摘要
为了提高短期交通流的预测精度,提出了一种基于小波包分解(wavelet packet decomposition,WPD)、粒子群优化(particle swarm optimization,PSO)算法和回声状态网(echo state network,ESN)的短期交通流预测方法。该方法命名为WPD-PSO-ESN。首先,在数据预处理阶段,采用小波包分解将交通流数据分解为不同频段的子序列,并将各子序列送入回声状态网预测模型;然后,在建立预测模型阶段,利用粒子群优化算法在线优化回声状态网的参数,以提高回声状态网的泛化能力和预测精度;进一步,针对粒子群优化算法存在的早熟收敛和易陷入局部最优的缺陷,通过检测粒子飞行过程中的状态信息,设计了惯性权重自适应调整策略,以期提高粒子群优化算法的寻优能力;最后,在结果输出阶段,采用加权平均法融合各子序列的预测值以得到模型的最终预测结果。试验结果表明:通过小波包分解和单支重构可以更加容易地抓住原始信号中的动态信息,更适合用于回声状态网的时间序列建模;带有自适应惯性权重调整策略的粒子群优化算法具备更强的跳出局部最优的能力,优化后的回声状态网模型精度更高;对于短期交通流预测,与前馈型误差反传神经网络、反馈型Elman神经网络和传统回声状态网等预测方法相比,WPD-PSO-ESN预测方法具有更高的预测精度,能够满足智能交通系统对预测精度的需求,对实现实时交通控制和建设智能交通系统具有重要意义。
In order to improve the prediction accuracy of short-term traffic flow, a short-term traffic flow prediction method based on WPD, PSO and ESN is proposed. This method is named WPD-PSO-ESN. First, in the data preprocessing stage, the traffic flow data are decomposes into sub-sequences of different frequency bands by WPD, and each sub-sequence is sent to the ESN prediction model. Then, in the prediction modelling stage, the parameters of the ESN are optimized by PSO algorithm to improve the generalization ability and prediction accuracy of the ESN. Furthermore, aiming at the premature convergence and easy to fall into local optimality of PSO algorithm, an adaptive adjustment strategy of inertia weight is designed by detecting the state information during particle flight to improve the optimization ability of PSO algorithm. Finally, in the result output stage, the predicted values of each subsequence are fused by weighted average method to obtain the final prediction of the model. The experimental result shows that (1) the dynamic information in the original signal can be more easily grasped by WPD and single-branch reconstruction, which is more suitable for time series modelling for ESN;(2) the PSO algorithm with adaptive inertia weight adjustment strategy has stronger ability to jump out of local optimum, and the optimized ESN model has higher precision;(3)for short-term traffic flow prediction, compared with feedforward error back propagation neural network, feedback Elman neural network and traditional ESN prediction method, the proposed WPD-PSO-ESN prediction method has higher prediction accuracy and can meet the requirements of prediction accuracy in ITS, which is of great significance for real-time traffic control and ITS construction.
作者
万玉龙
李新春
周红标
WAN Yu-long;LI Xin-chun;ZHOU Hong-biao(School of Management,China University of Mining and Technology,Xuzhou Jiangsu 221116,China;School of Business,Huaian Vocational College of Information Technology,Huai'an Jiangsu 223003,China;School of Automation,Huaiyin Institute of technology,Huai'an Jiangsu,223003,China)
出处
《公路交通科技》
CAS
CSCD
北大核心
2019年第8期144-151,共8页
Journal of Highway and Transportation Research and Development
基金
江苏省2017年度现代教育技术研究立项课题(基于现代信息技术的冷链物流智能配载推荐系统的设计与研究2017R58204)
关键词
城市交通
时间序列预测
回声状态网络
小波包分解
粒子群优化
短期交通流
urban traffic
time series prediction
echo state network (ESN)
wavelet packet decomposition(WPD)
particle swarm optimization (PSO)
short-term traffic flow prediction
