The Research And Implement On The Vehicle's Dynamic Shortest Path Based On Genetic Algorithm And Neural Network

In recent years, many developed countries such as Europe, the United States andJapan have begun extensive research jobs on intelligent transport system, for shortITS, but it started late in our country. A key of developing ITS is to offer real-timetraffic flow information and reduce traffic jam. However reducing the vehicle's traveltime is an important precondition to reduce traffic jam. At the same time for drivers,they also hope get dynamic shortest path under the real-time traffic condition. City traffic network is a complicated and dynamic network where road traveltime change dynamically and travel time is decided by traffic flowvolume and trafficdensity. It is very essential to forecast the future traffic flow in road networkaccurately and dynamically for offering real-time traffic flow information. Because the factors of influencing travel time and traffic flow have the highdynamic and non-linear characters. So it is difficult to give an accurate expressionformula. And the artificial neural network(ANN, Artificial Neural Network) havenon-linear description,self-learning and self-adapting,certain fault-tolerance and suitto deal with multi-variable system. This paper shows that an three-layer feedforwardBP neural network with combination of momentum and learning rate adaptiveadjustment win a satisfying short-term forecast result through survey data artificialexperiment. But BP network has many intrinsic defects, the structure is difficult to confirm,the blindness that initial weights is chosen results in slow convergence speed andeasily falling into local minimum. And the genetic algorithm (GA, Genetic Algorithm)is an iterative and self-adapting stochastic searching algorithm based on naturalselection and heredity mechanism, it has many virtues including global searchingability and fast convergence speed, etc. This paper makes use of genetic algorithm notonly to train initial weights and thresholds of BP network, but the structure of thenetwork, and overcomes the above-mentioned defects of BP network. Up to now, a lotof documents have only solely optimized weights or the structure by geneticalgorithm.This paper sets up a traffic flow forecasting model based on the combination ofgenetic algorithm and BP network. Utilizing the traffic flow data for every period ofsome time at present and in the past, the model can forecast the future traffic flowvolume accurately. Through survey data artificial experiment show genetic neuralnetwork model have better forecasting result in traffic flow short-term forecast thansimple neural network. This paper realizes the model in MATLAB language to forecast the future trafficflow volume and traffic density in road network, and then the real-time travel time canbe got. Considering the characteristic of dynamic path and limitation of restraintconditions, this paper has done some supplement on Dijkstra algorithm and someimprovement on its realization way to obtain a high efficient and dynamic shortestpath which can avoid the traffic jam at the same time, and this improved Dijkstraalgorithm is realized in JAVA language.