递推SOM神经网络在短时交通流预测中的应用

Application of the Recursive SOM Neural Network in Short-term Traffic Flow Prediction

针对短时交通流预测,提出递归自组织映射(SOM)神经网络方法。根据SOM神经网络的联想记忆技术,分别给出考虑了反馈的Rec SOM模型和能够利用结构化信息的SOMSD模型。递推SOM方法用全SOM作为重复神经元,用历史活动与当前信息的组合作为输入,通过训练神经元的权值及上下文信息学习时序动态。将递推SOM方法应用于预测某地区实测交通流数据,并与现有方法进行比较。试验结果表明,递推SOM方法能有效改善预测精度,在同等情况下优于其他方法。

For short-term traffic flow prediction, the method of recursive self-organizing map （SOM） neural network is proposed. On the basis of the associative memory technology of SOM neural network, the ResSOM model considering feedback and the SOMSD model capable using structured information are given respectively. The method of recursive SOM uses full SOM as the replicated neuron, and the eombination of historical activities and current information as the input; it learns time series dynamics by training the weight value of neuron and context information. The method of recursive SOM is applied in prediction of measured traffic flow data of an area, and comparison with existing method is conducted, the experimental results show that the method of recursive SOM effectively improves the prediction accuracy ; it is superior to other methods under the same circumstances.