Abstract: The methodology of short-term traffic flow forecasting is presented in this paper based on Elman neural network, which sets sub-network as objectives. To simplify road network analysis, and to reduce dimension of solution space, this paper extracts space characteristics of traffic flow and split the road network more scientific and rational based on general space distance. Then, it introduces Elman neural network to forecast traffic flow of multi-sections in the road network whose input vector is constructed by time series of traffic flow. In the last section, the methodology is tested using traffic flow data from the road network, which is compared to the method of BP neural network. The methodology can split the road network into several sub-networks satisfying short-term forecasting demand, and short-term traffic flow forecasting of the road network is realized by Elman neural network. The results are proved to be super to the method of BP neural network.

Key words: urban traffic, short-term traffic flow forecasting, road network splitting, general space distance, Elman neural network, BP neural network

摘要： 以道路子网为研究对象，采用Elman神经网络实现道路网多断面交通流短时预测. 首先通过提取交通流空间特性对道路网进行划分，降低道路网整体分析复杂度及解空间维数，提高交通流预测的计算精度和效率；其次以实时采集的交通流数据为基础，并以重构的交通流时间序列作为输入，采用Elman神经网络实现道路网多断面交通流同时预测；最后，基于城市快速路多断面交通流量数据对短时交通流预测方法进行验证，并与BP神经网络预测结果进行对比分析. 验证结果表明，本文提出的道路网划分方法能够划分出满足预测需求的子路网，在划分的子路网上，应用Elman神经网络能够实现道路网多断面同时预测，且预测效果优于BP神经网络.

关键词: 城市交通, 交通流短时预测, 道路网划分, 广义空间距离, Elman神经网络, BP神经网络