One-shot federated clustering under local differential privacy (LDP) requires a server to recover global cluster structure from a single communication round of privatized client summaries. Existing centroid- or field-aggregation methods can become unreliable when privacy noise is strong and client supports are heterogeneous, because global mixing may suppress weak but informative local structures before they are detected. To address this structural suppression problem, this paper proposes an extract-before-mix principle and instantiates it as multi-domain topology-aware recovery (MDTA). MDTA first constructs private fields inside support domains, extracts local candidate peaks, aligns candidates across domains, and then performs K-aware fusion. We additionally report adaptive MDTA (A-MDTA) as an optional label-free post-processing selector used for diagnostic cross-regime branch selection rather than as a guaranteed safeguard. Under a metadata-public client-level LDP protocol, we provide privacy accounting for privatized local representatives and compactness-aware weights, and we analyze the conditional recovery behavior of field approximation, peak preservation, cross-domain alignment, and final compression. Experiments on synthetic, semi-real, and public datasets show that MDTA improves structural robustness in strong-privacy and support-heterogeneous regimes, while the optional A-MDTA analysis shows how label-free branch selection can be used without consuming additional privacy budget.

Heavy mining trucks are key equipment in open-pit haulage systems,where the available roadway space is often narrow in relation to the vehicle's size,resulting in extremely difficult driving.With the rapid advancement of mining intelligence,autonomous-driving technology has become an essential means of improving production efficiency to ensure operational safety and reduce operating costs.As a core component of autonomous-driving systems,path tracking control plays a decisive role in ensuring stable vehicle motion along a reference path.However,heavy mining trucks exhibit pronounced steering-mechanism constraints and significant signal transmission delays.Under the combined influence of sharp curves and long delays,path tracking systems tend to exhibit sluggish responses that rapidly increase tracking errors and even instability.Existing control methods struggle to simultaneously handle the c ompound effects of steering constraints and time delay,limiting their engineering applicability.To address the response lag caused by front-wheel steering-rate constraints in sharp-curve environments,a preview correct control(PCC)algorithm was developed by introducing the future heading of the reference path as preview information and incorporating the keypoint displacement error.The preview component improves steering proactiveness,while the correction component enhances responsiveness to current deviations to enable stable posture adjustments during curve entry,mid-curve,and exit.The PCC does not rely on complex models or high-performance computing platforms,making it suitable for the real-time operation of low-power onboard controllers.To address signal transmission delays in autonomous-driving systems,a multistep motion-compensation delay compensator is established by analyzing the PCC output structure and dynamic characteristics of a heavy mining truck to predict the vehicle's posture evolution during the delay interval and generate new control inputs that counteract the delay effects.By integrating the PCC with the delay compensator,a path tracking control system capable of simultaneously handling steering mechanism constraints and long delays was achieved for heavy mining trucks.Simulations were conducted under no-load and full-load conditions,followed by full-load field experiments.In no-load simulations at 20 km·h-1 on a U-shaped curve with a radius of 35 m,the PCC achieved a maximum displacement error of 0.0892 m,which is significantly more accurate than proportional-integral-derivative(PID)and preview PID and close to the nonlinear model predictive control(NMPC).Its average computation time was only 0.1514 ms,outperforming NMPC in terms of real-time capability.Under fully loaded conditions with a 0.4 s signal delay,the PCC combined with the delay compensator maintained the maximum displacement error within 0.1537 m,while the uncompensated PCC showed error divergence in sharp-curve sections.This demonstrates the critical role of the proposed compensation strategy in ensuring system stability under long-delay conditions.The compensator increased the average computation time by only 0.0982 ms,which had a negligible impact on real-time performance.Two sets of full-load field tests were conducted,with an actual signal delay of approximately 0.4 s.The maximum displacement errors were 0.1976 and 0.2073 m.In both tests,the vehicle navigated the sharp curve stably,without any loss of control or noticeable yaw deviations.Overall,the simulation and experimental results demonstrate that the proposed control system maintained a stable and reliable path tracking performance under significant steering-mechanism constraints and long signal delays,achieving a favorable balance between accuracy,real-time capability,and engineering deployability.Therefore,it is well suited for practical autonomous-driving applications in heavy mining trucks.

In certain emergency maneuver scenarios, such as high-speed lane changes or collision avoidance, the trajectory-tracking controller must guarantee strict vehicle stability and maintain high control accuracy to prevent safety hazards. The strongly coupled dynamics and pronounced nonlinearities of a vehicle pose significant challenges in achieving both objectives. However, the four-wheel independently driven or steered, distributed electric-drive intelligent vehicle chassis provides a versatile platform for active safety technologies. In addition, the inherent strengths of model predictive control (MPC) in handling linear, multi-objective constraints offer theoretical support for achieving high-precision stability control. The prediction horizon determines both the step length of MPC’s receding-horizon optimization and extent of the predicted future vehicle state space, such that a longer horizon enhances control smoothness, whereas a shorter horizon improves the vehicle’s dynamic responsiveness to path-curvature variations and mitigates the control-accuracy degradation caused by accumulated model-prediction errors. To date, discussions on adaptive prediction-horizon optimization in high-speed stability MPC trajectory tracking controllers remain scarce, making it difficult to strike an optimal balance between curvature-response speed and vehicle stability. To this end, this study builds upon an integrated vehicle stability and trajectory tracking control framework, to propose an adaptive prediction horizon nonlinear model predictive control (NMPC) strategy that incorporates previewed curvature information. By leveraging a preview-based reference path curvature point sequence, the control parameters are dynamically adjusted. The proposed method enhances the controller’s responsiveness to path curvature variations and mitigates the tracking accuracy degradation caused by accumulated errors in fixed-horizon strategies during high-curvature trajectory tracking. A state-coordination optimization mechanism designed via optimization sub-objective, explicitly couples the controller to the vehicle state of the previous control cycle. This effectively suppresses the decoupling effects in multistep optimization problems induced by prediction horizon variations and minimizes the discontinuities in control inputs. Finally, the proposed algorithm was validated in a co-simulation environment built using MATLAB/Simulink and CarSim. Representative high-speed maneuvering control scenarios were selected to quantitatively assess its performance. Comparative evaluations against other methods demonstrated the superiority of the proposed algorithm: in high-speed single lane-change scenarios. The method reduced average/peak lateral deviations by 36.17%/15.25%, average/peak longitudinal deviations by 11.55%/38.58%, and average/peak heading deviations by 6.13%/25.27% compared to fixed-horizon NMPC. In high-speed double lane-change scenarios, it achieved reductions of 30.28%/29.77% (lateral), 25.07%/3.85% (longitudinal), and 11.02%/32.68% (heading). Under high-speed low-adhesion conditions (μ=0.4), the method maintained robust precision and stability with peak lateral deviation of 0.2017 m, peak longitudinal deviation of 0.9744 km/h, peak heading deviation of 1.1936°, and peak centroid sideslip angle of 1.9074°. These quantitative metrics demonstrate that adaptive predictive horizon optimization, which leverages preview curvature information and state coordination, can further improve vehicle trajectory tracking accuracy while maintaining adequate stability margins.

The measurement of online service reputation based on ordinal preferences has been proposed to address the issue of unreliable reputation measurement results due to inconsistent user evaluation criteria. When users' complete ordinal preferences are unavailable, these methods ignore unknown preferences or use collaborative filtering to predict preferences without verifying the accuracy of preference prediction, leading to an untrustworthy service reputation. This study proposes an approach that models users' complete preferences using the conditional preference networks (CP-Nets) and then measures service reputation by aggregating CP-Nets. The approach designs an adaptive Tabu search algorithm to learn users' CP-Nets efficiently and aggregating all the CP-Nets using the ranked pairs method. The service reputation ranking is then deduced from the aggregated CP-Net. Experimental results on real datasets show that the proposed method is more efficient compared to existing methods, with more accurate preference prediction, and the reputation ranking is more consistent with user preferences.

Accurately detecting substation cabinet screw status is critical to the reliable operation of the power grid. However, collecting and annotating such data in a substation is potentially hazardous and time-consuming, making large-scale datasets difficult to obtain. Consequently, screw status detection under small-sample conditions remains a challenging and underexplored problem. Moreover, the substantial domain gap between natural scenes and power equipment imagery limits the effective transfer of pre-trained weights, leading to degraded performance of existing object detection methods. To address these issues, we propose Data Augmentation, Transfer Learning, and Adapter Tuning-based YOLO (DTA-YOLO), a method for screw status detection in small-sample scenarios. First, we apply data augmentation strategies to diversify the training data. Second, we design a novel adapter module, MSCA-Adapter, which integrates multi-scale convolution attention to facilitate efficient feature adaptation. Third, we implement a transfer learning approach that incorporates adapter tuning to improve detection performance. Experimental results demonstrate that DTA-YOLO achieves mAP@0.5 accuracies of 78.8 +/- 0.4%, 84.3 +/- 0.5%, and 88.7 +/- 0.8% on three small-sample datasets of varying sizes, demonstrating the effectiveness of the adapter-based fine-tuning paradigm under limited data conditions. This work provides a promising solution for intelligent power grid inspection tasks where data acquisition is constrained.

Federated learning (FL) is a decentralized machine learning paradigm designed to address data privacy concerns by exchanging model parameters instead of raw data. Hierarchical federated learning (HFL), a form of FL, enhances communication efficiency through its client-edge-cloud hierarchy. However, HFL is vulnerable to poisoning attacks where malicious clients may corrupt the global model by manipulating local model updates. Moreover, in HFL, data heterogeneity among clients and actions by malicious clients cause their model updates to deviate from benign clients, making it a challenge to identify malicious clients. In this work, we propose a Byzantine-robust method for HFL, named RHFL, which integrates both Jensen–Shannon divergence and performance-weighted aggregation to mitigate the impact of poisoning attacks. Specifically, the model update differences of clients are evaluated by calculating the Jensen–Shannon divergence score between each client and the edge server. Subsequently, malicious clients are detected by conducting an adaptive threshold mechanism that analyzes the statistical characteristics of Jensen–Shannon divergence scores across clients. Finally, a model performance-weighted aggregation rule is developed at the edge server to enhance the robustness of HFL. Due to its high communication and computational demands, particularly in handling large-scale data and real-time updates, RHFL requires high-performance computing environments for efficient operation. Extensive experiments on four benchmark datasets demonstrate that RHFL outperforms current defenses in terms of prediction accuracy.

Despite the remarkable success and widespread application of deep neural networks (DNNs), the aspect of fairness in DNNs often goes overlooked. In particular, the unfairness threats is more practical and urgent in federated learning (FL), which allows multiple clients to collaboratively train models and greatly facilitates attackers to manipulate models. Surprisingly, this palpable risk has yet to be fully acknowledged and explored in the research community. To bridge this gap, this paper introduces the Federated Unfairness Distribution Attack (FedUDA), a pioneering strategy designed to deliberately induce bias covertly and realistically into FL systems, which poses serious and significant threats in real worlds. FedUDA is achieved by the strategic manipulation of data distribution within a single malicious client, without sacrificing the accuracy of the model's outputs by leveraging gradient decomposition to isolate and inject bias-inducing updates orthogonal to the main learning objective. It improves durability by focusing on stable neural connections. Our experimental results demonstrate the effectiveness and real threat of the proposed method, and reveal for the first time the vulnerability of FL to unfairness attacks. Our research sheds light on the often-neglected unfairness risks in FL, emphasizing the critical need for the development of stronger safeguards to promote fairness in federated learning.

The federated learning, characterized by multi-party collaborative training and the submission of model updates rather than raw data, is vulnerable to free-rider attacks. Free-rider disguise themselves as benign clients by submitting fake updates to minimize their data contribution, while expecting to benefit from the well-trained global model. Numerous free-rider attack methods have been proposed, but they just simply utilize the parameters of the global model to create low-quality fake model updates and do not consider the evolution frequency of the attacker's model weights. To address these challenges, we propose a novel free-rider attack method based on data-free knowledge distillation. Specifically, we use the global and local models as a joint discriminator, along with a lightweight generative network, to generate pseudo-data. This pseudo-data serves as input for knowledge distillation between the global and local model to create the model updates. Furthermore, we design a MixUp-based local-adaptive data augmentation method to augment the generated pseudo-data. Our method effectively transfers knowledge from the global model to the attacker's fake updates, making them more deceptive. Moreover, the knowledge distillation process ensures a reasonable evolution frequency of the attacker's model weights. Extensive experiments demonstrate that our method achieves superior attack performance under robust aggregation defense models.

Deep reinforcement learning has shown potential in autonomous driving decision-making. However, vehicle decision-making involves complex information, and limited state information often limits the ability of agents to make optimal decisions. We present a novel on-ramp decision-making method using the SAC (Soft Actor-Critic) algorithm, which integrates the driving intentions of surrounding vehicles. Our model captures the vehicle characteristics of the target lane and its adjacent lanes as the state space. Additionally, we develop a hybrid action space that combines discrete lateral actions with continuous longitudinal actions, enabling the agent to adapt more effectively to intricate driving scenarios. The efficacy of our approach is validated through simulations using SUMO (Simulation of Urban MObility) and real-world road datasets. Comparative analysis of experimental results illustrates that our model surpasses alternative approaches in terms of collision rate and success rate. Moreover, the model exhibits a stable success rate under various road traffic density conditions.

Autonomous shoveling of loaders is the key technology to realize automatic and intelligent operation,and the tracking control of the target working trajectory is one of its core parts.The actual trajectory of the bucket in the pile is related to the indicators such as the operation output,so it is of great significance to realize the effective track-ing control of the target working trajectory.The PID and other control methods without system models have prob-lems such as large overshoot amplitude and buffeting under the system constraints.Since the Model Predictive Con-trol(MPC)has the advantage of effectively dealing with system constraints to make the system operate smoothly,it was introduced into the motion control of the loader's working mechanism and a trajectory-tracking control method was proposed for the working mechanism based on the Nonlinear Model Predictive Control(NMPC).A kinematic model of the working mechanism in the drive space was established.Then,the description of the working trajectory was given.Furthermore,a trajectory-tracking controller for the working mechanism was designed based on the NMPC method.Finally,the Simulink/ADAMS co-simulation was carried out with the general PID as the compari-son group.The analysis showed that under the same system constraints,for the different target trajectories,the maximum absolute error of the bucket-tip displacement based on the designed controller didn't exceed±0.052m,which was 71％lower than the PID controller,and the maximum absolute error of the bucket angle didn't exceed±2.58°,which was 16％lower than the PID controller.Moreover,the designed controller had a smoother control effect.The designed controller had better performance than the PID controller in dealing with system constraints and smoothness.

Enhancing the operational speed and control accuracy of autonomous transport vehicles is crucial for meeting the efficiency and safety demands in cargo transportation. Although Nonlinear Model Predictive Control (NMPC), based on vehicle dynamics models and multi-point look-ahead rolling optimization, offers high precision, it suffers from poor real-time performance, making it unsuitable for medium- to high-speed conditions. Compared to two-axle vehicles, multi-axle vehicles have more complex dynamics models and constraints, which increase the computational burden of NMPC. To address these issues, a neural network-based trajectory tracking controller for multi-axle vehicles under medium- to high-speed conditions has been proposed, using NMPC as the training sample generator. The learning samples were generated by NMPC based on the dynamics and multi-point look-ahead rolling optimization of multi-axle vehicles. Additionally, to prevent the failure of the network controller due to vehicle position information deviating from the sample space under the presence of positioning errors, a sample fusion method was employed to enhance the network controller's robustness to localization disturbances. The neural network controller was obtained through offline training and validated using a MATLAB/Simulink-TruckSim co-simulation platform, where it was compared with other controllers. The simulation results indicated that the control accuracy of the neural network controller is very close to that of NMPC, with a nearly twofold improvement in real-time performance.

Accurate object recognition in open-pit mine environments is crucial for the safety of autonomous transport vehicles. Existing autonomous driving perception mostly focuses on urban structured road traffic, and it is hard to adapt to the challenging open-pit mine environment. Lacking of datasets further limits the development of the specific work. In this paper, we propose an object detection dataset for open-pit mine autonomous driving applications. This dataset encompasses data from several mines and includes different periods such as day, dusk, and night. It provides detailed annotations for diverse objects in the open-pit mines and incorporates additional attributes for evaluating occlusion detection. In addition, to address the multi-scale changes of objects in open-pit mines and the occlusion problems caused by dust, we propose a novel occlusion mine object general distribution detection method, utilizing soft labels and vehicle attribute location to reduce the positioning ambiguity in difficult backgrounds and achieve specific object detection in harsh open-pit mine environments. Our work explores the benchmark for open-pit mine object recognition involving occlusion. Comparison with mainstream techniques on the benchmark demonstrates that our approach outperforms existing state-of-the-art methods and can achieve 82.2%, 81.7%, and 76.7% average precision in easy, moderate, and hard modes, respectively.

In the domain of autonomous driving, the dynamic coupling between lateral and longitudinal movements poses significant challenges for trajectory tracking control, often leading to sub-optimal control states and interference between lateral and longitudinal controls. This issue is especially pronounced in front-wheel drive (FWD) vehicles due to their specific configuration. An integrated control architecture effectively minimizes the interference between lateral and longitudinal controls. However, in trajectory tracking tasks requiring coordinated lateral and longitudinal control, changes in steering angle and wheel drive torque lead to variations in tire slip ratio and side slip angle, impacting the current vehicle state at the tire force level. Ignoring the coupling of tire forces between side slip and slip ratio, as well as the time-varying nature of tire parameters, will lead to mismatches in the predictive model. To address these issues, this study thoroughly considers the dynamic coupling and nonlinear time-varying characteristics of vehicle dynamics, and develops an augmented dynamic model capable of accurately describing the vehicle's state within a limited time domain. Furthermore, considering the limitations of Linear Model Predictive Control (L-MPC) in updating tire side slip and longitudinal stiffness within a limited time domain and the introduction of errors due to linearization, an integrated lateral-longitudinal trajectory tracking controller based on Nonlinear Model Predictive Control (NMPC) has been designed. The accuracy of the predictive model was initially validated using a Matlab/Simulink-CarSim simulation environment, followed by testing the controller using four representative driving scenarios. Simulation results demonstrate that the proposed trajectory tracking controller surpasses other methods in terms of tracking accuracy, robustness, and smoothness of tracking during speed and curvature changes.

Device-to-Device (D2D) collaborative offloading is a critical task offloading paradigm in mobile edge computing (MEC) environments, addressing the challenges of wasted idle device (ID) resources and the limited computational power of the edge server (ES). However, the constant mobility of devices leads to dynamic changes in the channel state and network topology. The process of offloading dependent tasks involves significant data transmission, and changes in the network environment can result in transmission failures that affect the entire task offloading process. This situation poses challenges to the effectiveness and stability of dependent task offloading. Therefore, we investigate the problem of dependent task offloading on multicore computing nodes in dynamic D2D environments. We formalize the problem as a mixed-integer nonlinear programming problem, which takes device mobility, task dependency, and user selfishness into account. To solve this problem, we propose a task offloading algorithm based on genetic algorithm (GA) and design a prioritization algorithm that matches the offloading decision to determine the execution order of subtasks. The results of simulation experiments on the Alibaba clustering dataset and synthetic dataset show that our algorithm significantly reduces the cost of task offloading compared to existing algorithms.

Local Differential Privacy (LDP) has garnered considerable attention in recent years because it does not rely on trusted third parties and has low interactivity and high operational efficiency. However, current LDP frequency estimation mechanisms aggregate data using different privacy budgets within the same domain of attribute values, overlooking the aggregation requirements across different domains of attribute values. This limits the potential for enhancing the data utility under fixed privacy budgets and meeting user preferences in multiple domains of attribute values and privacy budgets. To address this issue, we define a Multi-Domains Personalized Local Differential Privacy (MDPLDP) model that allows users to freely choose domains of attribute values and privacy budgets according to their privacy preferences. Furthermore, based on the MDPLDP model, two new frequency estimation mechanisms are proposed: MDPLDP-Generalized Randomized Response and MDPLDP-basic Randomized Aggregatable Privacy-Preserving Ordinal Response. These mechanisms support cross-domains data aggregation and optimize data utility by adjusting the domains of attribute values and increasing privacy budgets. Theoretical analysis reveals that these new mechanisms have lower estimation errors than the traditional LDP mechanisms. Experiments on real and synthetic datasets demonstrate that the proposed mechanisms effectively reduce estimation errors and enhance the utility of data-frequency estimation.

To address the performance degradation in model predictive control (MPC) under vehicle state uncertainties caused by external disturbances (e.g., crosswinds and tire cornering stiffness variations) and rigid constraint conflicts, we propose a robust MPC framework with adaptive weight adjustment and dynamic constraint relaxation. Traditional MPC methods often suffer from infeasibility or deteriorated tracking accuracies when handling model mismatches and disturbances. To overcome these limitations, three key innovations are introduced: a three-degree-of-freedom vehicle dynamic model integrated with recursive least squares-based online estimation of tire slip stiffness for real-time lateral force compensation; an adaptive weight adjustment mechanism that dynamically balances control energy consumption and tracking accuracy by tuning cost function weights based on real-time state errors; and a dynamic constraint relaxation strategy using slack variables with variable penalty terms to resolve infeasibility while suppressing excessive constraint violations. The proposed method is validated via ROS (noetic)–MATLAB2023 co-simulations under crosswind disturbances (0–3 m/s) and varying road conditions. The results show that the improved algorithm achieves a 13% faster response time (5.2 s vs. 6 s control cycles), a 15% higher minimum speed during cornering (2.98 m/s vs. 2.51 m/s), a 32% narrower lateral velocity fluctuation range ([−0.11, 0.22] m/s vs. [−0.19, 0.22] m/s), and reduced yaw rate oscillations ([−1.8, 2.8] rad/s vs. [−2.8, 2.5] rad/s) compared with a traditional fixed-weight MPC algorithm. These improvements lead to significant enhancements in trajectory tracking accuracy, dynamic response, and disturbance rejection, ensuring both safety and efficiency in autonomous vehicle control under complex uncertainties. The framework provides a practical solution for real-time applications in intelligent transportation systems.

To improve the trajectory prediction performance of human-driven vehicles in mixed traffic flow, we propose a novel interaction-aware network framework based on mixed teacher forcing GRU (Gate Recurrent Unit). Firstly, we filter and normalize the vehicle trajectory, divide it into three categories (left lane change, lane keeping, and right lane change), and build a trajectory prediction dataset. Then, we encode the historical trajectory of the target vehicle and the information about surrounding vehicles into the context vector. Next, we decode the content vector into future trajectory by mixed teaching force mode. Finally, the model is verified on the real main road datasets NGSIM US101 and I-80 and compared with the state-of-the-art model. The experimental results show that the proposed model achieves the state-of-the-art accuracy. The code can be accessed at https://github.com/ColinFanghz/MTF-GRU.git.

The integrated path tracking control (PTC) of steering and braking is crucial for enhancing the stability of autonomous vehicles under extreme conditions. In the present study, a control input dimensionality-reducing method and an asynchronous sampling method are presented to address the problem of the high computational cost of the steering and braking integrated path tracking controller (PTCer) based on model predictive control (MPC). First, based on tire friction limit and tire force utilization, the control input dimensionality-reducing method is designed and a vehicle model with reduced dimensionality of control input is derived. Second, the rolling iteration mechanism of MPC is analyzed and a variable-scale asynchronous sampling method between the control loop and prediction horizon is designed with the control horizon as the boundary. Finally, the integrated MPC-PTCer based on control input dimensionality-reducing and asynchronous sampling is designed. The real-time performance (RTP), path tracking accuracy, and vehicle stability of the proposed integrated MPC-PTC are tested and evaluated through the simulation and the hardware-in-the-loop platform. The test results of different test conditions show that the proposed integrated MPC-PTC improves the RTP by more than 70% and ensures the path tracking accuracy and lateral stability of autonomous vehicles under extreme conditions.

Aiming at the problem of integrated scheduling of machines and AGVs in a flexible job shop, this paper constructs a scheduling model with the optimization objectives of minimizing the maximum completion time, minimizing the machine load, and minimizing the total energy consumption. This model is based on a comprehensive consideration of the payload time and no-load time of AGVs between the loading and unloading stations and the machining machines. An improved NSGA-II algorithm is proposed to address this problem. The algorithm adopts a three-level coding structure based on processes, machines, and AGVs, and employs differentiated cross-variation strategies for different levels to enhance its global search capability. A variable domain search algorithm is introduced to boost the local search capability by combining different neighborhood search methods within the three-level coding structure. Additionally, reverse individuals are introduced to improve the elite retention strategy, thereby increasing the diversity of the population. Ultimately, the case test results demonstrate that the improved NSGA-II algorithm exhibits superior performance in solving the flexible job shop scheduling problem involving AGVs, and the effect of the number of AGVs on the scheduling objectives conforms to the law of diminishing marginal utility.