Research On Short-term Traffic Flow Prediction Algorithm Based On Hadoop Platform

Short-term traffic flow prediction is an important research topic in the field of intelligent transportation, providing data support for the traffic flow guidance and control system. This paper focuses on the research of short-term traffic flow prediction based on BP neural network algorithm, using the BP network nonlinear mapping ability and the characteristic of self-learning, adaption to predict the complex short-term traffic flow. Traditional BP neural network prediction model and its improved algorithm normally couldn’t keep the balance between the high prediction accuracy and excellent calculation efficiency, therefore the problem of network training time of the traditional BP algorithm is not allowed to be ignored, especially in the condition of large traffic training sample data.This paper combines traditional BP algorithm and MapReduce parallel computing model to predict the short-term traffic flow, taking advantage of parallel computing and distributed storage of cloud computing. Meanwhile this paper uses genetic algorithm to optimize the BP neural network, considering the defects of traditional BP algorithm. The concrete research content of this paper is as follows:First, introduce the short-term traffic forecasts theoretical knowledge and analyze the advantages and disadvantages of various traffic flow prediction model,chooses the actual traffic flow to do the simulation experiment based on the BP neural network algorithm. The experiment includes the prediction of traffic flow on weekends and the prediction of traffic flow on weekdays. The result of the experiment shows that BP algorithm is easy to fall into local minima and it’s sensitive to the initial network weight variables. Slow convergence is another shortcoming of BP neural network algorithm. Facing the large scale of traffic data, the prediction of short-term traffic flow based on the BP neural network may result in the low calculation efficiency.This paper proposes improvements from two aspects against the defects of traditional stand-alone prediction mode based on BP neural network algorithm: Combine the MapReduce programming model and BP natural network algorithm to improve calculation efficiency due to the advantages of parallel computing on the cloud platform; Optimize BP neural network algorithm by generic algorithm to improve the prediction accuracy. Through the analysis and comparison of experimental result, the conclusion is that the accuracy and time consuming of genetic neural network short-term traffic flow prediction algorithm based on Hadoop platform are improved significantly.