摘要
短时交通流预测是动态交通管理与控制领域的重要问题之一,然而当前大多数交通流预测模型针对交通流的多种特性考虑不足,预测精度与稳定性较差。为提高快速路的短时交通流预测精度,建立了一种能够捕获多种交通流特性的交通流预测模型,提出了一种基于互信息量-自适应经验模态分解-自适应增强的组合预测模型。以快速路历史交通流数据为基础,利用MI加权算法处理交通流之间的时间相关性,筛选出与目标路段交通流相关性最高的历史数据。对相关性最高的历史数据进行CEEMDAN模态分解,得到多个本征模函数分量。计算了各个本征模函数分量的排列熵值(PE),筛选出能够反映目标路段交通流特性的有效分量,并将有效分量构建为重组子序列。之后,对重组后的子序列分别构建了BP-Adaboost短时交通流预测模型,将子序列的预测值叠加构成最终的短时交通流预测结果。最后,选取实际快速路交通流数据进行了验证。结果表明:所提出的MI-CEEMDAN-ADABOOST预测模型的均方根误差、平均均方误差、平均绝对百分误差均低于其他预测模型,预测误差最小,预测精度最高。针对同一预测模型而言,经过MI-CEEMDAN处理的模型各项预测误差指数明显降低,预测精度有所提高。
Short-term traffic flow prediction is one of the most important problems in dynamic traffic management and control field. However, most current traffic flow prediction models do not consider the various characteristics of traffic flow, and the prediction accuracy and stability are poor. In order to improve the prediction accuracy of short-time traffic flow, a traffic flow prediction model capable of capturing multiple traffic flow characteristics is established, and a new model MI-CEEMDAN-ADABOOST is developed based on the mutual information(MI), complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) and adaptive boosting(Adaboost) is established. Based on the historical traffic flow data of freeways, the time correlation on traffic flows is processed by using MI algorithm, the highest correlation historical data with the traffic flow of the target road section are screened. The CEEMDAN modal decomposition is performed on the historical data with the highest correlation, and multiple components of intrinsic mode function(IMF) are obtained. The permutation entropy(PE) of each IMF is calculated, the effective component that could reflect the traffic flow characteristics of the target section is selected, and the effective component is constructed as a recombination subsequence. Then, the BP-Adaboost prediction model is proposed for each restructured subsequence respectively, and the predicted values of the subsequences are superimposed to form the final short-term traffic flow prediction result. Finally, the selected traffic flow data of freeways are verified. The result shows that the that(1) the values of RMSE, MAE, MAPE of the proposed MI-CEEMDAN-ADABOOST prediction model are lower than those of other models, the prediction error is the smallest and the prediction accuracy is the highest;(2) for the same prediction model, the prediction error indexes of the model processed by MI-CEEMDAN are significantly reduced, and the prediction accuracy is improved.
作者
奇兴族
QI Xing-zu(Shenzhen Urban Transport Planning Center,Shenzhen Guangdong 518000,China)
出处
《公路交通科技》
CAS
CSCD
北大核心
2022年第6期136-143,共8页
Journal of Highway and Transportation Research and Development
基金
广东省重点领域研发计划项目(2019B110206001)。
关键词
智能交通
时间序列
短时流量预测
组合预测
预测模型
ITS
time series
short-term traffic prediction
hybrid prediction
prediction model
