The Research On Algorithm Of Short-term Traffic Flow Prediction

Nowadays, with the continuous expansion of city scale, the number of vehicles is creasing, and the city traffic situation gets worse. Governments are actively to build intelligent traffic guidance system to ease the traffic pressure and improve people’s quality of life. Short-term traffic flow prediction is the core technology to build the intelligent traffic guidance system, and it’s the main direction of this paper.Through a large number of historical traffic flow data, we conclude that the important characteristics of the main traffic flow is periodic, highly nonlinear and uncertainty. While the neural network algorithm is a kind of method which has good prediction ability for stochastic nonlinear data, with fault tolerance and good learning, but there are also some disadvantages such as easy to local optimization. In this paper, the establishment of short-term traffic flow forecasting model process is based on the conventional artificial neural network prediction algorithm. The method to obtain the ideal effect to try a variety of nonlinear field to improve neural network, prediction on traffic flow and applications, improves the prediction results. The research mainly includes the following aspects:(1) This dissertation analyzes the basic concepts of traffic flow, traffic flow forecasting data acquisition and pre-processing method, researches a variety of advantages and disadvantages on traffic flow forecasting models, and gives four evaluation algorithm performance indicators.(2) A large amount of historical data are collected and treated. The BP neural network algorithm carries on the forecast to the historical data, and the prediction performance is obtained through the simulation.(3) This dissertation adopts wavelet neural network algorithm model of short-time flow forecasting to improve the convergence speed and performance prediction. The simulation results verify that the prediction result is better than the traditional neural network.(4) This dissertation adopts the particle swarm algorithm of wavelet neural network weights threshold optimization which improves particle swarm optimization algorithm based on the algorithm. The simulation results verify that the improved algorithm’s prediction is slower in speed, but the accuracy obtains better effect.(5) Based on the particle swarm optimization neural network, the short-term traffic flow prediction algorithm is applied to the actual traffic prediction software, and the data is processed and sent through the Network Assistant. The client software through the algorithm realizes the prediction module function, and reflects its actual application value.