Information granularity provides a framework for machine intelligence to simulate human cognitive processes in problem-solving, enabling machines to make more flexible and adaptive decisions in complex data environments. However, most existing information granularities are based on fuzzy c-mean (FCM) or K-means clustering for granularity construction. This kind of convex data structure clustering has difficulty in recognizing some irregular nonconvex data types, and can only be filled irregularly by increasing the cluster classes, which is easy to lose the wholeness and accuracy of the data. The density-based spatial clustering of applications with noise (DBSCAN), a classical density-based spatial clustering method, is well-suited for both convex and nonconvex datasets. Building on the versatility of DBSCAN, a new density-based clustering method called stacked density granulation (SDG) is proposed, which constructs information granularities capable of effectively describing both convex and nonconvex data. This approach addresses the limitations of traditional information granularity. By applying the newly constructed information granularities to the feature space, the importance of features can be measured. To enhance this process, a heuristic feature selection method called density granular feature selection (DGFS) is introduced. DGFS constructs a low-dimensional feature space by aggregating and discretizing the information granularities, retaining only the most relevant features. To demonstrate the superiority and effectiveness of the DGFS algorithm, 12 publicly available datasets are utilized, and its performance is compared with other feature selection methods across four different classifiers. The experimental results and the statistical significance test indicate that DGFS consistently outperforms the competing methods.
Localized feature selection (LFS) partitions the sample space into multiple local regions and selects superior feature subsets per region. However, existing LFS methods treat each region separately when selecting features, ignoring the overlaps and correlations among neighboring regions. This approach limits classification performance in high-dimensional (HD) scenarios. To address this limitation, we model LFS as a many-task optimization problem, where each local region is treated as a separate multiobjective feature selection (FS) task while considering regional correlations. Generally, we propose a novel evolutionary many-task optimization-based LFS (EMaTO-LFS) framework (EMaTO-LFS), which enables collaborative solving of HD FS tasks by sharing knowledge among correlated regions. Specifically, we use a method that utilizes the ratio of positive and negative samples to adaptively construct local regions, while a filter-based method prefilters feature subsets for each local task in advance. We also design a subset-based mutation operator, improving the method's ability to search for superior feature combinations. In addition, a strategy for transferring knowledge between local regions is formulated based on neighborhood relationships, effectively avoiding negative transfer due to conflicting tasks. Empirical results on 14 HD datasets show that evolutionary many-task optimization (EmaTO)-LFS achieves competitive balanced accuracy with smaller feature subsets than state-of-the-art FS and LFS methods.
Detecting small objects in aerial images is significantly challenging due to their nonuniform distribution and severe scale variations resulting from changing view angles. Because autonomous aerial vehicles have limited computational power, balancing detection accuracy and efficiency remains a challenging problem. Existing methods, e.g., feature pyramid network (FPN)-based algorithms, concentrate on fusing deep low-resolution features with shallow high-resolution features and primarily rely on simple stacking and channel fusion. However, features of small objects are easily affected by unpredictable noise from the background, leading to high computation cost during feature fusing. This work tackles the issue by designing a novel one-step generative small object detection (SOD) framework. It leverages the self-consistency property provided by a consistency model, which enables the proposed model to convert random Gaussian noise to a single-scale output, thereby enabling the "one-step" inference. We formulate an SOD task as a noise-to-box procedure. We then apply a consistency model to initialize the diffusion process with Gaussian noisy bounding boxes derived from their corresponding ground-truth (GT) annotations. We next introduce a denoising sampling strategy to classify and locate small objects by iteratively refining their Gaussian distributions. We finally comprehensively evaluate our proposed framework on several SOD benchmarks for autonomous aerial vehicles, including DOTA, VisDrone, and AAVDT. Experimental results corroborate that it outperforms the state-of-the-art method (DiffusionDet) by up to 5.1% in terms of $AP_{S}$ (average precision on small objects) on DOTA. Code is available at https://github.com/BrainPotter/CEOSOD.
This article addresses the formation control of under-actuated multiple autonomous surface vehicles (MASVs) with input quantization under communication delay conditions, which is influenced by external marine disturbances and internal model uncertainties. A two-level distributed guidance and quantization control architecture based on the Nussbaum function is proposed. At the communication level, a time-delay distributed event-triggered extended state observer (ESO) is introduced to estimate the state of the single virtual leader, thereby further conserving communication resources. At the control level, the distributed formation guidance laws based on ESO are proposed in the kinematic subsystem, enabling effective tracking of the ideal trajectory while estimating the states of neighboring agents and unknown ocean disturbances. In the dynamics subsystem, a fuzzy logic system is used to estimate the uncertain terms within the model, and a linear model is introduced to handle the input quantization process. Additionally, the fuzzy adaptive quantization tracking control laws based on the Nussbaum function are proposed to achieve accurate tracking of the guidance signals and reduce actuator execution frequency, which makes the proposed scheme more applicable to practical marine engineering scenarios. The stability of the designed control structure is proven based on stability theory, and all signals within the closed-loop control system are uniformly ultimately bounded. Simulation experiments validate the rationality and effectiveness of the proposed method.
It is of great significance to obtain road preview information using sensors to improve the ride comfort of intelligent vehicles. The explicit model predictive control (EMPC) algorithm offers advantages under a single road condition by integrating the road elevation with suspension state. However, EMPC requires adjusting the weighting coefficient to achieve optimal control effectiveness across varying road conditions. Therefore, an active suspension preview controller with a knowledge-guided weight optimization method is proposed. This article introduces an offline method to enhance particle swarm optimization (PSO) through the augmented Lagrangian method (ALM) to mitigate the ill-conditioned problem. Meanwhile, the analytic hierarchy process (AHP) is employed to determine initial particles based on artificial knowledge, thus improving the efficiency of the PSO. During online implementation of the preview EMPC, a weight switch strategy is deployed to decouple the complex road condition by identifying the road class. Subsequently, the preview control experiment using actual sensing data is conducted on an electronic control unit (ECU) in the loop test platform, demonstrating a significant improvement over EMPC without optimization.
Zero-cost proxy fusion for neural architecture search (NAS) shows potential for efficient architecture evaluation. However, current proxy fusion methods still require extensive ground-truth samples for proxy generation and validation, incurring high computational cost. One of the keys to overcoming this challenge lies in developing automated fusion mechanisms that efficiently integrate the complementary strengths of diverse zero-cost proxies. To address this challenge, this work encodes such complementary information in the inverse covariance matrix. Based on the inverse matrix, the covariance-inverse fusion (CoIF) method is proposed to leverage the complementary strengths of existing proxies through covariance-weighted fusion. Theoretically, the proposed method is proven to provide optimal fusion weights and guarantee nondecreasing consistency. In addition, the variance of the estimated correlation is inversely proportional to the sample size, enabling reliable fusion with low ground-truth dependency. CoIF achieves competitive performance with 80% fewer samples and 99% less fusion time compared to existing learning-based fusion methods. Experiments across convolutional neural networks (CNNs), vision transformers, and large language models (LLMs) demonstrate that CoIF effectively integrates diverse architectural properties through proxy fusion. The closed-form fusion weights ensure both efficiency and theoretical optimality in the linear fusion case, establishing CoIF as an effective approach for zero-shot (ZS) NAS.
Safety, as a fundamental requirement in human-robot interaction, imposes high demands on autonomous decision-making. Existing studies often model such interaction as a perfect rational game with full observability, ignoring variation in rationality and perceptual constraints. To address these limitations, this article formulates the safe human-robot interaction as a partially observable system with bounded rationality and adopts the cognitive hierarchy (CH) model to characterize the interaction process. However, partial observability leads to discontinuous evolution of game states, resulting in a nonstandard differential game and causing difficulties for theoretical analysis. Based on hybrid system theory and the CH model, this work establishes a relationship between bounded rationality policies and Nash equilibrium, and provides interpretability for integrating CH with reinforcement learning. To further enhance safety, an adaptive low-conservative backward reachability analysis is introduced. Simulation experiments and evaluations on real-world traffic data validate the effectiveness, safety, efficiency, robustness, and predictive accuracy of our framework, reducing conservatism by more than 70% while causing negligible increases in collision risk and achieving approximately 18% higher prediction accuracy compared to another nonequilibrium game model, the level- $k$ model. Furthermore, compared with multiagent reinforcement learning and inverse reinforcement learning (IRL), our framework achieves a better balance between safety and efficiency.
This article presents a plug-and-play (PnP) distributed control framework for DC microgrids (DCmGs) to address scalability and reconfiguration issues under dynamic network topologies. First, a decentralized control architecture is developed based on the small-gain theorem, enabling PnP operations of distributed generation units (DGUs). Specifically, the plug-in operation only requires local and neighboring state feedback without relying on global information exchange, while the plug-out operation can be performed without interunit communication. In this way, controller deployment is decoupled from the global network topology, thereby reducing reconfiguration overhead and improving the modular scalability and operational flexibility of DCmGs. Second, an optimization-based controller synthesis method is proposed to minimize the coupling effects among DGUs. The proposed method facilitates distributed controller design for heterogeneous DGUs without requiring structured Lyapunov functions or free-weighting matrices, thereby simplifying the synthesis procedure. Theoretical analysis establishes the asymptotic stability and prescribed performance of the closed-loop system under PnP operations. Finally, simulation studies on a six-DGU DCmG prototype are provided to validate the effectiveness of the proposed framework.
The integration of power electronic interface-based renewable energy systems, along with fast dynamic loads, has significantly changed the dynamics of microgrids. This shift has reduced rotational inertia while introducing virtual inertia and fast frequency regulation strategies, enabling faster responses to fluctuations and necessitating real-time operation. Real-time operation, supported by advanced information and communication technologies, forms the backbone of this framework. However, it also exposes microgrids to cyber threats, which can disrupt operations by introducing delays in transmitted measurements and reference signals. To maintain microgrid resilient operations, an innovative control strategy is proposed for robust frequency regulation in microgrids under uncertainty against time-varying delay attacks. The proposed method exploits the advantages of fractional-order calculus and model-following control to enhance system operation resilience. Furthermore, a finite-time robust tracking control mechanism is utilized to ensure rapid responses. Performance of the proposed method is ensured through MATLAB simulations of a typical microgrid. Experimental test results obtained from the Opal-RT testbed are also presented. Results demonstrate the method's effectiveness in real-time operation within renewable microgrids, showcasing a swift response and reduced fluctuations.
This article investigates coordinated velocity tracking control for heterogeneous permanent magnet linear motors (PMLMs) under actuator faults and denial-of-service (DoS) attacks. Parameter heterogeneity, actuator degradation, and communication failures jointly threaten system stability and tracking accuracy. A prescribed-time performance-based hierarchical resilient adaptive data-driven control strategy is proposed to solve the formulated issue. First, a distributed resilient prescribed-time observer is designed to estimate the time-varying virtual reference velocity despite intermittent communication failures caused by DoS attacks. Subsequently, using the dynamic linearization (DL) method, the ideal controller is converted into an equivalent data model that relies solely on input/output (I/O) data, bypassing the need for accurate PMLM models. By introducing a global prescribed-time performance function (PTPF), the constrained velocity tracking errors are transformed into equivalent unconstrained conditions. Based on the estimated reference signal, a decentralized controller-dynamic-linearization (CDL)-based fuzzy adaptive tracking controller is designed, in which the unknown fault-related nonlinear term is approximated using a fuzzy logic system to compensate for actuator fault effects. The prescribed-time stabilization of the observation error and velocity tracking error is validated by Lyapunov stability analysis and contraction mapping techniques. Finally, simulation examples confirm the effectiveness of the proposed scheme.
This study investigates prescribed-time-synchronized (PTS) formation control for multiple autonomous surface vehicles (ASVs) subject to time-varying output constraints. A novel prescribed-time stability lemma is first established, which provides a general approach for constructing a new class of finite-gain adjustment functions that satisfy specified integral conditions. This approach effectively avoids the high-gain behavior typically caused by unbounded scaling functions in conventional temporal transformations. Based on this result, we develop a PTS constrained stable system (PTSCSS) that ensures simultaneous convergence of all state components to the origin within the prescribed time, with finite gains maintained throughout the time domain. Moreover, the time-varying constraints are strictly satisfied after a designated switching instant for any initial condition. Subsequently, a PTSCSS-based sliding manifold and a finite-gain sliding-mode formation controller are proposed for the ASV formation model. The PTS stability of the closed-loop system is rigorously proved via Lyapunov analysis. Finally, comparative simulations with multiple ASVs validate the effectiveness of the proposed control method.
This article copes with the problem of collaborative target tracking in multiagent systems (MASs) under the constraint of aperiodic output observations. First, a fully distributed adaptive event-triggered (AET) observer based on the aggregated observation measurement term is constructed to efficiently obtain aperiodic output observation information of the targets without relying on global properties of the interaction graphs. Within this framework, sensors exchange observations rather than raw measurements through the network, providing freedom and flexibility in information selection. A distributed control strategy is then developed to effectively accomplish the cooperative target tracking objectives subject to multiple constraints, including limited communication resources, aperiodic observations of the targets, and the requirement for a safe distance among agents. Moreover, the tracking errors are demonstrated to be ultimately bounded, ensuring the system's convergent behavior from a theoretical standpoint. Meanwhile, it is confirmed that neither the observation process nor the tracking control process exhibits Zeno behavior. Finally, comprehensive simulation findings are demonstrated to substantiate the efficacy of the designed control strategy.
Evolutionary multitask optimization (EMTO) aims to optimize multiple tasks simultaneously by transferring the related knowledge between tasks. Therefore, knowledge transfer (KT) between different tasks is crucial for facilitating the optimization of tasks. The traditional KT methods in EMTO achieve superficial KT through individual transfers, limiting their ability to transfer high-quality knowledge from other tasks. Therefore, in this article, a diffusion model-based KT (DMKT) method built specifically on cold diffusion (CD) is proposed to deeply mine the mapping relationships between different tasks and obtain transfer models for mapping individuals across tasks. Since CD supports arbitrary degradation operators, the interpolation between fitness-paired source and target task individuals can be defined as a task-oriented gradual degradation process, enabling the corresponding restoration process to learn directed cross-task mappings. In particular, we first construct two cyclic training diffusion models (DMs) for each source-target task pair. Subsequently, the trained DM can be used to generate new promising solutions for achieving efficient KT. The experimental results on the CEC2022 multitask optimization problem (MTOP) benchmark demonstrate that the proposed diffusion-based KT multitask optimization (DKTMTO) algorithm outperforms other state-of-the-art EMTO algorithms. Moreover, DMKT can be integrated into other EMTO algorithms to further improve their performance. Finally, DKTMTO is applied to real-world multitask planar kinematic arm control problems (PKACPs) and the WCCI2020 many-task optimization problem (MaTOP) benchmark, demonstrating its applicability and scalability.
This article studies privacy-preserving distributed Nash equilibrium (NE) seeking for aggregative games over directed graphs, where agents' cost functions contain sensitive information. A novel differentially private algorithm using decaying Laplace noise is developed to address two key issues: 1) how to design a distributed algorithm over directed graphs that achieves linear convergence while satisfying differential privacy requirements and 2) how to characterize the tradeoff between convergence accuracy and the privacy budget. First, sufficient conditions for linear convergence are established through the appropriate design of constant step sizes and convex combination parameters. Second, the differential privacy properties of the algorithm are analyzed without assuming bounded gradients, and a quantitative relationship between convergence accuracy and privacy budget is characterized. Furthermore, under additional restrictions on adjacent functions, the cumulative privacy budget admits an explicit expression and remains finite over an unbounded horizon, while the proposed algorithm is proven to converge to the exact NE. Finally, the effectiveness of the proposed algorithm is validated through a Nash-Cournot game and comparative simulations, which demonstrate its superior convergence performance compared to existing methods.
Physical laws widely exist in many industrial processes, which can be used to enhance model reliability and avoid physically inconsistent results. In this study, a probabilistic linear latent variable model is developed with readily accessible physical knowledge (PI-PQM) for quality-relevant monitoring applications. A block-matrix-based structure is adopted in the proposed method to satisfy the properties of different variables. Within this framework, physical principles (i.e., boundedness and monotonicity) are imposed on the selected variables as inequation constraints. In particular, both constraints are formulated with a skew normal distribution, which can be recast as a combination of normal and truncated normal distributions. After that, the variational inference (VI) algorithm is utilized for estimating the latent variables and model parameters of the proposed method. It is noted that the effect of the physical constraints is analyzed by comparing the latent variables and coefficient matrices of constrained and unconstrained variables. The performance of the proposed method is illustrated using two numerical simulations and a real industrial application. Experimental results verify that physical information helps the proposed method achieve accurate convergence to the real structure of the collected data. Besides, the monitoring sensitivity for both process- and quality-related anomalies is also effectively improved after adding physical constraints.
This article investigates the security control problem with guaranteed communication connection and collision avoidance (GCCA) for an unmanned aerial vehicle (UAV) swarm subject to multisource disturbances and cyber attacks. The network time-delay model of the communication link attacked is introduced first. To ensure communication, connection, and collision avoidance (CA), an improved artificial potential field (APF) function is developed. Then, a reinforcement learning (RL) model-free attitude controller scheme is proposed based on a disturbance suppression trajectory controller to accommodate the adverse effect of multisource disturbances, including matched and unmatched ones; meanwhile, it is combined with GCCA, which successfully accomplishes the control objectives. In particular, by utilizing the improved reciprocally convex combination approach, a sufficient criterion is given to guarantee that the UAVs satisfy the finite horizon H-infinity consensus performance. Finally, a comparative experiment is presented to verify the feasibility of the research.
Nonlinearity and dynamics are important issues in industrial process monitoring. The dynamic relationships between samples are often embedded in the latent variable space and are usually nonlinear. The fitting of nonlinear dynamic relationships largely determines the monitoring performance. In addition to the effectiveness of fitting, low computational complexity and high interpretability are also crucial. To address these issues, this study proposes a novel process monitoring algorithm called the dynamic inner neural network based on the simplified self-attention mechanism (DiNN-SSAM). This algorithm maps the process variables into the latent variable space that best reflects the dynamic relationships between samples through neural networks. Then, this algorithm fits the dynamic relationships between samples through the simplified self-attention mechanism. DiNN-SSAM not only establishes an accurate dynamic model but also reduces the complexity of online computation. In addition, this study creatively provides the fault detectability analysis of neural-network-based process monitoring algorithms and further fault sensitivity analysis, which compensates for the weak interpretability of neural networks. Finally, the superior performance of DiNN-SSAM is verified by the simulations of numerical examples, the Tennessee Eastman (TE) process, and catalytic cracking units.
This article intends to study the asynchronous $\mathscr {H}_{\infty } $ control problem for discrete-time networked Markov jump linear systems (MJLSs) subject to aperiodic denial-of-service (DoS) attacks. A more practical scenario of the simultaneous presence of imperfect mode detection and imperfectly known probability information is considered. More specifically, we use a hidden Markov model (HMM) to characterize the mismatched detected/system modes, and a general description of imperfectly known probabilities to characterize both the transition and observation probabilities (T/OPs). Then, we propose an HMM-based asynchronous control law for the concerned networked MJLS, and several sufficient conditions are established to ensure its stochastic stability and $\mathscr {H}_{\infty } $ performance, by means of an iterative method of piecewise Lyapunov functional and a nonconservative probability decoupling principle. We also show that the asynchronous control law can be designed via an LMI-based convex optimization algorithm. In the end, we use an example of a single-link robotic arm to demonstrate the effectiveness of the obtained results.
Predicting knee joint trajectory is critical for controlling intelligent walking-assistive devices, with surface electromyography (sEMG) emerging as a promising modality for motion intention decoding. However, accurate and continuous prediction remains challenging because both intersubject and intrasubject variability must be addressed simultaneously. To tackle this problem, this article proposes a semi-subject-independent deep learning framework that is pretrained on source subjects to learn shared cross-subject representations and then calibrated with only a few trials from an unseen target subject before testing on that subject's held-out trials. The framework contains two complementary components. First, gait kinematic decoupling (GKD) separates knee trajectory prediction into a shared normalized motion pattern and subject-dependent amplitude and offset terms, thereby reducing cross-subject label variability. Second, muscle activation filtering uses physiological activation priors to suppress motion-irrelevant sEMG components and enhance gait-related neuromuscular information. Experiments on both in-house and public datasets show state-of-the-art performance, with average root-mean-square errors (RMSEs) of $3.03^{\circ }~\pm ~0.49^{\circ }$ and $4.49^{\circ }~\pm ~1.14^{\circ }$ , respectively, while predicting knee angles 50 ms in advance. These results suggest that the proposed framework can support robust and practical control of intelligent walking-assistive systems.
Gene expression profiling (GEP) provides critical insights into the cellular states, phenotypes, and biological behaviors of biological systems. However, the traditional GEP acquisition methods have significant time limitations, which directly hinder the real-time monitoring of dynamic processes, such as cell development and disease progression. In this study, we propose the biological omics sparse sensing (BoSS) framework, which effectively resolves the compatibility issue between dimensionality-reduced sampling theory and biological reaction by introducing sparse sampling strategies and noise regularization constraints. Furthermore, we have first presented rapid capture and accurate reconstruction of GEP using multiplex RT-qPCR, increasing the efficiency of transcriptome data acquisition by 60-fold. The experimental results demonstrate that when evaluating transcriptional heterogeneity in cancer tissue blocks, BoSS exhibits excellent specificity and sensitivity in clustering analysis, while fully preserving key information related to biological states, thus ensuring that the accuracy of downstream data analysis is comparable to that of ribonucleic acid (RNA)-seq. This achievement indicates that the BoSS framework is expected to play an important role in scenarios, such as intraoperative assessment of cancer progression, personalized therapy, and predictive control of biological systems.