中文摘要：

交通流预测对于路径诱导、路网管控有着重要的作用.目前预测数据源未充分挖掘调用已有的沿路视频资源,而需特地另埋设专用地感线圈;在考虑上下游空间关系时,往往将关系并不密切的点也包含进来.为此,分析了路口交通流上下游关系,解析了BP神经网络模型机理及局限,提出了基于空间聚类的短时交通流预测Cluster-NN模型,选取了遥控视频摄像数据作为预测模型的输入,对区域内交通流进行了聚类分析预测.实验结果表明,短时交通流预测值优于神经网络模型6.5%以上.

英文摘要：

Traffic flow prediction is of great significance to route guidance and network control.At present,the data sources for forecast must have a layout of a special induction coil sensor,rather than fully excavating and using having existed video resources on the way.There used to include those infirmly dots when considering spatial relationships between upstream and downstreams.Therefore,this paper selects the parameters from remote video cameras as the input of forecast model.Based on having analyzed the traffic flow relationships between upstream and downstream and BP neural network model,we proposed a Cluster-NN model for forecasting short term traffic flow.We analyzed the regional traffic flow with clustering by choosing appropriate parameters and than forecasted.The experimental results show that the forecasting accuracy with the proposed prediction model of Cluster-NN has improved about 6.5% compared to the improved neural network model in short-term traffic flow.