There is a growing interest in the optimization of vehicle fleets management in urban environments. However, limited attention has been paid to the integrated optimization of electric taxi fleets accounting for different operations as well as complex spatiotemporal demand dynamics. To this end, this study develops a real-time recommendation framework based on deep reinforcement learning (DRL) for electric taxis (E-taxis) to improve their system performance with explicit modeling of multiple vehicle actions and varying travel demand across space and over time. Spatiotemporal patterns of urban taxi travels are extracted from large-scale taxi trajectories. Spatiotemporal strategies are proposed to coordinate E-taxis’ repositioning and recharging with optimized recommendation for next destinations and charging stations. A spatiotemporal double deep Q-network (ST-DDQN) is embedded in the DRL framework to maximize the daily profit. A prototype real-time recommendation system for E-taxis is implemented for the decision-making of E-taxi drivers and sensitivity analyses are carried out. The experimental results in Shenzhen, China suggest that the proposed framework could improve the overall performance. This study will benefit the promotion of connected E-taxis and the development of clean and smart transportation.

Segmentation of intracranial aneurysms (IAs) is an important step for the diagnosis and treatment of IAs. However, the process by which clinicians manually recognize and localize IAs is overly labor intensive. This study aims to develop a deep-learning-based framework (defined as FSTIF-UNet) towards IAs segmentation in un-reconstructed 3D Rotational Angiography (3D-RA) images. 3D-RA sequences from 300 patients with IAs from Beijing Tiantan Hospital are enrolled. Inspired by radiologists' clincial skills, a Skip-Review attention mechanism is proposed to repeatedly fuse the long-term spatiotemporal features of several images with the most obvious IA's features (sellected by a pre-detection network). Then, a Conv-LSTM is used to fuse the short-term spatiotemporal features of the selected 15 3D-RA images from the equally-spaced viewing angles. The combination of the two modules realizes the full-scale spatiotemporal information fusion of the 3D-RA sequence. FSTIF-UNet achieves DSC, IoU, Sens, Haus, and F1-Score of 0.9109, 0.8586, 0.9314, 1.358 and 0.8883, respectively, and time taken for network segmentation is 0.89 s/case. The results show significant improvement in IA segmentation performance with FSTIF-UNet compared with baseline networks (with DSC from 0.8486 - 0.8794). The proposed FSTIF-UNet establishes a practical method to assist the radiologists in clinical diagnosis.

In recent years, point clouds have been widely used in powerline inspection, smart cities, autonomous driving, and other fields. The deep learning-based point cloud processing methods have attracted more and more attention due to the developments of laser scanning technology and machine learning. However, the recent methods largely ignore global contextual relationships and do not make full use of the complementation between local feature and high-level geometric information. To the problem, we propose a novel deep neural network, namely, the Dense connection-based Kernel Point Network (DenseKPNet), which can greatly expand the receptive field of kernel point convolution to extract rich semantic context information and valuable geometric features from the local region effectively. Specifically, we first design a multiscale convolution kernel point module to extract initial geometric features from coarse to fine. Then, we design the dense connection module to efficiently learn more expressive local geometric features while capturing rich contextual information. In addition, we propose the kernel point convolution attention module (KPCAM), which can capture global interdependencies between points and strengthen the discriminativeness of effective features. We evaluate our method on public indoor and outdoor datasets. The qualitative and quantitative experimental results show the effectiveness of DenseKPNet. The mIoU of the proposed method on S3DIS and semantic3D datasets can reach 68.9% and 77.9%, respectively.

Mass gathering events occur frequently in urban regions. Not only serious traffic jams, but also safety risks are consequently caused. Although many evacuation strategies have been proposed, the spatiotemporal coordination issue of multiple-modal transport tools is not solved well to deal with the efficiency and safety risk during the traditional evacuations. This study presented a large-scale multi-modal transport macro-optimization evacuation model for urban mass gathering events. By taking full advantages of each kind of public transit vehicles and optimizing their spatiotemporal cooperation with pedestrian evacuation, our models can greatly improve the global evacuation efficiency. Experiments on a realistic event was carried to validate the proposed model. The numerical results demonstrated that, under the premise of no increasing additional vehicle supply, the efficiency of proposed multi-modal transport evacuation can be improved by 49.7%, 46.5%, 118.7%, and 20.8%, compared with pure metro-, bus-, taxi- single-modal-transport evacuation, and simulated multi-modal transport evacuation without adequate optimization, respectively.

Deep learning methods are being increasingly used for urban traffic prediction where spatiotemporal traffic data is aggregated into sequentially organized matrices that are then fed into convolution-based residual neural networks. However, the widely known modifiable areal unit problem within such aggregation processes can lead to perturbations in the network inputs. This issue can significantly destabilize the feature embeddings and the predictions - rendering deep networks much less useful for the experts. This paper approaches this challenge by leveraging unit visualization techniques that enable the investigation of many-to-many relationships between dynamically varied multi-scalar aggregations of urban traffic data and neural network predictions. Through regular exchanges with a domain expert, we design and develop a visual analytics solution that integrates 1) a Bivariate Map equipped with an advanced bivariate colormap to simultaneously depict input traffic and prediction errors across space, 2) a Moran's I Scatterplot that provides local indicators of spatial association analysis, and 3) a Multi-scale Attribution View that arranges non-linear dot plots in a tree layout to promote model analysis and comparison across scales. We evaluate our approach through a series of case studies involving a real-world dataset of Shenzhen taxi trips, and through interviews with domain experts. We observe that geographical scale variations have important impact on prediction performances, and interactive visual exploration of dynamically varying inputs and outputs benefit experts in the development of deep traffic prediction models.

Online-to-offline (O2O) commerce connecting service providers and individuals to address daily human needs is quickly expanding. In particular, on-demand food, whereby food orders are placed online by customers and delivered by couriers, is becoming popular. This novel urban food application requires highly efficient and scalable real-time delivery services. However, it is difficult to recruit enough couriers and route them to facilitate such food ordering systems. This paper presents an online crowdsourced delivery (OCD) approach for on-demand food. Facilitated by Internet-of-Things and 3G/4G/5G technologies, public riders can be attracted to act as crowdsourced workers delivering food by means of shared bicycles or electric motorbikes. An online dynamic optimization framework comprising order collection, solution generation, and sequential delivery processes is presented. A hybrid metaheuristic solution process integrating the adaptive large neighborhood search and tabu search approaches is developed to assign food delivery tasks and generate high-quality delivery routes in a real-time manner. The crowdsourced riders are dynamically shared among different food providers. Simulated small-scale and real-world large-scale on-demand food delivery instances are used to evaluate the performance of the proposed approach. The results indicate that the presented crowdsourced food delivery approach outperforms traditional urban logistics. The developed hybrid optimization mechanism is able to produce high-quality crowdsourced delivery routes in less than 120 s. The results demonstrate that the presented OCD approach can facilitate city-scale on-demand food delivery.

Sampling fraction is crucial to sampling-related studies and applications, especially in the big data era when most data are neither originally designed nor controllable in the data collection process. A common concern among researchers is ‘what’s the modelling accuracy when using a sample?’. Taking intra-city human mobility as the study objective, this study utilizes a simple and direct method to analyse the influences of various sampling fractions on modelling accuracy. Five common intra-city human mobility indicators (travel distance, travel time, travel frequency, radius of gyration and movement entropy) are evaluated considering mean value, median and probability distribution. Experimental results demonstrate that the representativeness of each considered indicator converges to 1 in its own unique rate and variances. The minimum required sampling fractions to satisfy specific accuracies differ for various indicators and evaluation measures. To further investigate how related factors influence the modelling accuracy of sampling fractions, additional experiments are conducted considering multiple sampling methods, study scopes, and data sources. Several interesting general findings are observed. This study provides a reference for other sampling-based applications.

The emergency supplies in large-scale natural disaster is affected by the dynamics of transportation network. The travel speed and link capacity change over time because of traffic accidents, landslides, and uncertain weather, etc., which block up the processing of emergency material transmission sometimes. Finding out critical links and knowing how these links affect the transportation are benefit to improve dispatching performance. In this paper, the quickest network flow model is utilized to calculate the significance of transportation network links. Not only the critical links could be identified, but also how much decrease of traffic volume will be caused by the decrease in critical link capacity, or travel speed, or both. Different from conventional methods, the proposed significance in this paper is not only for individual link, but also for multi-links with considering time dimension. The experimental results demonstrate that this method is effective in dynamic transportation network and the computational efficiency is acceptable.

Maximum flow is one of the important and classical combinatorial optimization problems. However, the time complexity of sequential maximum flow algorithms remains high. In this paper, we present a two-stage distributed parallel algorithm (TSDPA) with message passing interface to improve the computational performance. The strategy of TSDPA has two stages, which push excess flows separately along cheap and expensive paths identified by a new distance estimate function. In TSDPA, stage 1 enhances the parallel efficiency by omitting high-cost paths and decentralizing calculations, and stage 2 guarantees the achievement of an optimal solution through divide-and-conquer method. The experimental test demonstrates that TSDPA runs 1.2–15.5 times faster than sequential algorithms and is faster than or almost as fast as the H_PRF and Q_PRF codes.

The traffic congestion changes over time in the real transportation network, and the vehicle navigation is a real-time processing. Both the accuracy and time performance of navigation algorithm are important. To respond rapidly to transportation network changes, this paper proposes two new dynamic shortest path algorithms based on the re-optimization method. The algorithms take full advantage of the previous search results, thus they take relatively few computing time to obtain the optimal solution. Experimental results demonstrate that the proposed algorithms can be applied to vehicle navigation in the transportation network with dynamic variation of the travel times on arcs and runs much faster than the traditional Dijkstra algorithm.