To address the challenges of limited resilience and high operational costs in current expressway power supply systems under extreme conditions, a chain active distributed network model featuring dynamic SOPs is introduced. This model leverages flexible interactions among network nodes to facilitate optimal power flow distribution and energy dispatch. An uncertain dataset, based on the Wasserstein distance, is constructed to account for the diverse characteristics of power sources and loads on expressways. Using this dataset, a “Traffic-Electrical” coupling model for typical power scenarios is developed to identify the optimal SOP installation locations under improved flow capturing location model constraints, thus forming an active network with SOPs. A two-stage optimization model targeting both the “robustness” and “economy” of the power supply system is designed with the support of a novel “NSGA-II-PSO” heuristic algorithm. Simulations of the “Taohuaping-Dingyuan” segment of the G30 “Lian-Huo Expressway” demonstrate the model’s effectiveness, with a 9.74% reduction in average voltage offset and a 3.78% decrease in dispatch costs compared to traditional networks, confirming its robustness, flexibility, and cost-effectiveness.

This paper presents a force-sensorless collision detection approach based on virtual contacts. A series of indices based on virtual instantaneous powers are introduced associating to several virtual contacts defined on corresponding links of a manipulator. A possible contact link can be determined by comparing the introduced indices without any force or tactile sensors. Simulation results of 2D collision detection tasks by using a three DOF (Degree of Freedom) planar manipulator and 3D collision detection tasks by using a six DOF spatial manipulator validate the effectiveness of the proposed collision detection approach.

Pedestrian detection as a key problem of computer vision has many comprehensive applications such as smart transportation, video surveillance and self-driving car. In this work, we explore the application of Faster R-CNN for pedestrian detection. We found that Faster R-CNN has not shown competitive results as we expected. The main challenges for Faster R-CNN may be small objects and hard negative examples. More interestingly, we found that when the RPN and the proposal classifier are trained separately and simply combined, the detection results can be significantly improved. Based on these observations, we designed a multi-map network and a multi-ROI network to solve these two problems. The multi-map network not only utilizes the convolutional feature maps output by last convolutional layer but also exploits the previous feature maps which have higher resolution but less semantic information. The multi-ROI network converts one bounding box to three bounding boxes including enlarged one, shrinking one and original one. Then, the network fuses the features of these three bounding boxes into one robust and discriminative feature. In addition to comparing these two networks with Faster R-CNN, we also merged them into one unified and end to end network called multi-ROI-map (MRM) network. Finally, we trained the RPN and MRM separately and fused their detection results. Overall, our method achieves a competitive result compared with some leading methods on the Caltech dataset and executes 2 times faster than these methods.