Research On Section-Network Traffic Forecasting And Road Situation Assessment And Its Application

Short term traffic flow forecasting and network-level road situation assessment can not only improve efficient use of the road network,but also can provide decision support for traffic controllers.Therefore,it is crucial to predict and evaluate the road situation accurately and efficiently.However,there are lots of redundant iterations in the traditional prediction model,and the network spatial relations are not considered either.Besides,the method of situation assessment has the problem of depending on the single index.This paper is based on a project called "the highway traffic sensor network information detection,data mining,aggregation,release and auxiliary decision-making demonstration system",which is issued by Jiangsu provincial Science and Technology Department.So this research is very important and of application value.This paper introduced the mainstream of short-term traffic flow forecasting model and network situation assessment method in intelligent transportation system at home and abroad.the advantages and disadvantages of neural network prediction model were illustrated and the shortcomings of existing clustering method for situation assessment were analyzed.Then the methods of Lasso-NN pre measurement model aimed at network and the situation assessment based on reducing the dimension of the road network were proposed.With the assessment of MATLAB on actual traffic flow data,the accuracy of the model was verified.On the problem of low efficiency algorithm and vulnerable to local optima in traditional PSO neural network short-term traffic flow prediction model,this paper analyzed the principle of the improved PSO algorithm and the characteristics of basic traffic flow data and removed the redundancy iteration during PSO optimization process and replaced fixed weight in PSO algorithm with adaptive inertia weight in order to avoid the algorithm falling into local optima.On a single neural network model,that the effect of not fully considering the spatial relationships of upstream and downstream could lead to inaccurate prediction in single section traffic flow.This paper analyzed the complex relationships of road network space and the method of basic lasso variable selection and proposed a prediction model based on Lasso-NN combination traffic.The model firstly used Lasso to select the close section of the space,and then used the characteristics of these neural network model to train and predict these sections,thus the accuracy of prediction was improved.On the shortcomings of single evidence assessment based on clustering network situation assessment algorithm and poor real time,this thesis analyzed the principle of traditional local non negative matrix decomposition LNMF algorithm and proposed 2D-LNMF for multi parameter matrix of road network.through joining 2D-LNMF algorithm before clustering algorithm to reduce the dimensionality,we ensured realtime evaluation system and improved the accuracy of the evaluation algorithm.In conclusion,the innovations and characteristics of this paper are as follows:1)improved the weight update algorithm in PSO neural network model,removed the redundant iteration during optimization,improved the accuracy of the model.2)proposed the Lasso-NN combination model,analyzed the correlations of the downstream and upstream section of the road network and selected the training model of the related section,improved the accuracy of the model prediction.3)improved the LNMF algorithm,proposed a multi parameter 2D-LNMF algorithm,ensured the real-time performance of the model and improved the accuracy of the evaluation.