Guisong Yang

Accurate long-term traffic prediction in Low Earth Orbit (LEO) satellite networks is essential for proactive resource allocation and congestion avoidance, yet remains challenging due to highly dynamic topologies, intermittent connectivity, and scarce real traffic data. Existing approaches are largely limited to short-term prediction or assume static spatial dependencies, making them inadequate for non-stationary LEO environments. To address these challenges, this paper proposes DST-DEformer, a dynamic spatial–temporal Transformer framework that jointly models evolving inter-satellite topology and multi-scale temporal dependencies. Specifically, a topology-adaptive graph convolution module captures time-varying spatial correlations, while a dual temporal encoder decouples long-term global trend modeling from short-term local fluctuation learning. In addition, a hybrid simulation–calibration framework is developed to generate realistic satellite traffic by incorporating orbital dynamics, demographic information, and real-world traffic trends. Extensive experiments on simulated LEO satellite traffic and the PEMS08 benchmark show that DST-DEformer consistently outperforms state-of-the-art methods in long-term prediction, achieving 4%-13% reductions in MSE and MAE and significantly slower error accumulation as the prediction horizon increases. These results demonstrate the effectiveness and robustness of DST-DEformer for long-term traffic prediction under dynamic network topologies.

The rapid development of Low Earth Orbit (LEO) satellite networks has made in-orbit computing more feasible, offering a solution for processing real-time, diverse user tasks. Compared with traditional cloud computing in ground cloud computing center, directly computing on the LEO satellite can significantly reduce task-processing delay. However, challenges remain, including the limited sensing and computing capabilities of satellites, high delays in processing task requests, and frequent switching of control domains due to the relative movement between LEO satellites and nodes in other orbits. To address these challenges and improve task management, computing is treated as a Virtual Network Function (VNF), managed by Software-Defined Networking (SDN) controllers. This paper proposes a distributed SDN controller-based computing framework, where task information is forwarded to SDN controllers, which then use a task scheduling strategy to allocate tasks to suitable computing nodes for processing. To support the implementation of this framework, we first propose a heuristic SDN controller placement strategy that uses a tiling method to divide the LEO satellite network into SDN control domains and places the controller at the midpoint of each domain Then, we propose a Double Deep Q-Network (DDQN) algorithm for in-orbit task scheduling, which adaptively optimizes task scheduling strategy to minimize task-processing delay and ensure a high task completion rate. Finally, Simulations are conducted in two parts to evaluate the framework. The first part validates the DDQN-based task scheduling strategy, achieving significant reductions in task-processing delay and improved task completion rates compared to conventional strategies. The second part assesses the impact of SDN control domain shape and size on task-processing delay, confirming domain size as the dominant factor influencing delay.

To address the challenges of high uncertainty, inter-subject variability, and inefficiency multi-feature utilization in motor imagery electroencephalogram (MI-EEG) classification, this study proposes a multiview transfer Takagi-Sugeno-Kang (TSK) fuzzy classifier with soft variable embedded and discriminative structural preservation (MVT-TSK-SVDS). First, a transfer learning mechanism incorporating soft variable embedding in the consequent part is developed. This mechanism establishes cross-domain correlations via a shared consequent and representation matrix. Within this framework, soft variable embedding and low-rank constrained discriminative learning work in concert to effectively capture supervision information and cross-domain relationships. Second, a local-global structural preservation term incorporating graph embedding and low-rank constraint is implemented to maintain local discriminative information from source domain while integrating global geometric patterns across all data. Third, a multiview adaptive learning framework is designed to address feature representation diversity and information loss during knowledge transfer. MVT-TSK-SVDS dynamically optimizes view-specific contributions through entropy maximization criterion while ensuring collaborative decision via consistency constraints. Experimental results validate strong generalization between and across datasets. Our model achieves 62.16% and 72.71% accuracy in cross-subject tasks on BCI-IV 2a and OpenBMI, respectively. In cross-dataset evaluations, it attains 62.75% accuracy on BCI-IV 2a to OpenBMI and 65.08% accuracy on OpenBMI to BCI-IV 2a, respectively.

The classification of urban points of interest (POI) reflects the development of various industries in a city, making their distribution analysis significant. Traditional mapping methods often face inefficiency and high costs, leading to limited data quality and inaccuracies in classification. To address this, a low-cost, high-quality method is essential. Mobile Crowd Sensing (MCS) technology offers an innovative solution for identifying urban POIs. This paper introduces a hybrid MCS perception framework (MCS-APF) that includes a data collection module and a clustering module. The data collection module combines traditional participatory and opportunistic methods, incorporating a new recruitment criterion considering workers’ abilities, reputations, and POI popularity to enhance data quality. The clustering module employs an improved version of the Density-Based Spatial Clustering of Applications with Noise (DBSCAN-H) algorithm using Haversine distance, which effectively analyzes the combined data for accurate POI classification. Experimental results show that POI classifications derived from DBSCAN-H feature significant intra-cluster tightness and inter-cluster separation, outperforming traditional techniques. Overall, MCS-APF provides more accurate, efficient, and cost-effective POI sensing outcomes.

In multiplatform mobile crowdsensing (MCS), platforms attract mobile workers to participate in sensing tasks and collect data through incentive mechanisms to provide data-driven services. However, existing studies often focus exclusively on either competition or cooperation mechanisms between platforms, overlooking scenarios where both coexist. Additionally, most studies prioritize optimizing social welfare or market fairness, neglecting the core role of platforms as data service providers in a free market and the impact of their dominant market position, which limits the utility of platform benefit optimization. In addition, the issue of privacy protection has not received sufficient attention. To this end, this article proposes a two-stage hybrid mechanism (TS-HM) with the goal of maximizing the utility of the platform, including a decentralized pricing mechanism (DC-PM) and a two-substage cooperative matching mechanism (CMM). In the first stage, the DC-PM is used to help the platform learn the optimal pricing strategy under privacy-preserving conditions by modeling the pricing and worker contribution problem as a multileader–multifollower Stackelberg game; and in the second stage, the CMM is used to guarantee the matching stability and further enhance the platform utility. Experimental simulations demonstrate that the DC-PM effectively achieves rapid convergence of pricing strategies while ensuring privacy protection, and the CMM excels in matching stability and platform performance improvement. In general, the TS-HM outperforms existing approaches by increasing the total utility of the platform by an average of approximately 12.65%, significantly increasing the effectiveness of the platform and showcasing its strong advantages.

As the demand for computing in remote areas increases, low Earth orbit (LEO) satellites are transitioning from merely serving as communication roles to becoming providers of computing services. LEO satellite edge computing is being integrated into satellite networks, by placing servers on LEO satellites for task processing, to enable in-orbit computing. A key issue is how to optimally place servers in a constellation to support in-orbit computing of tasks effectively. This work proposes a LEO satellite server placement method that minimizes the average response latency of tasks processed on servers while reducing constellation construction costs by strategically selecting a subset of satellites in a constellation for server placement. First, we utilize real-world datasets from remote area devices and remote sensing satellites to derive the task distribution. To enable a unified representation of in-orbit computing tasks (ICT), we develop a unified model that standardizes task data from both sources. Second, we formulate the server placement process as an optimization problem that incorporates both spatial and temporal characteristics of the satellite network, aimed at minimizing the average response latency of ICT processing. To solve this problem, we design an improved genetic algorithm with simulated annealing (IGASA) to search for the effective server placement. Simulation results demonstrate that our method achieves superior performance in minimizing the average response latency of task processing compared to baseline server placement methods and converges rapidly.

Obstacle avoidance is the crux of formation control for a UAV swarm, no matter in urban or wild application environments. But there is a dilemma to promote both of obstacle avoidance flexibility and consistency in UAV formation control, which has not been solved effectively by existing researches. In view of this, this paper proposes a consensus based reinforcement learning method for differentiated formation control (DFC) of UAVs, to balance the obstacle avoidance flexibility and consistency. In this method, to promote obstacle avoidance flexibility of a UAV swarm, a consensus mechanism with differentiated formation control strategies is designed, it allows each UAV in a swarm changes its formation control strategy among aggregation, formation keeping and obstacle avoidance according to its local environment, and calculates its own current subgoal based on the selected strategy. Further, to improve the flight efficiency of the UAV swarm, a reinforcement learning model is provided to generate the optimal offset vector for each UAV according to its current subgoal. Moreover, to enhance the obstacle avoidance consistency of the UAV swarm, a collaborative obstacle avoidance algorithm is designed in the obstacle avoidance strategy, it requires UAVs to share their obstacle information and obstacle avoidance actions, and provides obstacle avoidance consensus rules to help UAVs to choose consistent obstacle avoidance directions. The experiment results show that the proposed method can combine obstacle avoidance flexibility and consistency of UAVs, thereby achieving higher flight efficiency and maintaining stable network connectivity.