基于改进狼群算法的小波神经网络短时交通流预测

发布时间：2018-03-20 14:10

本文选题：短时交通流预测　切入点：小波神经网络　出处：《西南交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：在智能交通系统中,实时准确的短时交通流预测一直是各国学者研究的重点。为了提高预测精度,越来越多的组合模型被应用到此领域,其中小波神经网络模型结合了小波分析与神经网络的优势,对短时交通流的预测具有较好的效果,狼群算法(Wolf Pack Algorithm,WPA)是新近提出的优化算法,具有较好的全局收敛性,为此,本文基于小波神经网络,将狼群算法与改进的梯度下降算法相结合,用于短时交通流预测。首先从美国明尼苏达德卢斯大学交通数据研究实验室以及加州运输性能测量系统中获得了 5个交通流数据集;然后对每个数据集的数据进行修复、小波降噪、相空间重构和归一化等预处理,用于所有模型的仿真实验;最后使用Morlet小波神经网络(Wavelet Neural Network,WNN)对短时交通流进行预测。仿真实验结果表明,小波神经网络模型能够对交通流的整体趋势进行预测,但稳定性和预测精度还有待提高。针对小波神经网络中梯度下降算法对权值和小波因子初值敏感,容易陷入局部极小值的缺点,本文将狼群算法与梯度下降算法结合,先利用狼群算法的全局寻优能力为小波神经网络找到一组较优的权值和小波因子,再通过梯度下降算法对权值和小波因子寻优,仿真实验结果表明狼群算法与梯度下降算法的结合是有效的。在此基础上对狼群算法进行了改进,仿真实验结果表明,IWPA-WNN模型有效地提高了短时交通流预测的稳定性和精度,同时缩短了运行时间。其次,为了进一步提高短时交通流预测的精度,本文将误差补偿方法(Error Compensation,EC)应用到小波神经网络短时交通流预测中,使用小波神经网络对交通流预测的误差数据进行二次信息提取,仿真实验结果表明,加入了误差补偿的小波神经网络能够有效地提高短时交通流预测的精度。最后将误差补偿方法与改进狼群算法的小波神经网络有机结合构成了 EC-IWPA-WNN短时交通流预测模型。仿真实验结果表明基于EC-IWPA-WNN模型的短时交通流预测在稳定性和精度上都具有良好的性能。
[Abstract]:In intelligent transportation system, real-time and accurate short-term traffic flow prediction has been the focus of scholars all over the world. In order to improve the prediction accuracy, more and more combined models have been applied to this field. The wavelet neural network model combines the advantages of wavelet analysis and neural network, and has a good effect on short-term traffic flow prediction. Wolf Pack algorithm is a newly proposed optimization algorithm, which has good global convergence. Based on wavelet neural network, this paper combines the improved gradient descent algorithm with the wolf swarm algorithm. First, five traffic flow data sets were obtained from the Traffic data Research Laboratory of the University of Minnesota, Duluth, and the California Transportation performance Measurement system; then the data of each data set was repaired. The pretreatment of wavelet denoising, phase space reconstruction and normalization is used in simulation experiments of all models. Finally, Morlet wavelet Neural network is used to predict short-term traffic flow. The simulation results show that, The wavelet neural network model can predict the overall trend of traffic flow, but the stability and prediction accuracy need to be improved. The gradient descent algorithm in wavelet neural network is sensitive to the weights and the initial values of wavelet factors. It is easy to fall into local minima. In this paper, the wolf swarm algorithm is combined with the gradient descent algorithm. Firstly, the global optimization ability of the wolf swarm algorithm is used to find a set of better weights and wavelet factors for the wavelet neural network. Then the weight and wavelet factor are optimized by gradient descent algorithm. The simulation results show that the combination of wolf swarm algorithm and gradient descent algorithm is effective. The simulation results show that the IWPA-WNN model can effectively improve the stability and accuracy of short-term traffic flow prediction and shorten the running time. Secondly, in order to further improve the accuracy of short-term traffic flow prediction, In this paper, the error compensation method is applied to short-term traffic flow prediction based on wavelet neural network. The error data of traffic flow prediction is extracted by wavelet neural network. The simulation results show that, Wavelet neural network with error compensation can effectively improve the accuracy of short-term traffic flow prediction. Finally, combining the error compensation method with the wavelet neural network of improved wolf swarm algorithm, EC-IWPA-WNN short-time traffic flow prediction is formed. The simulation results show that the short-time traffic flow prediction based on EC-IWPA-WNN model has good stability and accuracy.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491.14;TP18

【参考文献】