Research On Road Traffic Flow Short-Term Prediction Algorithm And The Application And Implementation Of System

With China’s auto demand and population obtained continuous and fast growth. The problems resulting from this growth such as traffic congestion, traffic safety and exhaust pollution have severely impacted the living environment of urban residents. The present studies show that the construction of informatization and intelligent traffic management system is an effective way to improve traffic management level and alleviate urban road traffic problems.In our research, we mainly study the road traffic flow short-term prediction and construction of the intelligent traffic management and control system based on the project o f Nansha District, Guangzhou. The research was launched from the following aspects:(1) We briefly reviewed the study background and research significance of this study. And we also explained the base knowledge of road traffic flow, road traffic flow short-time forecasting, the acquisition of traffic flow data and the pretreatment methods of the data based the equivalent road traffic flow.(2) From the study of the classical BP neural network prediction algorithm, we found it has some shortcomings such as slow convergence velocity and could easily fall into local minimum and so on. To overcome these shortcomings, gradient descent method and Newton’s method were utilized to modify it,then, the modified BP neural network uses the equivalent road traffic flow to do the simulation. The simulation indicated that the modified BP neural network has better prediction results, except in the case of traffic flow mutations.(3) Considering the drawbacks of the modified BP neural network model, two kinds of hybrid models based on K-means algorithm and experience were established to improve the problem. In the model, traffic flow was divided into several associated regions based on the tradition experience or K-means algorithm. Then we established the appropriate improved BP neural network forecasting model in each region. Simulation revealed that the two kinds modified hybrid models can significantly improve the prediction accuracy in the case of road traffic flow mutation and the two kinds of the hybrid models can improve the prediction with the increment of 5.62% and 11%. The comprehensive comparative study showed the modified hybrid model based on K-means algorithm was superior to the modified hybrid based on artificial expertise in all indicators in the the increment of 6.39 %.(4) The business requirements as well as the construction requirements of the intelligent traffic control system were comprehensively taken into account. We designed the logic functions of the entire system, the overall technical framework, the high availability of the database system and business model in detail. Meanwhile, the realization and the test of the entire intelligent traffic management and control system were also systemically introduced and studied.