Research On Short-term Traffic Flow Forecasting Based On RBF Neural Network

In recent years,the rapid increase in the number of motor vehicles has caused various traffic problems such as traffic congestion and traffic accidents,the emergence of intelligent transportation systems has effectively alleviated these problems.Short-term traffic flow prediction,as a key technology in the intelligent transportation system,can not only provide a certain decision-making basis for traffic guidance and control in the system,but also provide reasonable travel routes for residents.However,due to the strong randomness and complexity of the short-term traffic flow,it is difficult for the existing various prediction models to guarantee the accuracy of the prediction and the timeliness of the prediction results.This article has carried out research from traffic flow characteristics analysis and data preprocessing,RBF neural network and its optimization for short-term traffic flow prediction.The main work is as follows:1.Traffic flow characteristic analysis and data preprocessing.First ly,analyze the characteristics of the collected traffic data from three aspects: trend,ra ndomness and periodicity.Secondly,aiming at the problem of inaccurate data caused by interference from external factors in the process of traffic flow collection,this paper repairs the traffic data and reduces noise by heuristic wavelet threshold to pro vide data support for the following research.2.Established a traffic flow prediction model based on RBF neural network.Aiming at the problem that traditional models cannot predict short-term traffic flow well,this paper uses RBF neural network to predict traffic flow,studies the influence of the structure and structural parameters of RBF neural network on traffic flow prediction,and uses C-C method to determine the number of nodes in the input layer of the neural network,the K-means method determines the number of hidden layer neurons and the central value,and the prediction structure of the RBF neural network is established.3.Established a traffic flow prediction model based on IFA-RBF neural network.Firstly,aiming at the problem of sensitivity to the initial parameters of the network and easy to fall into local extremes when the RBF neural network predicts,this paper proposes to use the Firefly Algorithm(FA)to select the initial parameters to avoid the network from falling into local extreme values;Secondly,to solve the problem that FA is easy to fall into "premature",by introducing linearly decreasing inertia weight and chaotic search mechanism into basic FA,the prediction model of IFA-RBF neural network is established,which further improves t he prediction accuracy and convergence speed of RBF neural network.4.Established a traffic flow prediction model based on FCM-IFA-RBF neural network.Aiming at the problem of low prediction accuracy of a single RBF neural network and insufficient consideration of traffic flow characteristics,this paper proposes to start from the characteristics of the traffic data,use the FCM algorithm to divide the traffic mode of the data,and train different IFA-RBF neural networks for different modes,and all sub-models jointly complete the prediction,which further improves the accuracy of neural network prediction.