基于交通数据分解和时空特征提取的车辆平均车速预测模型  

Vehicle Average Speed Prediction Model Based on Traffic Data Decomposition and Spatio-Temporal Feature Extraction

作　　者：丁祥颖 胡尧 刘智权[1] 

机构地区：[1]贵州大学数学与统计学院,贵州 贵阳 [2]贵州大学公共大数据重点实验室,贵州 贵阳

出　　处：《运筹与模糊学》2023年第2期415-430,共16页Operations Research and Fuzziology

摘　　要：从交通数据分解和时空特征提取的角度出发,提出建立基于交通数据分解和时空特征提取(Traffic Data Decomposition and Spatio-Temporal Feature Extraction, TDD + STFE)的预测模型对城市道路交叉口车辆平均车速进行预测。模型首先借助Fourier变换将交通数据中的线性部分即周期项转换为Fourier级数的形式,再用原始数据减去线性部分得到非线性部分。继而借助卷积神经网络–门控循环单元(Convolutional Neural Network-Gated Recurrent Unit, CNN-GRU)模型提取交通数据非线性部分的时空特征,最后将两部分的预测值相加即为最终预测值。通过实际交通数据验证,表明本文所提车辆平均车速预测方法同时具备实用性和有效性,对交通运行状态的评估和预警具有一定的指导意义。From the point of view of traffic data decomposition and spatio-temporal feature extraction, a method based on traffic data decomposition and spatio-temporal feature extraction (TDD + STFE) is proposed to predict the average vehicle speed at urban road intersections. The model first converts the linear part of traffic data, that is, the periodic term, into the form of Fourier series with the help of Fourier transform. Then subtract the linear part from the original data to get the nonlinear part, and then extract the temporal and spatial feature of the nonlinear part of traffic data with the help of the Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) model. Finally, the final predicted value is the sum of the predicted values of the two parts. Through the verification of the actual traffic data set, the results show that the average vehicle speed prediction method proposed in this paper is both practical and effective, which has certain guiding significance for the evaluation and early warning of traffic operation state.

关 键 词：交通数据分解 Fourier级数理论 循环神经网络 卷积神经网络 时空特征 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]