关键词: BP神经网络, 集成预测, MapReduce, 加权平均值, 交通路口分流流量大小的预测

Abstract:

BP neural network provides a robust and effective learning method for approximating real values. It is fit for intersection traffic flow prediction. In order to resolve its slow convergence speed and easily falling into local minimum problem，an ensemble prediction model was proposed. This model integrated multiple BP neural networks which had different initial weights and training set and used weighted average as combination method. It had used an improved MapReduce method to implement every BP neural network of the ensemble prediction model. This ensemble prediction model took traffic shunt flow prediction at intersection as an example. At last，it was compared with simple single implementation BP model and MapReduce implementation BP model respectively. Finally，results prove that ensemble prediction model has a higher accuracy rate and real-time performance in traffic shunt flow prediction at intersection.

Key words: BP neural network, ensemble prediction, MapReduce, weighted average, traffic shunt flow prediction at intersection