基于BF-SVR-GRU的短时交通流预测方法

Short-term Traffic Flow Prediction Method Based on BF-SVR-GRU

短时交通流预测是智能交通管理的重要依据。为了提高短时交通流预测的精度,从交通流内在的稳态特征和动态特征着手,提出一种基于巴特沃兹滤波(Butterworth filter,BF),结合支持向量回归(support vector regression,SVR)算法和门控循环单元(gated recurrent unit,GRU)模型的预测方法,即BF-SVR-GRU模型。该方法先对交通流标准化处理,以加快后续模型计算的速度。通过设置适当阈值,利用巴特沃兹滤波将交通流信息分解为稳态分量和动态分量:稳态分量反映交通流总体变化趋势,动态分量反映突发因素(如交通事故、天气影响等)对交通流的影响。利用门控循环单元对稳态分量进行训练与预测,克服门控循环单元在预测变化剧烈的序列精度较低的问题;支持向量回归对非线性序列预测存在适应性较好、低泛化误差等优点,利用支持向量回归对动态序列进行训练与预测。最后,将稳态分量与动态分量的预测结果整合得到最终预测结果。采用某市不同的两个路口的数据集进行相关实验,结果表明,BF-SVR-GRU预测方法具有较好的预测精度,可为智能交通规划与管理提供有效的建议。

Short-term traffic flow prediction is an important basis for intelligent traffic management.In order to improve the prediction accuracy,starting from the inherent steady-state and dynamic characteristics of traffic flow,a method based on Butterworth filtering,SVR and GRU was proposed.The method was named BF-SVR-GRU.First,the traffic flow was standardized by this method to speed up the subsequent model calculations.To set the appropriate threshold,Butterworth filtering was used to divide steady-state components and dynamic components from the traffic flow information.The general trend of traffic flow was reflected in the steady-state component,and the impact of uncertain factors(such as traffic accidents,weather effects,etc.)was reflected in the dynamic component.Aiming at the problem of GRU’s low accuracy in predicting rapidly changing sequences,steady-state sequences were used to train and predict GRUs;considering the advantages of SVR’s better adaptability to non-linear sequences and low generalization errors,SVR was used to train and predict dynamic sequences.Finally,the results of steady-state components and dynamic components predictions were integrated as the final prediction result.The datasets of two different intersections in a certain city were used to carry out related experiments.The results show that the BF-SVR-GRU prediction method has better prediction accuracy,and can provide effective suggestions for intelligent transportation planning and management.