Short-term traffic volume prediction, which can assist road users in choosing appropriate routes and reducing travel time cost, is a significant topic of intelligent transportation system. To overcome the error magnification phenomena of traditional combination methods and to improve prediction performance, this paper proposes an improved Bayesian combination model with deep learning (IBCM-DL) for traffic flow prediction. First, an IBCM framework is established based on the new BCM framework proposed by Wang. Then, correlation analysis is used to analyze the relevance between the historical traffic flow and the traffic flow within the current interval. Three sub-predictors including the gated recurrent unit neural network (GRUNN), autoregressive integrated moving average (ARIMA), and radial basis function neural network (RBFNN) are incorporated into the IBCM framework to take advantage of each method. The real-world traffic volume data captured by microwave sensors located on the expressways of Beijing was used to validate the proposed model in multiple scenarios. The overall results illustrate that the IBCM-DL model outperforms the other state-of-the-art methods in terms of accuracy and stability.

Accurate and efficient lane-level traffic flow prediction is a challenging issue in the framework of the connected automated vehicle highway system. However, most existing traffic flow forecasting methods concentrate on mining the spatio-temporal characteristics of the traffic flow rather than increasing predictability of traffic flow. In this paper, we propose a novel hybrid model (CEEMDANXGBoost) for lane-level traffic flow prediction based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and extreme gradient boosting (XGBoost). The CEEMDAN method is introduced to decompose the raw traffic flow data into several intrinsic mode function components and one residual component. Then, the XGBoost methods are trained and make predictions on the decomposed components respectively. The final prediction results are obtained by integrating the prediction outputs of the XGBoost methods. For illustrative purposes, the ground-truth lane-level traffic flow data captured by remote traffic microwave sensors installed on the 3rd Ring Road of Beijing are utilized to evaluate the effectiveness of the CEEMDAN-XGBoost model. The experimental results confirm that the CEEMDANXGBoost model is capable of fitting the complex volatility of traffic flow efficiently at different types of lane sections. Moreover, the proposed model outperforms the state-of-the-art models (e.g., artificial neural networks and long short-term memory neural network) and other XGBoost-based models in terms of prediction accuracy and stability.

Lane-level traffic state prediction is one of the most essential issues in the connected automated vehicle highway systems. Accurate and timely traffic state prediction of the lane sections can assist the connected automated vehicles in planning the optimal route and making lane selection. In this article, we tackle the problem of forecasting lane-level short-term traffic speed and propose a novel mixed deep learning (MDL) model by coordinating the convolutional long short-term memory (Conv-LSTM) layers, convolutional layers, and a dense layer in an end-to-end structure. The introduction of the Conv-LSTM neural network enables the proposed MDL model to better capture the spatio-temporal characteristics and correlations of the dynamic lane-based traffic flow synchronously. To improve the efficiency of the proposed model, a feature correlation analysis method based on the maximum information coefficient is presented to measure the relevance between the historical traffic flows and the traffic speeds to be forecasted. Validated by the ground-truth traffic flow data collected by the remote traffic microwave sensors installed on the expressways in Beijing, the MDL model is capable of capturing the fluctuation of the lane-level traffic speeds at different types of lanes effectively during the whole day. Furthermore, the results confirm that the MDL model achieves better predictive performance than several state-of-the-art benchmark models in terms of prediction accuracy and space-time distributions. Our code and data are available at https://github.com/lwqs93/MDL.

Safety and efficiency have always been significant challenges to the development of road traffic. Detailed vehicle motion information is the prerequisite for achieving optimal control of the platoon and improving traffic safety and efficiency. The connected and automated vehicles (CAVs) system has offered unprecedented opportunities for the real-time collection and processing of these detailed vehicle motion data. Based on the model predictive control (MPC) framework and safety potential field (SPF) model, we developed an alternative CAVs platoon dynamic control method. The SPF model was applied to describe the road risk distribution under the complex driving environment and was embedded in the MPC framework to optimize the vehicle dynamics from the perspective of capacity, safety, and energy-saving. Also, some experiments were performed to verify the validity of our platoon control strategy. Compared with the fixed time-headway strategy, our proposed strategy can increase the traffic capacity by about 24.4%, while ensuring safety and improving fuel economy. The results indicate that the novel CAVs platoon control methodology proposed in this paper can be potentially applied to alleviate various traffic problems (e.g., traffic congestion, traffic accidents, and high emissions).