit parameters of the network,which include the hidden cell center,the width and the weights between the hidden units and output units.Similarly,the network is parallelized in conjunction with the MR framework,which is then used to predict the application of motorway traffic.Analysis of the prediction results of the improved RBF network,and using the genetic algorithm(GA)and particle swarm algorithm(PSO)for the optimization of the RBF network are compared,analyzed whether the effect of optimization network that based on FA algorithm prediction is the most close to the real.(3)MeansK-algorithm and improved network activation function are used to improve CNN And place the improved network on the MR framework.The network is applied to the direction of image recognition based on three databases,and the improved effect is analyzed.Different common activation functions and improved activation functions are compared and compared two improved CNN network structures and an optimal improved CNN network is obtained.(4)In the application of traffic sign image recognition,it is further verified that the improved three neural networks combined with the MR framework have faster computing speed and higher recognition accuracy.