The fuzzy flexible job shop scheduling problem (FFJSS) is a challenging NP-hard problem due to the combined effects of operation sequencing, flexible machine assignment, and fuzzy processing times. Existing multi-objective evolutionary algorithms mainly emphasize convergence and diversity in the objective space, while often overlooking the structure in the decision space induced by machine flexibility. This paper presents a multiform evolutionary framework for multi-objective FFJSS by co-evolving the original task and an auxiliary task with restricted machine choices. The auxiliary task reduces decision-space complexity and encourages stable scheduling structures, which are implicitly transferred to guide the search on the original task. Experimental results on benchmark instances demonstrate improved convergence and solution quality over representative algorithms. Further analysis shows that high-quality nondominated solutions tend to assign most operations to the most efficient machines, supporting the effectiveness of the proposed multiform design.

The deployment of unmanned aerial vehicle (UAV) swarms for collaborative intelligence in urban environments presents unprecedented opportunities for smart city applications, yet faces critical security challenges from Byzantine attacks in distributed learning scenarios. This paper presents a comprehensive framework that jointly addresses Byzantine-resilient federated learning and trust-aware task scheduling for heterogeneous UAV fleets operating under adversarial conditions. We propose a novel aggregation mechanism that employs multi-metric anomaly detection, combining gradient norm analysis, statistical residual examination, and directional alignment assessment to identify and mitigate malicious model updates. The framework incorporates a dynamic trust management system that maintains behavioral profiles of participating agents, enabling adaptive isolation of compromised nodes while preserving system performance. Furthermore, we develop a security-conscious task scheduler that considers real-time trust scores, resource constraints, and environmental threat indicators to optimize mission allocation across the swarm. Our approach uniquely couples learning objectives with operational requirements, allowing UAVs to adapt their training parameters based on perceived threat levels. Theoretical analysis establishes convergence guarantees under Byzantine presence and demonstrates trust score separation between honest and malicious agents. The proposed framework provides a practical solution for deploying secure collaborative learning in mission-critical UAV applications.

The deployment of unmanned aerial vehicles (UAVs) as dynamic infrastructure nodes in intelligent transportation systems requires simultaneous consideration of spatiotemporal traffic dynamics, complex urban airspace constraints, and stringent safety requirements. This paper presents a hierarchical learning framework that integrates physics-informed neural networks (PINNs) with adaptive control policies through differentiable safety layers. We develop a multi-level architecture where macroscopic traffic flow dynamics are captured via Lighthill-Whitham-Richards (LWR) PINNs, geometric and environmental constraints are encoded through Eikonal equation solvers, and deployment decisions are made by liquid neural network (LNN) policies. To guarantee safety during both discrete site selection and continuous navigation, we introduce a differentiable control barrier function (CBF) layer that enforces hard constraints through convex quadratic programming while maintaining end-to-end differentiability. The framework addresses uncertainty in learned physics models through risk-aware formulations incorporating distribution-free chance constraints via concentration inequalities and conditional value-at-risk (CVaR) penalties. We establish theoretical guarantees including forward invariance under CBF constraints, probabilistic safety bounds, and convergence to stationary points under joint physics-policy training. Comprehensive experiments on multi-region urban scenarios demonstrate that the integrated approach significantly outperforms both heuristic baselines and ablated variants, with the physics-informed features proving essential for achieving high deployment accuracy while maintaining safety. The modular architecture enables real-time operation and adaptation to diverse urban environments, offering a principled pathway toward autonomous aerial infrastructure management.

The Internet of Vehicles (IoV) provides a deployment environment for connected consumer electronics networks, including in-cabin assistants, telematics devices, and roadside electronics that support cooperative perception. Training or adapting generative models in this setting is constrained by private local data, limited energy budgets, tight communication links, and mobility-induced intermittency. This paper proposes EAFedGen, an energy-constrained federated learning framework that combines a shared generative backbone with per-device low-rank adapters and a dual controller for adapter rank and update sparsity. We formulate the training problem as a bilevel optimization with device-level energy constraints and analyze adapter stability, inexact inner adaptation, convergence under compression and optional differential privacy noise, and training behavior under participation and staleness effects. Experiments on V2X-Sim 2.0 BEV maps, V2X-Sim v1.0-mini BEV occupancy, and V2X-Sim 2.0-mini LiDAR BEV occupancy treat BEV completion as conditional generation. The experimental results characterize the communication-quality operating region induced by personalized adapters and dual energy control, and they show that attention-weighted latent fusion is the strongest learned cooperative strategy among the tested fusion schemes.

Effective urban traffic signal control in large-scale networks remains challenging due to complex interdependencies among intersections and unpredictable fluctuations in traffic conditions. To address these challenges, this paper proposes a novel Adaptive Hybrid Multi-Objective Optimization Algorithm with Reinforcement Learning (AHMOA-RL) for robust and scalable traffic signal management. The core innovation of AHMOA-RL lies in a hierarchical optimization framework that efficiently decomposes the problem into global region-level coordination and local intersection-level refinements, significantly reducing computational complexity while ensuring synchronized control across extensive urban networks. A Q-learning agent dynamically selects among multiple evolutionary operators-Genetic Algorithm, Differential Evolution, Particle Swarm Optimization, and Local Search-to strategically balance exploration and exploitation during optimization. Additionally, a memory-based evaluation mechanism leveraging historical data is integrated to smooth transient traffic anomalies and provide stable performance estimates. Extensive simulations on large-scale city networks inspired by Manhattan, Paris, S & atilde;o Paulo, and Istanbul demonstrate that AHMOA-RL consistently outperforms state-of-the-art methods, achieving substantial reductions in average vehicle delays, improved network stability, and enhanced robustness under diverse traffic conditions. The algorithm's compact Pareto fronts and superior convergence characteristics validate its effectiveness for practical deployment in complex urban environments.

Large-scale traffic control must remain effective under demand uncertainty and episodic disruptions. This paper presents a scenario-based risk-sensitive framework that adjusts grouped capacity multipliers to reduce tail travel time. The method couples static traffic assignment with a mean-CVaR objective and an L1 adjustment budget, solved via differential evolution. Because repeated exact user-equilibrium (UE) solves are costly at scale, we evaluate performance with incremental all-or-nothing assignment and validate ranking stability on Anaheim and Chicago-Sketch using a method-of-successive-averages (MSA) UE approximation. Experiments on Chicago-Sketch, Anaheim, and Chicago-Regional networks from the TNTP repository show consistent reductions in mean and tail travel time for CVaR-robust solutions, while a mean-only baseline can worsen worst-case outcomes. We report bootstrap confidence intervals, scenario-count and OD-coverage sensitivities, out-of-sample repeats, and domain-shift tests (lognormal, heavy-tailed, correlated, and extreme-demand scenarios), providing evidence that the relative improvements persist across uncertainty models and evaluation settings.

The digital twin, which integrates multidisciplinary attributes to achieve interactions between the virtual and the real worlds, has found numerous applications in the field of smart cities and transportation infrastructure. With the acceleration of urbanization, rail transit, as a key component of urban infrastructure, faces increasing challenges in maintenance and management. Traditional reactive maintenance methods can no longer meet the growing volume of user demands. This study presents an architecture of intelligent operation and maintenance platform for the rail transit infrastructure based on WebGIS, paving the way for a digital-twin solution driven by predictive maintenance. We present a case study of the Shanghai Metro Long Yang Road-Zhang Jiang High Tech experimental line, which demonstrates the significant advantages of our platform in real-time data integration, 3D visualization, and intelligent decision support. Dynamic simulation integrating finite element and multibody dynamics software further validate the effectiveness of digital twin technology in the identification and prediction of rail fastener damage.

Vehicular edge-computing (VEC) systems face unprecedented challenges in emergency scenarios, where mission-critical applications demand simultaneous optimization of latency, reliability, and energy consumption under rapidly evolving conditions. These systems are fed by dense vehicular IoT telemetry, including onboard perception sensors, roadside units, and uncrewed aerial vehicle (UAV) relays whose data streams must be scheduled in real time. Existing schedulers either rely on heuristic methods that lack theoretical guarantees or employ computationally intensive optimization that cannot meet real-time constraints. This article presents a novel bilevel scheduling framework that synergistically combines hierarchical decision-making with projected convex optimization to achieve both computational efficiency and provable performance bounds. The proposed hierarchical+projected (HP) scheduler operates through two complementary layers: a lower level mission-aware hierarchical scorer that generates warm-start allocations by evaluating vehicle-tier pairs using forecast costs, queue states, and group coordination factors; and an upper level projected subgradient coordinator that iteratively refines these allocations to enforce tier reservations and future commitment constraints derived from workload predictions. We establish theoretical foundations including convergence rates of O(1/root K) for the projected layer, bounded reservation violations decreasing exponentially with iteration count, and computational complexity of O(KNT\log T) enabling submillisecond execution. The framework integrates seamlessly with forecast-driven orchestration pipelines, incorporating mobility, link quality, and workload predictions, while maintaining safety fallback mechanisms for degraded conditions. Comprehensive evaluation across large-scale emergency traces, calibrated disaster scenarios, and urban mobility benchmarks demonstrates that HP substantially reduces reservation violations compared with pure heuristic methods, improves the system reliability over baseline particle swarm optimization, and significantly expands the Pareto hypervolume (HV) while maintaining the real-time decision latency. Comparative analysis against ten state-of-the-art approaches-including adaptive particle swarm optimization (PSO), convex alternating direction method of multiplier (ADMM), NSGA-II/-III, MOEA/D, deep reinforcement learning (RL), and theoretical hierarchical variants-reveals that HP uniquely balances real-time responsiveness with constraint satisfaction, achieving near-zero reservation gaps while preserving allocation diversity across heterogeneous mission profiles.

To alleviate computational load on RSUs and cloud platforms, reduce communication bandwidth requirements, and provide a more stable vehicular network service, this paper proposes an optimized pinning control approach for heterogeneous multi-network vehicular ad-hoc networks (VANETs). Within these VANETs, vehicles participate in multiple task-specific networks with asymmetric coupling and dynamic topologies. We first establish a rigorous theoretical foundation by proving the stability of pinning control strategies under both single and multi-network conditions, deriving sufficient stability conditions using Lyapunov theory and linear matrix inequalities (LMIs). Building on this theoretical groundwork, we propose an adaptive genetic algorithm tailored to select optimal pinning nodes, effectively balancing LMI constraints while prioritizing overlapping nodes to enhance control efficiency. Comparative analysis with alternative optimization methods demonstrates that our approach outperforms random search, high-degree selection, and greedy methods by achieving significant reduction in required control nodes while maintaining superior control performance. Extensive simulations across various network scales demonstrate that our approach achieves rapid consensus with a reduced number of control nodes, particularly when leveraging network overlaps. This work provides a comprehensive solution for efficient control node selection in complex vehicular networks, offering practical implications for deploying large-scale intelligent transportation systems.

We address multi-objective unmanned aerial vehicle (UAV) placement for motorway intelligent transportation systems, where deployments must balance coverage, link quality, and UAV count under geometric constraints. We construct a reproducible benchmark from highD motorway recordings with recording-level splits and generate Pareto-optimal labels via NSGA-II. A preference rule yields deployable targets while preserving multi-objective evaluation. We train fast surrogate models that map unordered vehicle positions to UAV count and continuous placements, using permutation-aware losses and constraint-regularized training across set-based and sequence-based architectures. The evaluation protocol combines Pareto quality metrics, success-rate curves, runtime benchmarks, and robustness studies, with uncertainty quantified by recording-level bootstrap. Results indicate that permutation-invariant set models provide the strongest coverage–SNR–count trade-off among learned predictors and approach NSGA-II quality while enabling real-time inference. Under shared budgets, they offer a more favorable success–latency trade-off than heuristic baselines. The benchmark, splits are released to support reproducible ITS deployment studies and to facilitate comparisons under shared operational budgets.

For autonomous systems influenced by multiple complex factors, generating test scenarios under these factors that comprehensively cover the system’s performance boundary is critical for thorough system testing. However, generating boundary test scenarios that are comprehensive (covering the entire performance boundary), accurate (concentrated near the performance boundary), and fast remains a significant challenge for systems with high-dimensional state spaces. To address this, we propose SIGIS, a support information guided and incremental SVM based boundary scenario generation algorithm. SIGIS iteratively couples a surrogate model with adaptive sampling. The surrogate model employs an SVM that incorporates a dynamic training scenario screening strategy and an incremental update strategy to balance between accuracy and efficiency when fitting large-scale, high-dimensional data. For adaptive sampling, it fully leverages the boundary information provided by the SVM and proposes a three-stage hybrid sampling strategy to rapidly explore and cover the performance boundary. A systematic evaluation platform composed of eight benchmarks, a path planning system, and six quantitative metrics is established. Extensive experiments demonstrate that SIGIS significantly outperforms six state-of-the-art comparison algorithms in coverage and accuracy while maintaining competitive efficiency. Furthermore, we show that the comprehensive boundary scenarios generated by SIGIS enable meaningful quantitative robustness comparisons between different autonomous systems, validating their practical utility for autonomous system safety assessment.

The difficulty of complex scheduling problems is significantly affected by the increase in task scale. A multi-agent deep reinforcement learning-guided adaptive construction and search (MADRL-ACS) algorithm is proposed to solve agile Earth observation satellite scheduling problems (AEOSSP). The MADRL-ACS algorithm, a combination of ensemble learning and data analytics strategies, was employed to address the challenges of complex, time-varying task constraints, multi-objective conflict optimization, and insufficient computational efficiency in large-scale AEOSSP. The multi-core computational resources are used in the ensemble learning approach by integrating classical heuristic algorithms, namely hill climbing (HC), Tabu search (TS), and simulated annealing (SA), with deep reinforcement learning (DRL). A multi-agent framework is adopted in the ensemble learning approach, where each agent employs a dueling deep Q-network (DQN) architecture to select local search strategies based on real-time state information and to execute them intelligently within each parallel thread. Density-based spatial clustering of applications with noise (DBSCAN) is used in the data analytics strategy to guide frequent pattern mining, whereby the elite solution set is periodically analyzed for density and dynamically partitioned into clusters. Frequent patterns are mined from representative solutions to construct new candidates, thereby enhancing global search capability and effectively preventing premature convergence. Experimental results demonstrate that MADRL-ACS outperforms the other compared algorithms in terms of both solution quality and computational efficiency, and exhibits strong robustness and practical applicability in large-scale, complex scheduling scenarios.

Fault diagnosis in Internet of Vehicles must handle rare faults, missing telemetry, changing class priors and asymmetric maintenance risk. A model with strong ranking quality can still make poor maintenance decisions when its probabilities and operating thresholds mismatch deployment cost. This article develops a probabilistic attention method for robust risk minimization. Each telemetry token is encoded by a Gaussian mean and variance pair, and attention scores are reduced according to a closed form distance between token distributions. A differentiable threshold layer maps deployment prevalence, cost ratio, and sample missingness to a decision threshold, while training minimizes the largest soft risk over a bounded deployment context set. Theoretical analysis gives complete bounds for attention logit stability, asymmetric cost regret, soft threshold approximation, prior change threshold transfer, and vertex reduction of the robust objective. Air pressure system (APS) experiments show targeted cost improvement against matched probabilistic variants and interpretable threshold adaptation. A controller area network (CAN) study further illustrates the gap between ranking quality and operating cost in temporal fault windows.

Urban Vehicle-to-Everything (V2X) services demand infrastructure plans that remain effective under pronounced spatiotemporal demand shifts and operational uncertainty. Existing studies often optimize roadside unit (RSU) deployment or unmanned aerial vehicle (UAV) operations in isolation, rely on scalar objectives, and simplify feasibility constraints, which limits their value for actionable planning. This paper develops a hierarchical multi-objective planning framework that jointly decides (i) where and how many RSUs to deploy and (ii) how to schedule a UAV fleet over time to complement the static backbone. The upper level searches for non-dominated infrastructure configurations that balance total cost, spatiotemporal coverage, and robustness, while the lower level evaluates each configuration through a state-aware greedy dispatch policy that explicitly models UAV operational modes and battery/charging dynamics. Robustness is optimized via a weighted composite of redundancy, coverage variability, and mission balance, and scenario-based Conditional Value-at-Risk (CVaR) is reported as a secondary tail-risk indicator under demand perturbations and component failures. Experiments on realistic urban traffic data, together with policy validation on reduced instances, baseline comparisons, and multi-seed stability analysis, demonstrate that the framework yields diverse and interpretable planning trade-offs.

Vehicular Ad-hoc Networks (VANETs) encounter significant communication challenges in highly dynamic highway environments due to rapid topology changes, varying vehicle densities, and unpredictable channel conditions. To address these challenges, this paper proposes a multi-objective robust optimization framework specifically designed to optimize communication delay, load balance, link quality, and temporal stability of dynamic multi-hop vehicular communication networks on highways. The proposed framework explicitly incorporates temporal continuity constraints to ensure stable and persistent communication paths despite frequent network changes. Moreover, a robust optimization model is formulated to mitigate performance deterioration arising from uncertainties in highway vehicle densities and channel fluctuations. An improved Non-dominated Sorting Genetic Algorithm II (NSGA-II) is developed to solve the multi-objective optimization problem efficiently, achieving optimal trade-offs among conflicting objectives. Extensive simulation experiments using realistic highway vehicle trajectories demonstrate that the proposed method achieves balanced performance across multiple objectives. Compared to traditional routing protocols, the framework provides substantial improvements in load balancing and link quality, while maintaining superior path stability through temporal continuity constraints. The framework demonstrates robust performance under various traffic scenarios and maintains effectiveness despite GPS positioning errors and network partitions. The results validate the framework's applicability and effectiveness in real-world highway-based intelligent transportation systems, offering a comprehensive solution that addresses the multi-faceted challenges of vehicular communications.

Balancing convergence and diversity is a crucial challenge for particle swarm optimization in addressing large-scale optimization problems. A main reason is that it is difficult to simultaneously effectively manage diversity with specific parameters in both objective space and decision space. To address this issue, this paper introduces a novel velocity update structure that integrates diversity preservation in both objective space and decision space alongside convergence. Aligned with the proposed update structure, a new diversity enhancement mechanism is proposed. This mechanism comprises an entropy-based diversity preservation strategy and an adaptive difference-mutation-based diversity preservation strategy, designed to preserve diversity in objective space and decision space, respectively. By utilizing a dynamic convergence learning strategy, a novel large-scale swarm optimizer capable of explicitly and simultaneously balancing convergence and diversity in both spaces is developed. This paper theoretically proves the stability and analyzes the search behavior of the proposed algorithm. Comprehensive experiments are then conducted using two large-scale benchmark test suites, a real-case application model and several state-of-the-art algorithms. The results demonstrate the competitiveness of the proposed algorithm in large-scale optimization and the effectiveness of the proposed strategies in balancing convergence and diversity.

The significance of transportation efficiency, safety, and related services continues to increase in urban vehicular networks. Within such networks, roadside units (RSUs) serve as intermediaries in facilitating communication. Therefore, the deployment of RSUs is of utmost importance in ensuring the quality of communication services. However, the optimization objectives, such as time delay and deployment cost, are commonly developed from diverse perspectives. As a result, it is possible that conflicts may arise among the objectives. Furthermore, in urban environments, the presence of various obstacles, such as buildings, gardens, lakes, and other infrastructure, poses challenges for the deployment of RSUs. Consequently, the deployment encounters significant difficulties due to the existence of multiple objectives, constraints imposed by obstacles, and the need to explore a large-scale optimization space. To address this issue, two versions of multi-objective optimization algorithms are proposed in this paper. By utilizing a multi-population strategy and an adaptive exploration technique, the proposed methods efficiently explore a large-scale decision-variable space. In order to mitigate the issue of an overcrowded deployment of RSUs, a calibrating mechanism is adopted to adjust RSU density during the optimization procedures. The proposed methods also address data offloading between vehicles and RSUs by setting up an iterative best response sequence game (IBRSG). Comparative analyses against several state-of-the-art algorithms demonstrate that our strategies achieve superior performance in both high-density and low-density urban scenarios. The results indicate that the proposed solutions significantly enhance the efficiency of vehicular networks.

The escalating need for efficient biodiversity monitoring motivates this investigation into lightweight deep learning solutions for automated fish species classification. While conventional convolutional neural networks achieve notable accuracy, their computational complexity hinders deployment in resource-limited ecological monitoring scenarios. To address this challenge, this paper presents a neural architecture search-driven framework utilizing differentiable architecture search (DARTS) to automatically design compact models optimized for edge devices. The proposed methodology systematically explores optimal operator combinations with DARTS, such as the size of the convolution kernel and the kinds of pooling operators, etc. Furthermore, the proposed framework incorporates data augmentation strategies to enhance generalization across degraded field images. In the experiment, our derived Model-D achieves 44.62

Unmanned aerial vehicles (UAVs) are increasingly deployed in dynamic environments for applications such as surveillance, delivery, and data collection, where efficient task allocation and path planning are critical to minimizing mission completion time while managing limited energy resources. This paper proposes a novel approach that integrates energy management into a rolling horizon framework for dynamic UAV task allocation and path planning. We introduce an enhanced Particle Swarm Optimization (PSO) algorithm, incorporating adaptive perturbation strategies and a local search mechanism based on simulated annealing, to optimize UAV task assignments and routes. The rolling horizon framework enables the system to adapt to evolving task demands over time. Energy consumption is explicitly modeled, accounting for flight, computation, and recharging at designated stations, ensuring practical applicability. Extensive simulations demonstrate that the proposed method reduces the mission makespan significantly compared to conventional static planning approaches, while effectively balancing energy usage and recharging requirements. These results highlight the potential of our approach for real-world UAV operations in dynamic settings.

For different types of complex terrain and scenes in natural environments, we could use some deep learning models such as YOLOv8, Deeplabv3+ to identify scenes and objects and separate them from background images. The results of the two methods are different, and each method has advantages and limitations. DeepLabv3+ excelled at segmenting large homogenous areas such as mud (88.61