Research On Railway Freight Volume Forecast Based On Optimized RBF Neural Network

China is a resource-rich country,but the distribution of resources is uneven,with more resources in the northwest and relatively less in the southeast.As the most important means of resource transportation in China,railway transportation is the top priority of the transportation industry.With the increasing demand for railway freight,the construction and investment of freight railways are also increasing.The forecast of freight volume has become an important indicator for the country to grasp the development of the railway transportation industry.With the development of modern society,the traditional prediction methods have become slightly fatigued.The development of machine learning and intelligent algorithms gives more choices and possibilities for freight volume forecasting.After analyzing the shortcomings of the present methods of freight volume prediction,this paper proposes the use of Radial Basis Function(RBF)neural network to predict freight volume.The choice of RBF neural network parameters is directly related to the prediction accuracy of the model.In this paper,particle swarm optimization(PSO)is used to optimize the RBF parameters.Through the analysis of the convergence of the standard particle swarm algorithm,it is concluded that the algorithm is easy to fall into the local extremum in the later stage,and there is a "premature" problem,and the search accuracy needs to be improved.In response to these problems,this paper proposes a hybrid particle swarm H-PSO algorithm based on differentially evolved particle swarms and simplex particle swarms.The algorithm inherits the excellent global search ability of the former and the excellent local optimization ability of the latter.It also passes the function test.It is proved that the optimization ability of the algorithm is significantly improved compared with the standard PSO.Subsequently,an RBF neural network prediction model based on H-PSO optimization was established.After experimental verification,this method has significantly improved compared to pure RBF and RBF based on standard particle swarm optimization,and it is proved that the number of neurons is 3 When the model converges most efficiently.This paper uses the railway freight volume in the past 27 years as a training sample to predict the freight volume in the following four years,and compares the experimental results with the relatively mature gray model and the BP(Back Propagation)neural network model.By comparison,the RBF model based on optimized particle swarm(HPSORBF)proposed in this paper reduces the average error of 6.36% compared to the gray model,and reduces the average error of 3.48% compared to the BP neural network.Finally,a simple combination of the HPSORBF and the gray model proposed in this paper is carried out.After verification,the combined model can combine the advantages of each sub-model to a certain extent,thereby obtaining a more complete prediction model for future Research provides direction.