Optimization of an intra-city express delivery network from three to two levels is of great interest to suppliers and customers for reducing costs and improving service efficiency. One feasible solution is to identify critical nodes in the three-level network and upgrade them as transshipment facilities in the two-level one. However, traditional optimization models seldom combine empirical business data, composite metrics, and objective evaluation rules. We proposed an approach integrating empirical data, multi-criteria decision-making methods based on the real-world application of the SF Express Chengdu branch. We also developed a mathematical optimization model using statistical and operations management techniques combined with logistics expertise for a location decision. First, the appropriateness of each service point as a candidate transshipment facility is evaluated from internal and external perspectives by applying multiple centrality assessment from complex network theory and fuzzy Technique for Order Preference by Similarity to an Ideal Solution, respectively. Second, 16 candidate transshipment facilities are selected by combining these two ways. Then, a multi-objective integer programming model is built to obtain the optimal number, locations of transshipment facilities, and the corresponding service points covered by each transshipment facility. Using this multi-methodologic approach, we show that the optimized two-level network is economically feasible and simply applicable, with the total cost and average delivery time reduced by 18.41% and 6 h, respectively. This article is of practical significance and provides an important reference for optimizing ground express service networks for other large cities.
Traffic congestion is a phenomenon that has been extensively explored by researchers due to its impact on reliability and safety. This research is focused on proactively detecting and mitigating congestion on freeways by fuzing conventional traffic data obtained from radar and loop detectors with newer sources, such as Bluetooth and connected vehicles (CV). Data-driven and signal-processing techniques are explored to develop algorithms that use near- or real-time traffic measurements to predict the onset and intensity level of traffic congestion. The developed algorithm can be applied to both conventional and low penetration CV-based datasets to identify four types of congestion, that is, normal, recurring, other non-recurring, and incident. This research also demonstrates the advantage of using CV-based travel time estimates to calibrate microsimulation models over fixed point-based derivations of travel time from spot speeds. Finally, a set of mitigation strategies consisting of speed harmonization and dynamic rerouting are implemented in the calibrated simulation network to demonstrate their effectiveness in proactively reducing recurring and non-recurring congestion. The final derived algorithm is effective in proactively predicting the onset of congestion and its intensity level, with an overall mean prediction error of 30.2%. A limitation to the algorithm’s methodology is that it cannot disentangle the type of congestion when two or more are occurring simultaneously and only predicts/classifies the anticipated highest level. However, this does not impair the user’s ability to readily deploy appropriate mitigation strategies to alleviate the predicted intensity of congestion.
Short term forecasting is essential and challenging in time series data analysis for traffic flow research. A novel deep learning architecture on short-term traffic flow prediction was presented in this work. In conventional model-driven prediction method, a critical deviation in prediction accuracy was occurred in face of large fluctuations in traffic flow, while machine and deep learning-based approaches performed well in accuracy study than conventional regression-based models. Moreover, a fusion attention mechanism bidirectional long short-term memory model (ATT-BiLSTM) was proposed due to its bidirectional LSTM (BiLSTM) and attention mechanism units. The model not only dealt with forward and backward dependencies in time series data, but also integrated the attention mechanism to improve the ability on key information representation. The BiLSTM layer was exploited to capture bidirectional temporal and spatial features dependencies from historical data. The proposed model was also trained and validated using freeway toll datasets from Humen Bridge. The results showed that compared with ARIMA and SVR models, the indicators of the proposed model have been significantly improved. The ablation experiments were conducted to evaluate the role of the attention mechanism module. Compared with BiLSTM, CNN and 1DCNN-ATT-BiLSTM models, the MAE, RMSE and MAPE indexes of proposed model were reduced by 0.6–5.9%, 1.6–4.7% and 0.6–22.8%, respectively. More accurate predictions were obtained by the proposed model. The research results are of great significance to improve the level of traffic management.
The global COVID-19 pandemic has had a great impact on transportation across the United States. However, there is a lack of studies investigating the pandemic’s impact on vehicular traffic at the later stage of the pandemic. Therefore, this paper studies the change of freeway traffic patterns in two metropolitan counties in the State of Utah at the latter stage of the pandemic. We found that with the relaxation of travel restriction and the COVID vaccine, vehicular traffic has recovered to equaling, if not exceeding, pre-pandemic levels. Truck traffic is higher than the pre-pandemic level due to the growth of online shopping and on-demand delivery. To help responsive agencies to prepare for the near-future traffic pattern, a traffic prediction model based on an innovative approach integrating machine learning with graph theory is proposed. The evaluation shows that the proposed prediction model has a desirable performance. The mean absolute percentage prediction error is between 0.38% and 1.74% for different jurisdictions. On average, the modal outperforms the traditional long short-term memory model by 31.20% in terms of root mean squared prediction error.