This paper addresses the construction of observable and maximum distance profile convolutional codes over finite fields that exhibit good performance with some available decoding algorithms for convolutional codes. Our construction is based on the use of input/state/output representations and the invariance of certain properties of linear systems under various group actions. This framework allows us to systematically generate new convolutional codes from existing ones while preserving key decoding and distance properties.
This article considers the problem of fish monitoring in an underwater environment, where many problems might occur, including occlusion, pose changes, and complexity of the scene. Recognizing fish behavior is very important to develop various types of technologies able to provide more precise estimations and monitoring of fish populations in a long term. In this paper, we propose a novel method for underwater fish monitoring (shape modeling and pose estimation). Two main aspects of underwater image processing will be studied: classification and localisation. Additionally, we extract key point features from fish patterns. The fish position and motion are not sufficient features to avoid scene problems. Skeleton extraction could offer us a large range of additional information. It models an object as a set of points of a certain manifold. The 3-dimensional fish pose, along the track of its 3D motion, could depend on curve segments of the underlying manifold. Faster reccurent conventional neural networks (faster R-CNNs) will be used to extract the fish skeleton in different poses. Also, a 3-dimensional trajectory of multiple fish will be derived using a Kalman filter based on the previous feature matching process. The simulation is made for live fish in a fish tank. Experimental results show that our method outperforms relevant models in terms of precision, achieving a minimal accuracy of 94.2%.
In this work, a velocity tracking control strategy is presented using passivity methodology for a permanent magnet synchronous machine with uniformly distributed windings, which is known as a brushless direct current (BLDC) motor. The trapezoidal components of back electromotive force (EMF) are here modeled by using a novel differentiable approximation. The proposed control scheme is constructed by employing only measurements of the shaft position and stator currents. The use of a second-order filter makes it possible to solve the velocity tracking control problem without measuring the actual rotor velocity. A formal stability analysis of the closed-loop system consisting of the motor, the controller, and the second-order filter is presented using partial stability theory. By applying partial stability, the relevant variables are shown to converge, while the behavior of the rest of the variables that are not relevant to the control objective can be disregarded, which makes the theoretical analysis simpler. A numerical simulation is constructed to verify the applicability of the control law. The results show that all the variables of the BLDC motor are below the nominal values of the machine. Therefore, the physical implementation of the controller is feasible.
The global stability of fractional multi-inputs multi-outputs continuous-time nonlinear feedback systems with interval matrices of positive linear parts and application to electrical circuits is investigated. New sufficient conditions for the global stability of fractional nonlinear systems are given. The new stability conditions are applied to nonlinear electrical circuits and demonstrated on a simple example of a fractional nonlinear feedback system with a positive linear part.
This paper presents a nonparametric interval forecasting method that combines circular block bootstrap resampling with complexity-invariant K-nearest-neighbor time-series prediction. Prediction intervals are obtained directly from bootstrap-resampled training series, thereby preserving temporal dependence while accounting for forecast uncertainty. Under weak dependence and local stability assumptions, asymptotic validity of the resulting prediction intervals is established. The proposed method is evaluated using twelve time-series datasets drawn from economic, environmental, industrial, and energy applications. Empirical performance is compared with seasonal autoregressive integrated moving average models and long short-term memory neural networks using the mean absolute percentage error, empirical coverage probability, and interval score. The results show that the proposed approach yields prediction intervals of moderate width with competitive forecasting accuracy across most datasets, while empirical coverage remains close to the nominal level. Mild undercoverage is observed in short samples, attributable to limited data availability and fixed tuning parameters.
The paper is devoted to the development of algorithms for the odd-time discrete Fourier transform (type II DFT, or simply DFT-II). It presents efficient computational solutions related to the implementation of a small-sized DFT-II for input sequences of lengths 3, 4, 5, 6, 7, and 8. The derivation of each algorithm is described in detail, and computational complexity estimates are provided. The developed algorithms are implemented on field-programmable gate array (FPGA) platforms such as Spartan 3 and Spartan 6, demonstrating the advantages of their implementation in hardware environments where performance and resource utilization are critical. Key performance indicators such as the number of multiplications and additions as well as FPGA resource utilization are evaluated. The values of the maximum operating frequencies achieved are given. The results show a significant improvement in computational performance compared to the direct matrix-vector product, which proves the effectiveness of the obtained solutions.
This paper presents and tests several task scheduling methods in a real-time multitasking system based on the thread interleaving mechanism. The original configurable multicore time-predictable system architecture for multitasking is briefly discussed. The essential requirement of the system is the predictability of tasks, regardless of their number and when they are initialized. The paper focuses on the appropriate configuration of core interleaving registers, which determine the order and frequency of execution of individual tasks. This study develops various heuristic algorithms based on genetic programming and task execution rate analysis, and implements these in the Python language and PROLOG. The experiments are conducted with different task scenarios and system work requirements: with minimization of resources, energy, and operating frequency. A special task analyzer is used to evaluate the quality of the resulting system configuration. The results obtained are tested on a real hardware structure implemented in an FPGA chip. The proposed approach can be a useful tool for configuring real-time multitasking systems.
This paper proposes an event-triggered neural network control approach to address the voltage control issue in distribution networks under false data injection (FDI) attacks. Firstly, a mathematical model of voltage deviation in distribution networks considering the impact of FDI attacks is established to accurately represent the dynamic behavior of the attacked system. To optimize the utilization of communication resources within the network, an adaptive event-triggered mechanism is designed, which can dynamically adjust the triggering conditions based on the system state, effectively reducing unnecessary communication instances. On this basis, an event-triggered voltage control (VC) system model is established. To effectively mitigate voltage over limit caused by FDI attacks, an adaptive neural network controller is designed, which can compensate for the attack signals and keep the voltage within the allowable range. By combining Lyapunov-Krasovskii stability theory with linear matrix inequality (LMI) techniques, the stability of the system is analyzed, and sufficient conditions for ensuring it are derived. Finally, simulation results demonstrate that this method can not only effectively resist FDI attacks, but also significantly reduce the communication burden while ensuring system stability.
The increasing complexity and dynamism of modern networks pose significant challenges for effective fault management. Temporal network faults, characterized by their evolving nature and cascading effects, are particularly difficult to diagnose. Traditional root cause analysis (RCA) methods often struggle with the high dimensionality, non-linearity, and temporal dependencies inherent in network monitoring data. This paper proposes a novel framework, ESN4TRCA—an echo state network for temporal root cause analysis, for identifying the root causes of temporal network faults. ESN4TRCA leverages the inherent capabilities of echo state networks (ESNs), a paradigm of reservoir computing, in modeling complex temporal dynamics with remarkably low training overhead. We formulate the temporal RCA problem as a sequence classification task, where sequences of multivariate key performance indicators (KPIs) and alarm data are mapped to their underlying root causes. The proposed framework encompasses modules for data preprocessing, ESN model construction specifically tailored for heterogeneous network fault data, and a robust inference mechanism. We introduce specific mathematical formulations for the leaky-integrator reservoir dynamics and the output weight training via ridge regression, optimized for the RCA context. Comprehensive experiments are conducted on both a synthetic dataset generated using the NS-3 network simulator and a real-world public dataset. The results demonstrate that ESN4TRCA significantly outperforms state-ofthe-art RCA methods, including traditional machine learning approaches and other recurrent neural network architectures like LSTMs and GRUs, in terms of accuracy, F1-score, and robustness to noise, while maintaining superior computational efficiency. The study highlights the potential of ESNs as a powerful and practical tool for advanced automated network fault management.
This study explores the development of two control strategies based on the sliding mode approach for quad rotorcraft trajectory tracking. A dynamic and an integral-type controller are designed to ensure that the sliding surface reaches zero within a finite time, resulting in a PD-like structure. Since this structure aids in determining the gains of robust PD controllers, it is utilized for comparative analysis. To account for uncertain dynamics and external disturbances, this work proposes an offline linear matrix inequality (LMI) algorithm that guarantees ultimate uniform stability for both sliding mode controllers. The primary advantage of the proposed LMI-based strategy is its ability to simplify the implementation of a sliding mode controller in complex systems, overcoming challenges associated with their intricate tuning process. Since the proposed algorithm applies to all three controllers, it facilitates the identification of the most effective one based on the system’s dynamic response. A comparative analysis based on error criteria is performed through numerical simulations to validate the effectiveness of the proposed strategies. In addition, a second comparative analysis is conducted between two widely used robust control strategies from the literature and the proposed ISMC. Finally, the effectiveness of the designed algorithm is evaluated using a complex reference trajectory featuring high maneuverability and high-speed flight.
With the rise of the digital era, handwriting examination has become crucial for identity verification and document provenance. However, determining whether samples from different texts are written by the same person remains challenging. The challenge is greater in few-shot settings, where data are scarce and writing styles vary widely. Traditional methods often lack sufficient accuracy and robustness. We propose a dual-branch Siamese network for handwriting verification. It fuses attention mechanisms with a feature-bank matching strategy. This design improves adaptation and generalization under few-shot conditions. It also suppresses background noise and emphasizes key writing traits. We evaluate the method on CCSbC, the mixed Chinese-English hard-pen dataset named MDC, and CCD-CQU. The model attains high accuracy on multi-class few-shot classification tasks. It shows strong robustness and adaptability. With data augmentation and feature optimization, it could deliver more efficient handwriting identification in real-world applications.
In the paper, new, fractional order, variable parameter discrete transfer function models of an elementary inertial plant are proposed. One uses a time variable quasi time constant while the other employs, a variable time quasi constant and variable fractional order. Both models apply the variable fractional order backward difference to approximate the fractional operator. Both variable parameters are described by discrete, bounded functions. The accuracy and stability of the models considered are discussed. Theoretical results are validated by simulations. The proposed models can be applied to the modeling of various physical phenomena, where the constant parameter transfer function is not enough.
Design of offset-free model predictive control (MPC) with a linear state-space process model is discussed in the paper, for deterministic constant or asymptotically constant external and internal (modeling errors) disturbances. The three existing, established methods assuring offset-free control are briefly compared and relations between them are recalled. The main result of the paper is a new, more flexible but still simple formula defining unmeasured disturbance estimates in the model-based prediction error method (abbreviated as MPE). The new formulation eliminates the only deficiency of this approach, i.e., being possibly highly sensitive to noise. Equivalence of the proposed technique with the augmented process-and-disturbance model method (AM for short) is proved, indicating how to define arbitrary choice of disturbance matrices in the equivalent formulation of the AM approach. With the new formulation, the MPE method seems to be a competitive, best design choice. Theoretical results are validated and illustrated by extensive simulation results of a MIMO control problem, with unmeasured disturbances from the mentioned class and also with noises added. The capability of the proposed method to tune sensitivity to noises of the MPC control system is shown.
The production and supply chain system is a complex nonlinear network comprising suppliers, manufacturers, distributors, retailers, and other entities. The collaborative evolution process for such a system is vitally important, especially when the system undergoes design changes. In this paper, the data-driven inventory consensus problem for production and supply chain systems is studied while considering design changes. Firstly, the production and supply chain system is modeled as a multi-agent one and the collaborative process of the production and supply chain is formulated as the system topology structure switching. Secondly, a digraph data-driven consensus control protocol is proposed to achieve inventory level consensus tracking with respect to a predefined reference trajectory in production and supply chain systems. Thirdly, when a design change occurs, the saturation function is introduced for the protocol design, which helps address the productivity limitation problem. Lastly, with the help of the contracting mapping principle, the convergence analysis is carried out, demonstrating the effectiveness of the proposed data-driven consensus approach through a numerical simulation example.
This paper presents a trajectory tracking error detection method for autonomous vehicles (AVs) via an optimal control scheme, where two online learning algorithms are designed: (i) an adaptive learning algorithm (ALA) and (ii) a finite-time adaptive learning algorithm (FTALA). We first construct an error dynamic system by combining the kinematic equation of AVs and an ideal tracking trajectory. To realize the adaptive optimal control for AVs, the ALA is designed, which provides us with an online solution by resolving the derived Hamilton-Jacobi-Bellman (HJB) equation. Then, the FTALA is presented, which further relaxes the requirement of the system dynamics, i.e., the system drift is not required. The adaptive critic and control action in both online learning algorithms continuously and simultaneously interact, eliminating the need for iterative steps. This approach also avoids the use of an actor neural network (NN) and an initial stabilizing control policy. Moreover, the finite-time convergence can be ensured via adopting a sliding mode technique in the FTALA. Finally, both online learning algorithms are applied to control AVs, and the simulations show their effectiveness and practicality feasibility. The FTALA reduces the time-accumulated tracking error and the cumulative control effort by about 39% and 57% in lane-keeping, and by about 38% and 57% in the S-curve, respectively.
The multiple hypothesis testing problem occurs when a number of individual hypothesis tests are considered simultaneously. The larger the number of inferences (tests) made, the more likely erroneous inferences become. Several classical statistical techniques have been developed to address this problem. Unfortunately, very often, in studies that aim to examine the significance of multiple effects (for example, correlations), the effect of multiple testing is ignored. This leads to an inflated number of significant findings (e.g., correlation coefficients). On the other hand, in the case of a large number of tests about relatively small effects, and thus a large family of inferences, the power of individual tests decreases rapidly when classical procedures for controlling multiplicity are applied, often resulting in too few significant findings. In this paper, we propose an alternative and powerful approach to face the problem of testing multiple hypotheses and develop a new MCPerm method for testing the significance of multiple correlations. The proposed solution is definitely more effective than Holm’s, which proved to be more conservative in the presence of several dependencies and tended to indicate fewer of those.
The concept of an autoencoder is a key element of modern machine learning methods. It is used, e.g., to compress the input data into a lower-dimensional representation, which is later employed in other machine learning algorithms. In the area of quantum computing application methods, quantum machine learning is also developing dynamically, with the concept of a quantum autoencoder also present. In the article, we discuss a variant of building a quantum autoencoder based on a quantum convolutional network. The proposed autoencoder is characterized by an architecture based on adjacent quantum gates, which is especially important for the near-future noisy intermediate-scale hardware implementation of such a type of quantum circuits. The proposed circuits are built with an elementary set of gates, i.e., controlled negation gates and rotation gates. The described architecture of a quantum autoencoder properly uses the quantum phenomenon called quantum exponential capacitys, i.e., a linear number of qubits which allow encoding an exponential amount of classical information. In our case, for n qubit, it is possible to encode a classical gray coloured image with dimensionality 2n x 2n. The conducted numerical experiments on a set image of handwritten characters and letters show that, for a small number of parameters, the quantum autoencoder offers a reconstruction quality comparable to the currently used classical autoencoders. An important assumption is the omission of additional dimensionality reduction techniques, e.g., PCA, for the preparation of classical data. In the discussed experiment, the data after reconstruction can be read directly from the quantum register, through a series of quantum register measurements.
Collecting large amounts of real-world data can be expensive, time-consuming, or even impossible in some cases. This situation can be particularly acute for IMU-based gait-analysis data due to participant fatigue, changing walking surfaces, and instrumentation positioning. These factors can provoke changes in the characteristics (drift or distribution shift) of data collected over long durations or on different days, making analysis difficult. Ideally, one would prefer a larger data set whose distributional characteristics do not change. Data augmentation can help address this limitation by creating more training data from what is already available. It is a technique used to artificially increase the amount of data available for training machine learning models. It works by creating new data points from existing data through various modifications. We present a new data-augmentation technique based on modelling three types of common sensor disturbances: alignments, vibrations and drift. The new approach is compared to three state-of-the-art methods, and validated on three different data sets: two from publicly accessible repositories, and one from our own laboratory with 100 participants and 30 gait cycles per person. Improvements in classification accuracy are obtained for each data set: 35%, 90% and 21%, respectively. The experiments show that the use of data augmentation has a positive impact on the metrics of the gait biometric system. It enables increased efficiency in identifying subjects from a limited sample of data and does not require a problematic data acquisition process.
Liquid-based cytology (LBC) is a widely used diagnostic tool for cervical cancer diagnosis. However, the accuracy and efficiency of LBC-based cervical cancer classification are still limited due to the lack of standardized, scalable, and objective cytological assessment protocols. To address these gaps, this study develops and evaluates a machine learning framework that integrates various feature extraction techniques, feature selection methods, and machine learning classifiers to improve cervical cancer detection. The results demonstrate that handcrafted and local binary pattern features achieve the best overall performance, with the SVM, gradient boosting and histogram-based gradient buffering reaching a 95.92% accuracy, highlighting the strength of combining morphological and texture descriptors to maximize their discriminative potential. Moreover, we provide a systematic comparison of different classification pipelines, offering insights into the feasibility of hybrid approaches, particularly in resource-constrained medical environments. The promising results obtained in this study highlight the potential impact of machine learning in modern medical diagnostics, providing a clinically relevant, highly accurate, and efficient classification method for LBC slides.
The aim of this work is to find a compromise between the accuracy of reproducing the behaviour of a nominal (Wiener-type) object examined in a laboratory under noise-free conditions and its robustness to intentional external attacks disrupting the input signal. By linearizing the model at the operating points and replacing the computationally expensive minimax optimization criterion with a simpler one, we construct a technique that leads to models robust to adversarial attacks of bounded intensity. Simulation experiments demonstrate the robustness of the obtained models against adversarial disruptions, highlighting the method’s potential applications in fields requiring high resilience, such as control systems and safety-critical environments.
