交通流量VNNTF神经网络模型多步预测研究

研究了VNNTF 神经网络（Volterra neural network trafficflow model，VNNTF） 交通流量混沌时间序列多步预测问题. 通过分析比较交通流量混沌时间序列相空间重构的嵌入维数和Volterra 离散模型之间的关系，给出了确定交通流量Volterra 级数模型截断阶数和截断项数的方法，并在此基础上建立了VNNTF 神经网络交通流量时间序列模型；设计了交通流量Volterra 神经网络的快速学习算法；最后，利用交通流量混沌时间序列对VNNTF 网络模型，Volterra 预测滤波器和BP 网络进行了多步预测实验，比较了多步预测结果的仿真图、绝对误差的柱状图以及归一化后的方均根；实验结果表明VNNTF 神经网络的多步预测性能明显优于Volterra 预测滤波器和BP 神经网络.

Research on the Multi-step Prediction of Volterra Neural Network for Traffic Flow

This paper studies multi-step prediction of traffic flow chaotic time series based on Volterra neural network traffic flow model (VNNTF). Firstly, by analyzing the relationship between the embedding dimension of phase space reconstruction of traffic flow chaotic time series and Volterra discrete model, we give the method to determine the truncation order and items of Volterra series. Secondly, based on the first step, we build the VNNTF neural networks model of chaos time series and design the fast learning algorithm of Volterra neural network traffic flow. Thirdly, we describe multi-step prediction experiments based on chaotic time series VNNTF traffic network model, Volterra prediction filter and BP networks. Finally, we compare the multi-step prediction simulation diagram with the absolute error histogram and normalized root mean square are compared. The experimental results show that the VNNTF neural network multi-step prediction performance is significantly better than those of the Volterra filter and BP neural network.