Short-term Traffic Flow Prediction Based On Continuous Temporal Convolution Network

Traffic flow prediction and analysis is the key link of intelligent transportation system,which can provide powerful support for the command and decision of traffic management department.With the development and application of deep learning,there are numerous traffic prediction methods based on neural networks,but the existing methods have the problems of difficulty in capturing the long-time dependence of traffic flow sequences and single prediction features.Therefore,the research of this paper is to establish a multi-source information fusion model to predict short-time traffic flow,and the main work of this paper is as follows.(1)To address the problem that Temporal Convolution Network(TCN)is difficult to capture the long-time dependence of traffic flow sequences on short-time traffic flow prediction tasks,a short-time traffic flow prediction model based on Continuous Temporal Convolution Network(CTCN)is proposed.The model introduces a continuous convolution kernel function to replace the original discrete convolution,which can improve the perceptual field of the network without the influence of expansion coefficient and network depth.The feature extraction of the Continuous Temporal Convolutional Network can be done at the bottom layer to solve the problem of over-deep network when the convolutional network is processing long sequence data.The prediction model is validated on the dataset provided by PeMS platform in California,U.S.A.The experimental results show that the CTCN model outperforms other classical models in prediction,and the mean square error in predicting future 5-minute traffic is reduced by 9.3%compared to TCN.(2)To address the problem of single prediction feature of traditional traffic flow forecasting methods,an information fusion-based short-time traffic flow forecasting model based on Information Fusion Continuous Temporal Convolution Network(IFCTCN)is proposed.The model uses CTCN as the base structure and considers the effects of multiple influencing factors on road traffic flow,which include traffic speed,lane occupancy,weather and holidays.Before the feature variables are input into the model,the weather and holiday factors need to be converted quantitatively,with weather quantified according to weather warning levels and holidays quantified according to the length of legal holidays.The model was validated with the traffic dataset provided by the PeMS platform and the weather dataset provided by the Darksky platform.The experimental results showed that the mean square error of the IFCTCN model decreased by 5.1%on average compared to the CTCN model.Figure[23]Table[17]Reference[64]...