This paper addresses the fuzzy adaptive finite-time control problem for nonlinear systems with unknown nonlinearities and external disturbances. Fuzzy logic systems are employed to approximate the unknown nonlinear functions within the control framework. A dynamic surface control approach, integrated with first-order filters, is adopted to avoid the “complexity explosion” issue inherent in conventional backstepping design. By incorporating adaptive backstepping and finite-time stability theory, an adaptive fuzzy control scheme is developed, ensuring that all closed-loop signals remain bounded within finite time despite uncertainties and disturbances. Simulation results demonstrate the effectiveness and feasibility of the proposed method.
Short-Term load forecasting (STLF) is essential for the secure operation and effective planning of power systems. Accurate load forecasting not only ensures system reliability but also supports economic dispatch, optimal resource allocation, and risk management. However, due to the highly nonlinear and volatile nature of load patterns, as well as the influence of various external factors such as weather conditions, holidays, and human activities, single-step forecasting is often insufficient to meet practical requirements, making multi-step STLF particularly important for operational decision-making and planning. To address this challenge, this paper proposes a multi-population genetic programming (MPGP) approach to multi-step STLF. This approach evolves prediction models through multiple independent populations, with each population specifically trained and optimized for a fixed forecasting time step, thereby generating accurate predictions corresponding to each time step. Experimental results on real-world datasets demonstrate that the proposed approach can effectively capture complex load variation patterns, adapt to nonlinear dynamics, and maintain high accuracy across multiple steps, confirming the feasibility, robustness, and reliability of MPGP for multi-step STLF in power systems.
Dynamic characteristics are inherent in industrial processes. With increasing system complexity and the growing number of sensors, data-driven modeling faces challenges in terms of computational burden, and incipient faults become harder to detect promptly. To address these issues, this study proposes a vector autoregressive-based dynamic process modeling method that decomposes the monitoring space into dynamic and static components. An efficient sparse dynamic matrix estimation algorithm is further developed for offline model optimization, and a dissimilarity analysis-based approach is introduced for incipient-fault detection in the static component. Experiments on the Tennessee Eastman process benchmark model validate the effectiveness of the proposed method.
Relational database querying in Tibetan remains a markedly low-resourced endeavor. Existing Text-to-SQL models have only been evaluated on English or Chinese corpora, leaving Tibetan entirely unexplored. To address these issues, TSpider as the first Tibetan Text-to-SQL dataset is constructed by translating the CSpider corpus and comprising 11,840 Tibetan natural-language queries. Furthermore, the Input Feature Enhancement (IFE)-LLM framework is employed: its cross-encoder is adapted to Tibetan by integrating CINO, and the downstream open-source large language model is then prompted to generate executable SQLite SQL statements. Experimental evaluation on TSpider reveals that our framework attains 31.8
Drug target identification is a constraint multi-objective optimization problem with NP-hard characteristics, which aims to select a set of nodes from a gene network to control the transition between disease and normal state. However, existing related studies mainly focus on the evolutionary strategies, ignoring the relationship between nodes and constraint optimization. Therefore, this paper proposes a novel node importance method by analyzing the relationship between adjacent nodes with different connectivity differences. Then, the node importance is utilized to design a population initialization strategy to select better nodes. In experiments, this strategy is embedded into three algorithms to verify its effectiveness. Experimental results on three cancer genomic datasets show that the algorithm incorporating node importance knowledge achieves better performance compared to the original algorithm regarding two multiobjective optimization indicators and one biological significance indicator.
In fault diagnosis methods for neutral point clamped (NPC) three-level inverters based on signal analysis, the frequency domain analysis method lacks local time-frequency analysis capabilities, and the extracted fault features cannot locate the time of fault occurrence. Additionally, traditional time-frequency analysis methods have issues with window function selection and lack of window width adaptability. A method combining generalized S-transform with singular value decomposition (SVD) is proposed for fault feature extraction. Based on this, SVD is further applied to reduce data dimension, enabling feature extraction for open-circuit faults in three-level inverters. The extracted fault feature vectors are input into a limit learning machine for fault classification. Finally, the proposed algorithm is validated on a simulation platform. Simulation results demonstrate that fault features can be effectively extracted from the transient nonlinear current signals of a neutral-point-clamped inverter during an open-circuit condition, enabling diagnosis of single-tube and dual-tube open-circuit faults.
Ensuring the safety of aircraft engines during flight operations and mission execution is a matter of paramount importance. While deep learning-based approaches have shown promise in fault diagnosis, their effectiveness typically depends on the availability of large-scale and well-balanced labeled datasets. In real-world aviation scenarios, however, fault samples are often scarce and difficult to acquire, resulting in a significant class imbalance that degrades diagnostic performance. To better address the aforementioned challenges, this paper proposes an algorithmic framework for engine rotor fault diagnosis based on pre-trained transformers and generative adversarial networks, termed PT-Trans-WGAN-GP. The approach integrates selected layers of a pre-trained Transformer model into the discriminator and auxiliary classifier of a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP). This configuration guides the generator to produce high-quality synthetic samples for under-represented fault categories, thereby augmenting and rebalancing the training dataset. A diagnostic classifier is subsequently trained on the enhanced dataset to achieve accurate identification of multiple engine fault types. By effectively addressing the data imbalance issue, the proposed method yields a significant improvement in the performance of fault diagnosis under few-shot conditions, as evidenced by the experimental findings. The framework offers a promising new direction for intelligent health management in aircraft engines.
Coverage Path Planning (CPP) is a critical component for autonomous Unmanned Aerial Vehicle (UAV). However, classic methods struggle with non-convex environments, often yielding suboptimal paths or requiring complex area pre-processing. This paper introduces a novel Voronoi-based Full Coverage Path Planner (VFC-CPP) to generate path-length optimal trajectories for arbitrary polygonal areas. Our approach models the path as a dynamic mass-spring system guided by a potential field. The core innovation is a composite force model that integrates a Voronoi-based centroidal attraction force for uniform coverage, a spacing force to maintain ideal waypoint distance, and a path smoothing force to prevent clustering in sharp corners. VFC-CPP iteratively optimizes waypoint locations in continuous space, naturally adapting to complex boundaries without explicit decomposition. Comprehensive simulations demonstrate that VFC-CPP consistently generates shorter, more efficient paths than boustrophedon decomposition and grid-based TSP methods, especially in non-convex polygons, while achieving full coverage. Our work offers a robust and flexible solution for UAV coverage tasks in realistic, complex-shaped environments.
Accurate forecasting of future demand for civil aircraft is of great significance for formulating development strategies, allocating resources and planning future development in the aviation industry. However, civil aircraft demand is influenced by multiple complex nonlinear factors such as macroeconomic conditions and air transport volume, making it difficult for traditional single prediction models to fully capture its inherent regularities. In order to improve prediction accuracy, this paper proposes an adaptive weighting combination prediction model based on XGBoost and SVR. Historical data from 2013 to 2024 with multiple factors are utilized as predictive indicators and Bayesian optimization is employed to tune the hyperparameters of the XGBoost and SVR models separately. An adaptive weighting mechanism is then applied to combine the two prediction models, enabling accurate prediction of future civil aircraft demand. Experimental results show that, compared to standalone XGBoost or SVR models, the proposed model significantly reduces prediction errors and improves all evaluation metrics, fully demonstrating its effectiveness and superiority in civil aircraft demand prediction.
To reduce the risk of work-related musculoskeletal disorders (WMSDs) in aerospace component assembly, this study evaluates the performance of exoskeletons in typical component assembly scenarios. Surface electromyography (sEMG) quantitatively assesses upper-limb exoskeleton assistance and hand-vibration-damping exoskeletons, while a subjective satisfaction questionnaire evaluates human-exoskeleton interaction performance. Results demonstrate that upper-limb exoskeletons reduce upper-arm fatigue by up to 45.8
Natural language-based control commands for large-scale UAVs hold promise in applications such as voice-centric air traffic control (ATC) environments, single-operator management of multiple UAVs, and eVTOL taxi services. Large language models (LLMs) can enable command translation, knowledge base queries, and contextual understanding of diverse phrasings, while providing operational recommendations. However, challenges including speech recognition errors, harmful command inputs, and timeout issues hinder practical deployment. This paper proposes an LLM-based UAV interaction agent designed to parse commands from both UAV cockpits and ATC systems, while generating actionable operational advice. We evaluated multiple LLM variants and implemented engineering optimizations to improve recognition accuracy and response speed. Surveys with UAV stakeholders indicate that LLM-driven systems partially meet operational requirements but underscore the need for improvements in training methodologies and accuracy. These findings highlight the potential of LLMs in onboard UAV communication while emphasizing unresolved technical challenges requiring further refinement.
To address performance degradation caused by distribution shifts in vibration signals across operating conditions, we enhance the Domain-Adversarial Neural Network (DANN) with a lightweight multi-scale 1D convolutional feature extractor (MSCNN1D) and a Minimum Class Confusion (MCC) regularizer. MSCNN1D uses three parallel branches (kernel sizes 3/5/7) followed by 1 × 1 fusion, capturing short-and mid-range temporal patterns at low computational cost. Complementary to adversarial alignment, MCC is applied only to target-domain predictions and minimizes inter-class correlation via class-wise normalization, thereby reducing class mixing and encouraging clearer decision boundaries. On cross-condition transfers of the CWRU dataset, the proposed method improves target-domain accuracy over baseline DANN under the reported protocol while preserving real-time feasibility. Qualitative analyses (t-SNE, confusion matrices, and class-correlation heatmaps) indicate tighter and better-separated target-domain clusters. These results suggest that coupling multi-scale temporal representation with alignment-and-discriminability constraints yields robust cross-condition generalization under limited labels and varying operating settings.
This paper studies the optimal flight path planning of civil aircraft in the landing stage and carries out simulation verification. The six degrees of freedom model of civil aircraft is established and the landing path of civil aircraft is planned by serial convex optimization method. The optimization loss function is defined from the perspective of energy, the convexity of civil aircraft motion equation is completed by linearization, and the calculation amount of solution is reduced by Chebyshev collocation method. The flight path planing strategy is verified by simulation of Boeing aircraft model.
The Rapidly-exploring Random Tree (RRT) algorithm is one of the most popular algorithms for motion planning problems. However, the slow convergence of RRT* in searching the initial solution caused the low efficiency. Inspired by the mechanism of plant tropism, an environmental index-guided heuristic algorithm, the Heuristic-RRT*(H-RRT*) is proposed in this paper to improve the search efficiency: the growth direction of the search tree consists of a random term and a directional term; the directional term is determined by the location of the current node, the goal point the local microenvironment; the weights of the random and directional terms are determined by the environmental index; the search pattern is determined by the environmental index. Three search patterns are defined for various environments. The optimal solution can be obtained by combining the informed sampling strategy. The directional guidance of the H-RRT* algorithm can provide a better initial searching tree and a higher convergence rate. Compared to the RRT and GB-RRT algorithms, the H-RRT* algorithm demonstrated an initial solution speed improvement of 24.16
The proliferation of Internet-generated text, particularly in domains such as news and sentiment classification, has led to increased reliance on short texts characterized by brevity and high generalizability. While fine-tuning pre-trained models has shown success, these approaches typically require substantial labeled data and are unsuitable for few-shot settings. To address this, we propose a novel few-shot classification method enhanced with external knowledge. Our approach constructs a tailored prompt template integrated with input texts, reformulating classification as a cloze-style task. Additionally, external knowledge is leveraged to expand label words, with predictions mapped to original labels via a scoring mechanism. Experiments on sampled subsets of three datasets—THUC-News, Toutiao, and Chinese News Titles—demonstrate that our method significantly outperforms baseline models under 1-shot, 5-shot, and 10-shot conditions. Notably, in 1-shot settings, accuracy improvements average 10.1
In response to the problem that traditional feedback anomaly detection method based on fixed-threshold is prone to failure when facing continuous abnormal signal or changes in system states, an algorithm of servo feedback anomaly detection and suppression based on adaptive confidence is proposed in this paper. Firstly, through theoretical analysis, a control system model is established and feedback values are predicted according to the system state matrix and input signals. Then, a dynamic confidence evaluation framework based on the feedback error distribution is constructed to assess the credibility of feedback sampling values. Finally, through the confidence-weighted fusion strategy, the true feedback value of the system is obtained for subsequent feedback control. Simulation results show that the algorithm can effectively suppress abnormal feedback and significantly improve the system’s control performance under both short-term and long-term anomaly conditions; in the case of long-term abnormal feedback. Compared with traditional feedback anomaly detection algorithm, the algorithm proposed in this paper can effectively suppress the propagation of cumulative errors and improve the stability of the control system.
To solve the segmentation of convergence zones and acoustic shadow zones in marine acoustic Transmission Loss (TL) maps, this paper proposes an improved ResNet-34-based method. It introduces cross-stage feature fusion in the backbone, uses deformable convolution to fit slender bent region boundaries, and combines a global statistical feature module and joint loss function for optimization. Validated via the gray value difference and uniformity indices on a Bellhop simulation-based TL dataset, experimental results show it effectively enhances segmentation performance, offering an interpretable and reusable technical path for automated TL map analysis and related engineering applications.
This paper puts forward a reinforcement learning (RL)-based algorithm specifically tailored for the online solution of the optimal tracking problem in Markov jump systems (MJSs) where the system models are only partially known. To address the challenges posed by the partial unknownness and jump characteristics of MJSs, the research first elaborately designs a system augmentation method under the premise of decoupled subsystem conditions. On this basis, it conducts a rigorous theoretical demonstration to verify the equivalence between the optimal tracking problem of the original MJSs and the linear quadratic tracking (LQT) problem of the constructed coupled augmented systems (CASs). Leveraging the inherent jumping attributes of MJSs, the complex overall system is then effectively decomposed into multiple independent continuous-time linear subsystems, which simplifies the subsequent solution process. Afterwards, reinforcement learning techniques are employed to numerically solve the Riccati equation closely related to the augmented system, realizing the online computation of the optimal tracking strategy. Finally, through comprehensive numerical simulations, the tracking performance and convergence properties of the proposed online algorithm are extensively validated.
Aiming at the problems of traditional neuro-fuzzy models that rely on manually defined fuzzy rules and are prone to “rule explosion,” this paper proposes a Fuzzy Broad Learning System based on Fuzzy C-Means clustering (FCM-Fuzzy BLS). In the proposed framework, the FCM algorithm is employed to automatically generate fuzzy rules, thereby eliminating subjective rule design and enhancing model interpretability. The fuzzy membership degrees obtained through clustering serve as the basis for constructing the mapping between input features and fuzzy rules, ensuring adaptive rule generation. Meanwhile, the broad learning system (BLS) structure is integrated to enable efficient incremental learning and fast network expansion without retraining from scratch. By combining the fuzzy inference mechanism with BLS’s linear mapping and pseudo-inverse optimization, the proposed model achieves both high computational efficiency and robust generalization performance. Experimental results on multiple benchmark datasets demonstrate that the FCM-Fuzzy BLS not only alleviates the rule-explosion problem but also significantly improves accuracy, convergence speed, and scalability compared with conventional fuzzy neural networks.
This paper proposes a finite-time adaptive neural learning-based event-triggered robust tracking control for robotic manipulators with uncertain dynamics. First, a nominal system is introduced, and an appropriate cost function is designed to successfully transform the original robust control problem into an optimal control problem for the nominal system. Meanwhile, the optimal controller is synergistically designed with an event-triggered mechanism, significantly reducing the computational resources required for the robotic system. Next, a novel parameter update law is developed using adaptive control techniques to ensure the finite-time convergence of the neural network. Rigorous stability analysis proves the finite-time convergence of the parameters and the uniform ultimate boundedness of the tracking error. Finally, the effectiveness of the proposed control algorithm is validated through simulation experiments.
