This article investigates the fixed/prescribed-time synchronization problem for fuzzy inertial memristive neural networks with hybrid delays. Within a unified framework, a comparative study was conducted on the interval matrix method and the maximum absolute value method with respect to the state-dependent switching parameters induced by memristive characteristics. Correspondingly, two different controllers are designed, with theorem constraints expressed algebraically and as linear matrix inequalities (LMIs). Numerical experiments demonstrate that, under identical initial system parameters, the interval matrix method constructs a Lyapunov-Krasovskii functional incorporating an integral term, enabling finer handling of time-delay effects. As a result, it provides a more accurate estimate of the upper bound of the settling time compared to the maximum absolute value method, thereby achieving faster and more efficient synchronization control. Ultimately, the proposed results are applied to image encryption, thereby demonstrating its theoretical significance and practical utility.

Given the existence of random disturbances and dynamic modeling uncertainties, this article addresses the fixed-time synchronization in probability for a type of fuzzy stochastic dynamic networks under the leaderless and leader-following frameworks. Firstly, a relaxed fixed-time convergence condition for stochastic systems is proposed, which improves the accuracy of convergence time estimate and refines the range of condition parameters. Subsequently, by skillfully introducing an intermediate signal subject to random interference, two types of distributed fuzzy control strategies are developed, which enhances the system’s security and robustness. Building upon the derived stochastic stability result, several stochastic synchronization criteria in fixed time are obtained. Lastly, numerical simulations are conducted to corroborate the theoretical findings.

This paper investigates the fixed-time synchronization (FXTS) problem for an array of delayed memristive reaction-diffusion neural networks (DMRDNNs) under semi-intermittent switching control (SISC). The proposed model, formulated as a partial differential system, incorporates both temporal and spatial effects, thereby offering broader applicability than models considering only time. Instead of adopting the traditional fixed-time stability criterion, which is commonly used with two power exponent terms, a novel lemma based on a hyperbolic-cosine function is established, containing only one power term. This study improves upon previous related works by achieving a more precise estimation of the settling time (ST). Furthermore, by designing an appropriate controller and applying Green’s formula, the Lyapunov method, and inequality techniques, sufficient conditions are derived to guarantee the FXTS of the drive-response DMRDNNs. Finally, numerical simulations are conducted to validate the effectiveness of the proposed theoretical results.

This paper investigates the distributed adaptive finite/fixed-time synchronization problem for uncertain complex dynamical networks subject to state-dependent uncertainties and partial actuator faults. To derive its synchronization criteria, we firstly introduce the Radial Basis Function neural networks, and employ them to approximate unknown system dynamics. Then, we propose a novel distributed adaptive control strategy to compensate in real time for the performance degradation caused by actuator faults without requiring prior knowledge of fault information. Finally, we use a unified Lyapunov-based stability framework to analyze the synchronization problem of the closed-loop system and establish its practical finite-time and fixed-time synchronization conditions. It is worth mentioning that the convergence time in the fixed-time case is explicitly independent of initial conditions. A generic complex network is employed to verify the theoretical results, and the practical applicability of the proposed scheme is validated through a networked robotic manipulator system.

Accurately forecasting real-world vehicle emissions is increasingly important as global carbon and pollutant reporting requirements become more rigorous, which creates a growing need for robust and reliable emission modelling approaches. However, existing vehicle emission models often struggle to simultaneously capture complex nonlinear emission dynamics, long-range temporal dependencies, and model interpretability under real-world driving conditions. This study proposes a hybrid Kolmogorov-Arnold Network and Transformer architecture (KANFormer) that integrates KAN with Transformer to jointly capture complex nonlinear relationships and long-range temporal dependencies in exhaust-emissions data. Using real driving data from 30 drivers operating a diesel van across urban, rural, and highway conditions, the model was benchmarked against non-temporal models, classic time-series models, and alternative different combinations of KAN and Transformer models. Results show that incorporating temporal dependencies greatly improves accuracy, particularly when sequence lengths are sufficient to account for gas-transport lag. The KAN-based nonlinear layers further enhance prediction performance by adaptively modelling complex combustion-related relationships, especially for highly nonlinear pollutants such as nitrogen oxides (NOx). Compared with conventional models, the proposed architecture demonstrates stronger robustness under limited training data while maintaining competitive performance across both CO2 and NOx prediction tasks. Improved interpretability is achieved through spline-based sensitivity analysis, which provides transparent and physically meaningful insights into how driving behaviours and engine operating conditions influence emissions. The proposed KANFormer therefore offers an accurate and interpretable framework for real-world vehicle-emission forecasting and for investigating the relationships between driving behaviour, vehicle operation, and emission dynamics.

Multisynchronization of multistable stochastic neural networks (MSNNs) is investigated herein. First, an MSNN multistable stochastic neural network (SNN) model subject to both time-varying delays and parameter uncertainties is constructed, which accurately describes neural dynamics in realistic environments. Subsequently, to reduce control costs, a coupled impulsive control strategy is adopted for studying the multisynchronization problem among MSNN systems. By developing an appropriate Lyapunov functional and leveraging the average impulsive interval concept, sufficient conditions for multisynchronization of the considered SNNs under both fixed and switching topologies are derived. Finally, the effectiveness of the proposed control scheme is verified through a numerical example.

This paper investigates the problem of zero-dynamics attack (ZDA) detection for networked power systems with electric vehicles (EVs). The communication channel of networked power systems is assumed to be compromised by malicious zero-dynamics attackers. A watermark-based auxiliary function detection method is proposed to identify the occurrence of ZDA. To overcome the disadvantage that the watermark defense strategy can degrade system controller performance, we remove the encrypted watermark before the control signal is transmitted to the actuator. Considering that static watermarks are easy for attackers to obtain, we propose the use of time-varying dynamic watermark encryption for transmission. The stealthiness of the zero-dynamics attack is verified by theoretical derivation. The simulation results reveal the harms and characteristics of ZDA, and the comparative experiment verifies the differences and superiority of the proposed method compared with common X2 detection methods.

This manuscript rigorously investigates finite-time projective synchronization in distinct delayed state-dependent switching neural networks governed by Caputo fractional-order, working on novel upper bounded estimate for setting time. The initial phase involves the integration of master-slave systems with distinct structures and parameters, and the state-dependent discontinuities are handled using the Filippov differential inclusion framework, then a synchronous error system is subsequently established. Subsequently, novel parameter conditions for accomplishing finite-time projective synchronization are posited, and corresponding upper bounded estimate for setting time is appraised using fractional-order inequality. In addition, based on output observation and sliding mode control, a derivative-based sliding surface and its associated controller are constructed, and the reachability of the sliding mode surface is subsequently analyzed by means of relevant lemmas and the fractional-order Lyapunov direct method. Furthermore, the stability of the sliding path is corroborated, indispensable prerequisites and a novel upper bound estimate of setting time necessary for achieving finite-time projective synchronization are delineated. To demonstrate our approach, three-dimensional distinct fractional-order delayed state-dependent switching neural networks are simulated with continuous frequency distribution model.

Recently, the synchronization problem of variable-order fractional networks has garnered considerable attention. However, the intrinsic variability of the fractional order poses substantial challenges to convergence analysis, which hinders a rigorous and complete solution to the synchronization control. To address this, we investigate the passivity and bipartite synchronization of variable-order fractional spatio-temporal signed networks. Firstly, a new variable-order fractional integral inequality and a convergence result based on a variable-order fractional condition are developed, which provide a key theoretical foundation for adaptive control of variable-order systems. Furthermore, some pinning adaptive schemes are proposed, and sufficient criteria are derived for ensuring the passivity and bipartite synchronization, by decomposing the Laplacian matrix of the signed graph rather than the traditional gauge transformation. Finally, the validity of the proposed control schemes and derived criteria is demonstrated by numerical examples.

In the above article, reference [30] is incorrect. The correct reference should be the following:

This paper investigates the memory-based resilient control problem for cyber-physical systems (CPSs) under hybrid cyber attacks and actuator saturation constraints. First, a hybrid cyber attack model that incorporates both aperiodic denial-of-service (DoS) attacks and deception attacks is developed. To mitigate the impact of cyber attacks on the sampling process, a switching period sampling mechanism (SPSM) is introduced, and a rigorous analysis is conducted on the impact of attacks on the sampling timing under SPSM. Next, an improved adaptive dynamic memory event-triggered scheme (ADMETS) is proposed. This mechanism significantly reduces the number of unexpected triggering events caused by uncertain disturbances, thereby effectively alleviating the network bandwidth load and improving control performance. Then, Lyapunov-Krasovskii functionals (LKFs) and linear matrix inequality (LMI) techniques are employed to derive sufficient conditions ensuring exponential mean-square stability under a prescribed HPo performance criterion. Finally, simulation results on an autonomous ground vehicle (AGV) trajectory tracking task validate the effectiveness of the proposed method in improving communication efficiency and system robustness.

This study explores fixed-time (FXT) stabilization, predefined-time (PDT) stabilization for a specific category of nonlinear dynamic systems. Through the introduction of the hyperbolic cosine function and the application of the Lyapunov approach, innovative FXT and PDT stabilization theorems are put forward. Leveraging these theoretical outcomes, FXT and PDT stabilization is realized for memristive neural networks (MNNs) with mixed delays by devising a suitable control strategy. Eventually, numerical simulation cases are presented to verify the core research results.

This article investigates finite/fixed-time synchronization (FTS/FXTS) for delayed inertial memristive neural networks (IMNNs) with time-varying actuator faults, a more practical scenario compared to the widely studied constant-fault case. Novel fault-tolerant controllers, including state-feedback and event-triggered schemes, are proposed to achieve synchronization under mixed delays. By establishing algebraic criteria via a nonreduced-order approach, explicit settling time estimates are derived while excluding Zeno behavior. The theoretical results are verified through simulations, and the proposed method is further applied to secure communication using synchronized IMNNs for image encryption.

We conducted an in-depth investigation into the impact of conditional variational autoencoders (CVAEs) and Bayesian neural networks (BNNs) on high dynamic range (HDR) image reconstruction. A parallel multiattention module (PMAM) is introduced in the improved YOLOv10 framework to enhance computational efficiency and detection performance. To reconstruct HDR images, we enhance the asynchronous Kalman filter (AKF) algorithm to improve image detail quality. We introduce BNN and CVAE into the AKF algorithm to reduce noise effects and improve the logarithmic intensity of the reconstructed image. The BNN estimates noise covariance, thereby reducing its impact during the reconstruction process. Simultaneously, the CVAE leverages polarity as a conditional input, and uses spatial and temporal information through a CVAE to generate more accurate logarithmic image intensities. In the object detection stage, we integrate a parallel module combining the self-attention mechanism and the ECA module to improve training efficiency without increasing the number of parameters. This PMAM module, based on improved YOLOv10, strengthens the model's ability to capture global and channel-specific features. Finally, the proposed method's accuracy and robustness are validated through extensive simulations and comparative experiments. Comprehensive experiments on public datasets show that our model achieves 81.32% mAP@0.5 and 59.67% mAP@[0.5:0.95], demonstrating significant improvements in detection accuracy and image reconstruction quality.

This paper is devoted to exploring prescribed-time stability of impulsive systems and prescribed-time cluster/complete output synchronization for heterogeneous network models under hybrid impulsive control. Firstly, by defining a general type of regulatory function, a flexible prescribed-time stability theorem about nonlinear impulsive models is established, which improves some well-known results. Furthermore, for a type of heterogeneous networks with output coupling, two output-based hybrid impulsive controllers are developed, and a novel impulsive-time-dependent Lyapunov function is constructed to relax the restriction on the control gains and derive less conservative criteria of prescribed-time cluster/complete output synchronization. A typical numerical example and an application in heterogeneous vehicular networks are presented finally to confirm the effectiveness of the results obtained.

Recent tudies have shown that most spiking neural networks (SNNs) for speech classification only use a fixed temporal resolution, limiting their capacity to capture multiscale temporal information. To address this challenge, we propose the temporal reconstruction (TR) method, which enhances its learning ability on multiscale temporal information by leveraging the hierarchical temporal processing mechanisms of the human brain in speech comprehension, enabling efficient reconstruction of temporal representations. Additionally, we propose temporal alignment (TA) method to enable the residual connection being used between the data with different temporal dimensions. According to the experiment results, we have obtained a SOTA result 96.04% on spiking Heidelberg digits (SHD), and we have achieved a classification accuracy of 81.02% on spiking speech commands (SSC). Furthermore, on the nonspiking dataset Google speech commands v0.02 (GSC), we can achieve the classification accuracy of 95.63%.

Traffic violations constitute one of the principal contributors to serious road accidents and remain a persistent threat to public safety and property worldwide. For this reason, accurate identification and prediction of traffic violations are of considerable importance for improving traffic governance and supporting early intervention. To address the challenges posed by traffic violation data with complex structures and heterogeneous feature distributions, this paper proposes a new classification framework: TrafficViolationNet. The proposed model integrates an enhanced residual architecture with a lightweight attention mechanism to improve feature learning from structured traffic data. At the architectural level, TrafficViolationNet is built upon ResNet Plus, in which auxiliary residual branches and dense shortcut connections are introduced to facilitate information propagation, improve gradient flow, and strengthen feature representation. In addition, an attention module is incorporated into each residual block to adaptively emphasize informative features and capture complex dependencies among variables. Experimental results on traffic violation datasets from the United States and Qatar show that the proposed method consistently outperforms mainstream machine learning baselines and achieves state-of-the-art classification performance.

Electroencephalogram(EEG)-based emotion recognition has become a prominent research focus in affective computing. Existing studies mainly analyze the temporal, spatial, and frequency domain characteristics of EEG signals to construct methods, but few incorporate neuroscientific theories. Neuroscience shows that EEG signals from different brain regions correlate differently with emotions. Even further, our preliminary experiments found that although whole-brain signals performed well on average, certain important regions often achieved higher accuracy rates. Inspired by this, we propose a Local-guided Global Learning Multi-Stream Network (LGL-MSNet) for EEG emotion recognition, which utilizes the differential of local brain regions to guide the model in effectively learning from the entire brain, ensuring that its performance is at least as good as that of key brain regions. Specifically, LGL-MSNet contains four local streams corresponding to individual brain regions and one global stream. Each stream is a straightforward model with independent classification capabilities, consisting of a feature extractor and a classifier. Subsequently, the LGL block is used to inject local-stream information into the global stream, providing weighted guidance from two levels: data-driven adaptive weight and result-driven prediction confidence. Furthermore, a prediction-stability pre-training strategy is adopted to enhance the initial guidance capability of local streams. Experiments on the SEED, SEED-IV, and SEED-VII datasets demonstrate the effectiveness of LGL-MSNet, and validate that local information can assist global learning in EEG emotion recognition. The code is available at https://github.com/braverSheep/LGL-MSNet.

This paper investigates the stability and stabilization of nonlinear time-delay systems via a novel Takagi-Sugeno (T-S) fuzzy hybrid impulsive controller that explicitly accounts for both input delay and saturation, enhancing its practical applicability. The main contributions are threefold. First, a new impulse-time-related Lyapunov function (ITRLF) is constructed, which synergistically integrates the impulsive Razumikhin technique with an improved convex hull representation to handle saturation nonlinearities effectively. Second, sufficient conditions in the form of linear matrix inequalities (LMIs) are established to ensure local exponential stability. A key advantage of these conditions is that they depend only on the bounds of impulse delays and intervals, eliminating the restrictive requirement of a specific relationship between them, thus reducing conservatism. Third, novel LMI-based optimization algorithms are proposed to maximize the estimation of the region of attraction (ROA), effectively trading off computational complexity for significantly reduced conservatism compared to conventional methods. The effectiveness and advantages of the proposed approach are validated through numerical simulations using the MATLAB LMI toolbox.

This paper studies the stability and the intelligent event-triggered control (IETC) problem for the T-S fuzzy load frequency control (LFC) power system. Firstly, different methods are adopted to model various nonlinear factors in the LFC power system. Due to the influx of electric vehicles (EVs), the nonlinearity of the charging state characteristics needs to be considered in the LFC system modeling. Thus, the gain of EVs is treated as a time-varying function based on the change of the state of charge (SOC). Meanwhile, the nonlinearity of the valve position of the steam turbine is modeled using the T-S fuzzy theory. In summary, a T-S fuzzy LFC system model containing uncertainties is established. Secondly, an intelligent event-triggered mechanism (IETM) based on the grey wolf optimization (GWO) algorithm is developed. Furthermore, an IETC integrating the proposed IETM is established. The GWO algorithm is used to find the optimal triggering parameter of IETC to achieve both optimal bandwidth resource utilization and system state stability. In addition, the novel delay segmentation set-dependent Lyapunov-Krasovskii functionals (LKFs) are constructed to reduce the conservativeness of the results. Finally, the superiority of the proposed stability criterion and the effectiveness of the IETC are demonstrated by some case studies.