| Traffic forecast as an important research direction of intelligent transportation research can effectively provide effective safeguards for the construction of urban road, the control of intelligent traffic and soon on.Through the analysis of a large number of past traffic data and combining with neural network theory, we can build an accurate data predictive model, the traditional network predictive model is limited by training sample is too small and the initializing weights is chosen inappropriately. So these reasons lead to the failure for the prediction. We can easily solve these two problems by parallel and distributed storage technologies. In order to establish a more accurate predictive model, this paper combined the traditional neural network algorithm with the parallel programming techniques of MapReduce.At the same time,we choose HBase as the distributed database to store sample data, establishing a neural network prediction model based on Hadoop clusters, which used to predict the actual problems in traffic.The main works and research results of this article are as follows:(1) Through researching and analyzing the more extensive parallel programming technology, we determine to use the MapReduce parallel programming framework, and the Hadoop cluster as an experimental environment. We realize the visual monitoring of the Linux cluster and Job monitoring services and provide a components which can install cluster without network.(2) Based on the analyzing of the neural network forecasting model and the serial MapReduce programming model, try to merge MapReduce and the neural network algorithm, then we propose BP-MR which predict well also through enhanceing the randomness of the initial weights and providing parallel training data.(3)After Data cleansing we save data into the HBase as the data source of MapReduce program. Through to compare the BP-MR forecasting model with the BP model can get the advantages and disadvantages of the BP-MR, then we give an improved ideas. |
PDF Full Text Request