AT present,the world is experiencing profound transfor-mations unseen in a century,with a new round of scien-tific and technological revolution increasingly inter-twined with major-power competition.Against this backdrop,artificial intelligence(AI)has emerged as a key strategic technology that is profoundly reshaping global economic,social,and military systems.Through top-level policy design and national devel-opment planning,China has been continuously promoting the deep integration of AI with industry,energy,education,and other key domains.Therefore,the intelligent era should not be regarded as a future prospect,but rather as an ongoing transformation that is already reshaping technological development and social progress.
Dear Editor, This letter is concerned with the robust tracking control problem of automated vehicles(AVs).First,the dynamics of an AV is con-structed by taking into account model uncertainties.Then,to guaran-tee the successful completion of tracking tasks,an actor-critic learn-ing-based robust tracking control scheme is designed.At last,formal stability analysis and experiment results are provided to verify the tracking performance of the designed control scheme.
Dear Editor, Nonlinear dynamic system is difficult to model by traditional mod-eling methods,because of high nonlinearity and limited precision.Koopman operator,an effective technique,maps nonlinear dynamic systems to a higher-dimensional linear space,allowing complex non-linear behavior to be described linearly.However,conventional Koopman-based approaches often suffer from computational com-plexity,this paper employs a deep neural network with random search to learn feature mappings and estimates the Koopman opera-tor by subspace identification.Experimental results show the approach simplifies modeling and significantly improves modeling accuracy over traditional methods.
Dear Editor, The crafted generalized replay attacks(GRAs)can bypass tradi-tional multiplicative watermarking(MW)-based active detection in networked control systems(NCSs),which will seriously destroy sys-tem stability or even crash.To address this problem,this letter pro-poses a novel secure control method by using MW-based detection and data compensation.First,the limitation of traditional MW-based detection method is analysed,and a novel MW-based active detec-tion scheme is proposed by adding an irreversible watermarking detection unit.Then,according to the detection result,an online data compensation scheme based on cubic spline interpolation algorithm is provided,and the maximum allowed attack rate of GRAs is given to maintain the exponential stability of NCSs.Finally,experimental results confirm the effectiveness of the proposed method.
Dear Editor, This letter deals with the security control for nonlinear cyber-physical systems (CPSs) under mixed deception attacks. Both sensors and actuators are assumed to be injected deception data during the data transmission via networks. In order to identify the unknown dynamics of the attacked system, a neural network (NN) is adopted, on basis of which an NN-based secure observer is designed to diminish the attack impact on state estimation. Then, by resorting to the reinforcement learning approach, the secure control strategy is presented via actor-critic and zero-sum games. At last, the designed control scheme is proved via a numerical simulation.
Dear Editor, This letter proposes a fully distributed multi-agent reinforcement learning(DMARL)algorithm for the coordinated optimization and scheduling of source-load-storage in networked microgrids.To accommodate the rapid development of networked microgrids,we have designed a DMARL algorithm with an event-triggered mecha-nism(ETM).Unlike centralized approaches,DMARL empowers individual agents to cooperatively learn and optimize based only on local observations,while reducing the communication burden.Simu-lation results validate the performance of the proposed algorithm,demonstrating its effectiveness in efficient microgrid resource man-agement.
Dear Editor, This letter presents an intelligent fault diagnosis method for vari-able speed rolling bearings based on the adaptive short-time frac-tional Fourier transform(ASTFrFT)and the time-frequency BoTNet(TFB)to address the challenge of extracting fault characteristics of rolling bearings under variable speed conditions and the poor classifi-cation of classical deep learning models.Firstly,to address the limi-tations of FrFT in time-varying signal processing,the physical mech-anism of traditional STFT is extended into the FrFT domain by mini-mizing fuzzy entropy values to construct the order matrix.Subse-quently,each signal segment undergoes rotation filtering to derive its time-frequency(TF)distribution.Finally,the resulting TF images are inputted into the TFB,which utilizes fully connected and classifica-tion layers to identify two-dimensional TF domain features for intel-ligent diagnosis automatically.Experimental validation and compara-tive studies confirm the effectiveness and superiority of the proposed method.
Dear Editor, This letter investigates the fixed-time fault-tolerant control(FTC)problem for small unmanned underwater vehicles(UUVs)subject to the full-state error constraints involving position-layer and velocity-layer.First,a dual-level evolving performance boundary is devised by integrating the fixed-time performance functions and low-com-plexity error transformation techniques.This novel formulation con-verts the full-state constrained control issue into a dual-layer uncon-strained stabilization,thus ensuring fixed-time convergence regard-less of initial conditions.Subsequently,based on the fuzzy logic sys-tem(FLS)and barrier Lyapunov functions(BLF),a robust FTC scheme is derived to enforce full-state constraints within desired behavioral boundaries and achieve the robustness against dynamics uncertainties,external disturbances and actuator faults.Finally,it is proved that the closed-loop system is stable and state constraints are never violated.Simulations verify the effectiveness of the proposed scheme.
Dear Editor, Industrial processes have become increasingly complex and dynamic,making robust and interpretable modeling techniques indis-pensable for safe and efficient operation.Although contemporary machine learning algorithms exhibit strong predictive capabilities[1],[2],it remains challenging to ensure trustworthy performance-i.e.,models that reliably generalize to diverse operating regimes while offering intuitive explanations of their behavior.Achieving high accuracy with robust interpretability and stability(especially under process upsets or nonstationarities)stands as a pressing need in modern industrial environments[3],[4].
In this paper, the problem of proportional-integral observer (PIO) design is investigated for a class of discrete-time multi-rate systems with multiple sensors, with the sensor sampling periods being allowed to differ from the system updating periods. The facilitation of communication between sensors and the remote PIO through wireless networks, which are subject to probabilistic packet dropouts, is achieved through the utilization of a decode-and-forward relay-based strategy. The occurrence of packet dropouts is governed by a Bernoulli-distributed random variable whose probability is dependent on the available transmission power. A decode-and-forward relay-based strategy, developed based on different components, is capable of processing information from different encoders at different physical locations. For the convenience of observer design, the lifting technique is employed with aim to cast the multi-rate system into a single-rate one. By establishing sufficient conditions, the combined effect of external noises and relaying-aided communication on estimation performance is intuitively illustrated. Subsequently, a PIO with an adjustable parameter is designed by solving certain optimization problems. A simulation example is finally provided to validate the theoretical results.
Quality of service (QoS) data that characterize historical user-service invocations that vary over time are vital to discovering patterns of cloud services and understanding user behaviors. Though effective, prevalent approaches never consider higher-order spatiotemporal connectivity within QoS data, thus suffering from inferior performance. To address this critical issue, this paper presents spatiotemporal graph convolutional network (GCN) that is equipped with the functionality of latent factorization of tensors (SGLFT). It is achieved by introducing three key innovations: 1) Proposing a tensor graph convolution based on the generalized tensor product technique for uniformly modeling the temporal and spatial patterns within dynamic user-service graphs; 2) Incorporating the built layer-wise graph convolution into tensor factorization for efficiently capturing the implied spatiotemporal high-order connectivity; and 3) Developing a node-level attention pooling mechanism to perceive feature differences among neighbors and across time slots. Theoretical derivations are conducted to demonstrate that the expressivity of the graph neural network proposed in this paper is evidently higher than that of vanilla GCNs. Empirical studies on eight large-scale testing cases arising from two real-world dynamic QoS datasets show that SGLFT substantially outperforms state-of-the-art QoS estimators regarding estimation accuracy for missing dynamic QoS data.
With the rapid integration of modern information technologies into agriculture, smart agriculture enables increasingly precise phenotyping and yield prediction. Flower counting is a key phenological indicator; however, achieving high precision remains a significant challenge due to severe occlusion, density variations, and environmental variability (e.g., lighting/weather). Moreover, existing studies remain fragmented without a comprehensive synthesis. To bridge this fundamental research gap, we present the first systematic survey of computer-vision-based flower counting. In this work, we propose a novel taxonomy that categorizes methods into static (single-image) and dynamic (video/multi-view) paradigms and elucidates their evolutionary trajectory. Unlike conventional reviews, we conduct a multi-scale evaluation encompassing both horizontal (methodological evolution from traditional to deep learning) and vertical (cross-species and scene-condition) performance analyses. Crucially, we validate representative algorithms on deployed platforms-UAV and ground robots-through engineering case studies that quantify real-world trade-offs (e.g., height-accuracy and latency-robust-ness). Finally, we discuss prevailing limitations and propose future directions, including graph-based reasoning and AgriVerse inte-gration (i.e., agriculture-centric metaverse or digital twin ecosys-tems), establishing a foundational framework for both academic research and industrial deployment.
Deep neural networks(DNNs)have been widely applied in the field of synthetic aperture radar(SAR)image while they are facing more and more serious threats from a variety of malicious attacks.As one of the malicious attacks with strong destructiveness and stealth,backdoor attacks have severely affected DNNs,but there are no related research studies concern-ing the backdoor attacks against the DNNs-based SAR image classification models.In this work,we make the first attempt to automatically design a constrained multi-objective invisible and adaptive backdoor attack termed as CMo-IABA for DNNs-based SAR image classification.In the CMo-IABA,we firstly generate an initial trigger-based backdoor attack randomly by a random combination of pixels with random noise conforming to the Gaus-sian distribution.Then,we design multi-objective functions by considering the trade-off between maximizing the attack success rate and minimizing L2 distance-based invisibility.The classifi-cation error between the backdoor DNN and the clean model is considered as the constraint to maintain the original perfor-mance of the model.To solve the optimization problem,a dis-crete non-dominated sorting genetic algorithm-Ⅱ is introduced as the search engine with the developed crossover operation and mutation operation.The superiority of the proposed CMo-IABA to five state-of-the-art backdoor attacks on six different types of DNNs-based SAR image classification models has been demon-strated by the experimental results on Fudan University SAR(FUSAR)-ship and moving and stationary target acquisition and recognition(MSTAR)datasets in terms of attack success rate and L2 distance-based invisibility.
The heuristic function in Petri-net-based A＊ search directly influences both the search efficiency and solution quality for scheduling resource allocation systems(RASs).In the litera-ture,some heuristic functions have been proposed,but most of them fail to consider key aspects such as token remaining time,alternative routes,weighted arcs,multiple resource copies,and batch processing ability,which are common in the place-timed Petri nets(PNs)for RASs.This paper proposes two novel heuris-tic functions.Both are admissible,guaranteeing the optimality of the obtained schedules.In addition,they are designed not only for ordinary PNs but also for generalized ones,which may have arc weights greater than one.They can effectively handle RAS PNs with alternative routes,weighted arcs,multiple resource copies,and batch processing capability.Most importantly,the new heuristics,especially the second one,are highly informed,leading to faster searches for optimal schedules compared to existing heuristics for generalized PNs.Experiments on several bench-mark PNs of RASs have been conducted to demonstrate the effectiveness and efficiency of our methods.
In this paper, the problem of adaptive event-triggered (ET) fixed-time tracking control (FTTC) for a class of uncertain nonlinear systems (NSs) with partially unmeasurable states is investigated. An observer and a set of radial basis function neural networks (RBFNNs) are introduced to reconstruct the unmeasurable states and to approximate the unknown nonlinear functions, respectively. Moreover, in order to reduce the communication resource consumption, an ET mechanism with the relative threshold is adopted. The designed ET controller can ensure that the tracking error converges to a small neighborhood of the origin, all the signals of the closed-loop system are uniformly ultimately bounded (UUB), the settling time depends solely on the design parameters, and the Zeno behavior is successfully avoided. A practical example is given to verify the feasibility of the proposed method.
This paper addresses the protection-strategy-based distributed state estimation (SE) problem for time-varying non-linear complex networks, where uncertain inner coupling and random false data injection attacks are considered. Owing to the fact that the measurement signals can be easily injected with false data by potential attackers before being transmitted to the estimator, a novel protection strategy is firstly proposed from the perspective of the defender to mitigate the effects of malicious attacks on the estimation performance. Based on the proposed protection strategy, more suspicious and unreliable measurements received are identified and discarded respectively. Accordingly, the zero-order holder strategy is employed to compensate for the discarded measurements. Subsequently, the purpose of this paper is to design a protection-strategy-based distributed SE scheme such that an optimized upper bound (UB) on the estimation error covariance (EEC) is obtained. Furthermore, a sufficient criterion is provided to guarantee the uniform boundedness of UB on the EEC. Finally, a localization problem involving multiple mobile robots is used to demonstrate the effectiveness and practicality of proposed variance-constrained optimized SE method.
In this paper, a zonotopic distributed fusion estimation problem is investigated for a class of 2-D nonlinear systems subject to unknown-but-bounded noises over a binary sensor network. An auxiliary innovation is constructed to reduce the influence of the less measured information, and FlexRay protocol is employed to schedule the innovation transmission between nodes. By resorting to the set-membership filtering, a variable-independent zonotope is achieved to constrain local estimation error, and gain parameters are obtained by minimizing the upper bound of zonotope in the F-radius sense. Subsequently, a zonotopic distributed fusion scheme is implemented through the matrix-weighted fusion criteria, and an optimized weighted matrix is obtained using the Lagrange Multiple method. Furthermore, the monotonicity of the local zonotope is analyzed with the gain-constraint parameter. Finally, a numerical example is considered to verify the effectiveness of the developed fusion algorithm.
With the growing emphasis on intelligent agriculture, scheduling problems involving multiple harvesters in large-scale croplands have attracted increasing attention. In these scenarios, harvesters collaboratively perform tasks in expansive croplands, with each harvester taking on a portion of the workload. This study defines such problems as the harvester scheduling problem with splittable workloads (HSPSW). Unlike most multi-robot task allocation problems, HSPSW requires collaboration among several harvesters and involves more harvesters than croplands. These features increase the complexity of har-vester-cropland interactions. To address these challenges, this paper proposes a novel two-step iterative local search (TSILS) algorithm. A greedy strategy that prioritizes croplands with higher workloads is proposed to generate initial solutions. Subsequently, a two-step iterative search strategy is employed to obtain high-quality solutions: the first step uses a predefined objective to facilitate rapid workload allocation for feasible solutions, and the second step integrates specific neighborhood operators and fine-grained local search to improve its search capability. Experimental results demonstrate that the proposed method TSILS significantly outperforms existing approaches in both algorithmic performance and computational efficiency.
It is crucial to forecast passenger flows for optimizing subway operations and improving travel experience for passengers. While traditional methods usually estimate the average travel time of passengers, this study delves into the modeling and prediction travel time distributions from all origin stations to all destinations. We construct a matrix that includes the travel time and destination for passengers originating from each station at a time slot. We propose a traveltime and destinations distribution modeling approach for one-day ahead predictions for each origin station. By quantifying the multi-time-scale similarities of week-to-week, day-to-day, and time-to-time, we unveil the underlying mechanisms of passengers' travel patterns, revealing predictable and repetitive mobility behaviors. We validate the proposed model using real subway passenger data sets and demonstrate its superior performance in predicting the passenger flows and the associated travel time.
A well-designed charging matrix (CM) is crucial for advancing green and low-carbon production in the blast furnace (BF) ironmaking process. Over recent years, metaheuristics algorithms have been applied to optimize CM, partially reducing reliance on on-site workers. However, CM optimization is a challenging mixed-variable constraint optimization problem. Prior studies predominantly simplify CM to either continuous or discrete forms via variable fixation or type conversion, which hinders the efficient joint optimization of heterogeneous variables, limiting optimization accuracy and search efficiency. To tackle this barrier, this study proposes a novel method named Hybrid Encoding-based Adaptive Coordinated Differential Evolution (HE-ACoDE), marking the first attempt to optimize CM from a mixed-variable perspective. First, a hybrid encoding scheme is devised to provide a unified representation for the mixed variables in CM. Then, a coordinated mixed-variable mutation strategy is developed, effectively facilitating the synchronized evolution of continuous and discrete variables. Moreover, a constraint-aware selection operator and a weight-guided parameter adaptation strategy are proposed, which collaboratively guide the population toward feasible, high-quality solutions across different evolutionary stages and problem landscapes. Extensive comparison experiments on two industrial scenarios demonstrate that HE-ACoDE outperforms state-of-the-art CM optimization and mixed-variable optimization methods in terms of accuracy, stability, and convergence performance.
