Combinatorial optimization (CO) concerns decision-making over discrete solution spaces under constraints and lies at the core of many complex systems across science, engineering, and society. Yet current approaches to solving CO problems still rely heavily on expert-designed solvers and manual tuning. Large language models (LLMs), trained on massive corpora, offer a promising new avenue for enhancing CO solution methods with stronger capabilities for generalization and adaptation in complex or previously unseen scenarios. Some existing surveys provide high-level taxonomies without examining solver-level methods, while others focus narrowly on specific perspectives. To fill this gap, we present a methodology-oriented synthesis of recent LLM-assisted CO work, covering three mainstream solution paradigms: mathematical programming, heuristics, and end-to-end solvers. We first develop a hierarchical taxonomy that groups existing methods by solution paradigm and, within each paradigm, further classifies them by the workflow stage and the interaction pattern through which LLMs engage with the solver. We then formalize the workflow of each paradigm, clarify the functional role of LLMs, and analyze the core techniques in key modules. We further summarize the problem domains studied in the literature and discuss representative applications. Finally, we identify current limitations and outline future directions in architecture, methodology, and problem scope.
This paper focuses on the hybrid flow shop scheduling problem with sequence-dependent setup time (HFSP-SDST) with minimizing the makespan. To address this problem, this paper designs a hybrid migrating birds optimization (HMBO) algorithm that integrates migrating birds optimization (MBO) algorithm, variable neighborhood descent search (VND) algorithm, problem-based local search (LS) algorithm, and constraint programming (CP) model. Specifically, HMBO consists of three primary stages. The first stage employs a hybrid algorithm (MBOVND) that integrates MBO and VND with permutation encoding and decoding. Because the solution space of permutation encoding and decoding cannot cover the full solutions of HFSP-SDST, LS and CP are used to enlarge the solution space of MBOVND in the second and third stages respectively. Specifically, LS algorithm is used in the second stage to explore the solutions that are not in the solution space of MBOVND, and CP model is used in the third phase to further enlarge the solution space of MBOVND and LS algorithm. The efficacy of the proposed VND, LS, CP, and HMBO are verified. Experimental results demonstrate that VND, LS, and CP are effective to improve the solving ability of MBO, and HMBO improves 89 out of the 120 best-known solutions for the benchmark instances.
BRT-lane-sharing, which allows regular buses to operate on Bus Rapid Transit (BRT) lanes without disrupting scheduled BRT operations, has gained prominence. It is helpful to increase the utilization of BRT lanes and enhance the efficiency of the bus transit system. However, the current approaches for designing bus transit networks and setting frequencies do not incorporate BRT-lane-sharing, thereby neglecting its potential benefits, including improved speeds, easier transfers, and cost savings. This paper proposes a pioneering study for the Bus Transit Network Design and Frequency Setting (BTNDFS) problem incorporating BRT-lane-sharing. A novel road network description is presented, specifically tailored to accommodate BRT-lane-sharing through the introduction of BRT nodes and BRT-lane arcs. Furthermore, a bi-level model is built for the BTNDFS problem incorporating BRT-lane-sharing. To solve this model, a Priority-Based Genetic Algorithm (PBGA) is proposed, in which a priority-based chromosome is defined, whilst priority-based crossover and mutation operators are devised. Experimental results on the standard Mandl's benchmark instances indicate that the PBGA outperforms other metaheuristic approaches, with outcomes closely approximating optimal solutions. Further experiments are carried out on a real-world network featuring BRT-lane-sharing in the city of Linyi. The results show that the proposed model and the PBGA can reduce costs for passengers and operators, while simultaneously increasing the utilization of BRT lanes.
In aerodynamic design, high-quality multiblock mesh generation serves as the cornerstone for accurate lift-drag analysis of wings. However, conventional computational fluid dynamics (CFD) workflows offered by industrial software reveals critical limitations in labor-intensive manual operations. While automatic meshing in 2D wing cross-section is well-established, its extension to 3D wing meshing faces unresolved challenges, including excessive block generation, complex feature alignment, and incomplete simulation-ready pipelines. This paper presents an automatic multiblock mesh generation framework specifically targeting these commercial software shortcomings for 3D wing analysis. Our methodology addresses three persistent challenges: (1) a parameterization-based tip surface meshing method with optimized cell quality, (2) an extrusion-based wing body meshing approach ensuring precise feature alignment at joints and leading edges, and (3) an end-to-end pipeline from geometric pre-processing to CFD-ready multiblock mesh generation. Experiments demonstrate the framework's ability to produce high-quality meshes with minimal manual intervention, streamlining the workflow for aerodynamic simulations.
Driven by the “dual-carbon” goals, new energy power generation is gradually dominant, yet distributed energy's intermittency and volatility challenges power system operation. Focusing on potential flexible resources in power systems and large-scale grid connection of new energy vehicles (EVs), this paper innovatively introduces EVs penetration rate and the weight coefficient for encouraging new energy power generation, and constructs a virtual power plants (VPP) optimal scheduling model considering such penetration rate and flexible electrical and thermal loads. To achieve coordinated optimization of flexible loads on both supply and user sides, with the goal of reducing the total operating cost of the VPP, an electrical load model with load shifting, transfer, and reduction functions is built. Case studies analyze the impact of different EVs penetration rates on its optimal dispatching, verified by multiple scenario comparisons. The research results show that as the penetration rate of EVs increases to 50%, the total operating cost of the VPP continues to decrease, with new energy output increasing by 9.2% and gas turbine output decreasing by 15.4%. This reduces reliance on traditional fossil fuels and accelerates the green and low-carbon transformation of energy systems, but also strengthens the substitution effect of new energy power generation for traditional energy based power generation.
Ensuring robust visual perception under adverse weather conditions is a critical prerequisite for the safety and reliability of autonomous driving systems. Camera-based perception, despite providing rich semantic information, suffers significant degradation in rain, fog, snow, and low-light scenarios, thus presenting a key bottleneck to all-weather, all-scenario autonomy. This paper systematically reviews the evolution of image restoration techniques for road traffic environments, tracing the progression from traditional physical-model based approaches to data-driven deep learning paradigms. The domain gap between synthetic training datasets and real-world applications is identified. Misalignment between restoration metrics and downstream perception performance is also highlighted, revealing that improvements in image quality do not necessarily translate into enhanced perception outcomes. The review advocates for a paradigm shift toward perception-oriented, semantically-guided, and unsupervised domain-adaptive joint frameworks, which integrates image restoration and perception in an end-to-end manner, utilizes semantic priors for region-specific enhancement, and employs self-supervised signals to fully exploit unlabeled real-world data. At last, this work concludes with prospects for future research, emphasizing the necessity of bridging the restoration-perception gap, extending solutions to more complex degradations and tasks, and leveraging foundation models as intelligent controllers for robust visual systems in autonomous driving.
Online content generation enables automatic and adaptive creation of diverse and playable game content for maximizing player experience or testing Artificial Intelligence (AI) algorithms. Multiple diversity metrics have been formulated on different content facets in the literature, while some of them conflict with one another. Existing work addresses this multi-dimensional diversity nature by converting those metrics into one term that is further used to direct the training of content generators. However, each generator is trained to meet the preference specified by the weights and fails to fully interpret the relationships among these metrics or provide different trade-offs. This paper proposes a multi-objective procedural content generation via reinforcement learning to train a set of generators that create diverse game content in an online manner while balancing the trade-off between multiple diversity metrics with playability as a constraint. Our framework is compared with state-of-the-art approaches on the commonly used Mario-AI benchmark. Results show that our framework is capable of increasing the diversity of the generator distribution while accelerating the convergence during the early stages of model training. Our approach enables researchers, designers, and practitioners to gain a better understanding of the relationship among conflicting diversity metrics, allowing them to generate content more efficiently and accurately tailored to specific needs.
Glass insulator defect detection faces challenges such as noise data interference, small defect targets, and low accuracy. Current most methods aimed at enhancing the accuracy of insulator defect detection inevitably increase model complexity, requiring additional computational resources. To address this issue, this paper proposes a Lightweight Multi-scale Attention Convolutional Network (LMACNet). A general Spatial-Channel Cross Attention (SCCA) is proposed to perform cross-spatial learning on multi-scale features. SCCA-Mobile Inverted Bottleneck Convolution (SCCA-MBConv) is designed for lightweight feature extraction. Additionally, we improved the Cross Stage Partial Bottleneck with 2 convolutions (C2f) module to enhance multi-scale feature fusion and high-level semantic feature learning. Specifically, an auxiliary network is designed to train the lightweight model, alleviating the information bottleneck and ensuring sufficient gradient flow. Extensive experiments were conducted on the public Vietnam Public Merged dataset of Broken Glass Insulator (VPMBGI) dataset and a custom dataset, the results demonstrate that LMACNet achieves detection performance comparable to other related detection models, while having a parameter count of only 2.54 & times;10(6) and requiring 3.01 & times;10(9) operations. LMACNet exhibits efficiency and practicality.
Remanufacturing contributes to achieving economical, environmental, and social sustainability, and one of its main steps is disassembly aiming to acquire a set of recyclable and reusable components from end-of-life products. This research considers a multi-objective multi-product disassembly sequence planning problem under uncertain circumstances to realize a trade-off among economic, environmental, and social sustainability. Firstly, a multi-objective chance-constrained programming model is formulized to achieve maximal disassembly profit and minimal noise pollution while satisfying energy consumption requirements and obeying various complex product structures. Secondly, a multi-objective group teaching optimization algorithm combining a stochastic simulation approach is particularly devised to handle the problem. In the designed approach, problem-specific encoding and decoding methods are employed to represent and produce feasible solutions. The stochastic simulation approach is utilized to assess the feasibility and performance of the obtained solutions under uncertain environments. Rank and crowding distance approaches are introduced to realize ability grouping, namely, dividing the population into two groups. Precedence preserving crossover and mutation operators are separately utilized on the two groups to achieve population evolution, and an adaptive local search method is developed to enhance exploitation. Thirdly, comparison experiments on some real-world test problems with different scales are carried out. Through dissecting the experimental results with three performance metrics, it can be observed that the devised approach outperforms its competitors by 9.39%-10.00%, 11.37%-59.86%, and 2.36%-7.73% regarding performance, respectively. The experimental results demonstrate the efficiency and excellence of the devised approach in providing high-quality disassembly schemes for managers and engineers.
Copy number variation (CNV) is a remarkable manifestation of genomic structural variations that affect human health. However, CNV detection in low coverage and low purity data is one of the challenging issues. To fill this gap, a hybrid algorithm combining an improved whale optimization algorithm (IWOA) and backpropagation (BP) neural networks (hereafter called IWOABP) is developed for CNV detection. First, to enhance the precision of detection, the detectable categories for the gain and loss are respectively expanded to two types, where gain is divided into tand_gain and inte_gain, and loss is divided into hemi_loss and homo_loss. Then, IWOA is introduced to tune the weights and bias values of BP neural network, which can improve the BP neural network abilities to jump out of the local optimums. Next, to ensure the population diversity and the uniform distribution of solutions, a pooling mechanism and a migration search strategy are designed. In addition, to balance the exploitation and exploration abilities, three position update strategies based on an adaptive inertia-weight are used. Finally, to evaluate the detection performance of IWOABP, seven state-of-the-art detection methods are chosen to make detailed comparisons with the proposed algorithm. The results show that IWOABP has outstanding performance in sensitivity, precision, and Fl-score using both simulated and real data.
Within the context of Massive Open Online Courses (MOOCs), the application of Human-in-the-loop systems is becoming increasingly prevalent. Given the issue of high dropout rates in MOOCs, developing accurate dropout prediction models is essential to improve educational outcomes and ensure the sustainability of the platform. Researchers usually use interactive information between students and platforms to build predictive models. However, existing methods have the problem that the single construction of behavioral characteristics leads to misclassification of special student groups. Therefore, after in-depth analysis of the behavioral data of MOOC learners, this paper constructs the characteristics of learning activity based on the two dimensions of learning time and number of clicks to reduce the misclassification of special groups. On this basis, we propose a hybrid neural network model (CNN-LSTMAE) to model and predict dropout behavior. CNN-LSTMAE uses the Convolutional Neural Network (CNN) to extract local features of students' behaviors, and uses the Long-Short-Term Memory Autoencoder (LSTMAE) module to perform a high-level abstract representation of local features. Experimental results on the KDD Cup 2015 dataset show that the proposed model has better predictive performance than the baseline methods.
Artificial electric field algorithm (AEFA) is a meta-heuristic optimization technique recently developed that has demonstrated efficacy in scientific research and engineering applications. However, it exhibits limitations such as premature convergence jointly with constrained search capability, particularly in complex optimization scenarios. To rectify these defects, this article proposes an enhanced artificial electric field algorithm (EAEFA) by incorporating a hybrid position updating strategy. To fully utilize all agents in the population, simultaneously improving the exploration and exploitation capabilities of AEFA, EAEFA divides its swarm into two groups. The elite group uses Levy flight for better solution precision, while the non-elite group combines a spiral update approach and the intrinsic update method in AEFA for a more thorough exploration of the search space. This combination effectively balances the exploration and exploitation trade-off to facilitate a robust search. Meanwhile, a stagnation interrupt strategy is employed if EAEFA confronts stagnation or premature convergence. Whereafter, the proposed EAEFA is evaluated over thirteen classical benchmark test functions and CEC2014 benchmark test suits. Numerical results unequivocally demonstrate that the proposed algorithm has exceptional performance surpassing recent variants of AEFA and eight well-known existing metaheuristic methods. Moreover, EAEFA's success in solving four real-world engineering design problems showcases its practical applicability.
Reentrant is widespread in many manufacturing scenarios even if there is a few concerns in literature. This study explores a distributed reentrant flow shop scheduling problem with sequence-dependent setup time (DRFSP-SDST). The goal is to minimize the maximum factory completion time (Makespan). Initially, the mathematical model of DRFSP-SDST is formulated by considering the sequence-dependent setup time. Second, four meta-heuristics, including iterated greedy (IG), artificial bee colony (ABC), Jaya, and particle swarm optimization (PSO) algorithm, are used and their variants are proposed for solving the concerned problems. Third, to enhance the performance of the algorithms, five local search operators are designed based on the nature of the problems. Then, two algorithms for reinforcement learning, Q-learning and state-action-reward-state-action (Sarsa), are integrated into the iterative process to select high-quality local search strategies. Finally, the effectiveness of the proposed improvement strategies is evaluated through comprehensive numerical experiments on 90 instances. The performance of the proposed algorithms is further verified through the Freidman test. The ABC algorithm with Sarsa-based local search exhibits the highest competitiveness for solving the DRFSP-SDST, according to the experimental findings and debates.
This article addresses distributed consensus and optimization problem for multiple Euler-Lagrange systems by proposing a novel distributed control scheme, which is motivated by the practical problems including source seeking and communication relay through coordination of multiple mobile robots. The proposed scheme ensures that each Euler-Lagrange agent reaches the optimal solution for the team's objective function in a prescribed time, overcoming the limitations of finite-time and fixed-time distributed optimization algorithms in estimating convergence time. The control algorithm employs a distributed estimator to calculate the gradient vector of the team's global cost function. Subsequently, distributed prescribed-time controllers for each Euler-Lagrange individual are designed based on the gradient descent method, enabling all the Euler-Lagrange agents' states achieve optimal solution for the team's objective function. The system's prescribed-time convergence is examined via theoretical analysis using Lyapunov stability theory, and simulation experiments are conducted to confirm the controller's effectiveness.
The online scheduling of electric vehicle (EV) charging stations faces significant challenges due to the volatility of real-time electricity prices and the uncertainty of charging demand caused by stochastic EV arrivals. Furthermore, the participation degree of station operators is pivotal to the effective execution of the grid's demand response. To effectively capture the complex variability of electricity prices, a novel ensemble prediction-probability mapping framework is proposed to probabilistically characterize real-time peak-valley price features. Unlike traditional methods that rely on predetermined price prediction curves, the framework quantifies price uncertainty through probability interval values, offering a more robust and adaptive decision-making process. To respond to these real-time electricity signals, a risk-aware dynamic price subsidy mechanism is designed to provide differentiated incentives, aligning station operators' economic interests with grid load-balancing objectives. A hybrid ambiguity set that integrates both moment-based and distance-based information is developed to accurately characterize the uncertain charging demand. Additionally, a distribution robust optimization (DRO) model with this ambiguity set is proposed to mitigate conservatism in worst-case scenarios. The model is reformulated as a finite-dimensional convex program to ensure that the set remains sufficiently expressive without becoming intractable. Experimental results demonstrate real-time response in dynamic EV-grid interaction scenarios, validating the robustness and cost-risk balance of the proposed method against price and demand uncertainties.
The rapid advancement in circuit manufacturing technology has enabled the reduction in circuit width. Consequently, a single wafer can now accommodate more chips, allowing for smaller lot sizes to meet diverse customer needs. Manufacturers aim to use a single production line to simultaneously produce multiple types of wafers for various customizations. However, this presents two challenges: (1) Different types of wafers require different processing routes, necessitating a highly adaptable scheduling algorithm for producing various wafer types in the same production line; (2) producing multiple wafer types involves several state transitions, whereas existing research primarily focuses on single-state transitions. To address these challenges, we propose an end-to-end scheduling method for single-arm cluster tools handling multiple wafer types (ESM-SMWT). First, ESM-SMWT employs a genetic algorithm to optimize the sequence in which wafers enter the cluster tool, improving the utilization of shared and parallel processing modules. Next, integer programming is used to achieve end-to-end scheduling, from the processing of the first wafer at the start of a lot to the completion of the last wafer. Additionally, the new robotic arm strategy we propose significantly reduces the number of robotic arm activities. Through theoretical proofs and extensive experiments, the practical effectiveness of ESM-SMWT is validated.
The research on Optimal Power Flow (OPF) problems in power systems has undergone significant development, yet a significant gap persists in effectively addressing the multi-objective nature of these problems, particularly within the context of modern power systems characterized by increasing renewable energy integration. This study posits that the incorporation of the Non-Uniform Rational B-Spline (NURBS) curve concept into the Multi-Objective Particle Swarm Optimization (MOPSO) algorithm can significantly enhance its performance in resolving OPF problems. This paper introduces a novel methodology, termed NURBS-MOPSO, which capitalizes on the geometric structure of the Pareto front to facilitate co-evolution among subpopulations of particles. This approach aims to optimize the OPF problem within power systems, with the objectives of minimizing generation costs, reducing pollution emissions, balancing line loads, and maintaining node voltage stability simultaneously. Comprehensive simulations are conducted on the IEEE 30- bus and 57-bus systems, demonstrating the superior performance of the proposed algorithm.
Hydrogen production from wind-solar generation is of great importance for consuming renewable energy and it is meeting industrial hydrogen demand. In this paper, the modelling of the off-grid hydrogen production system from wind-solar generation and the simulation of its operating characteristics are investigated. Firstly, the network architecture and hierarchical control architecture of the off-grid hydrogen generation system are designed with the goal of efficiently utilising wind-solar generation output. Then, the components of the off-grid hydrogen generation system are characterised and modelled. Finally, the operating characteristics of the hydrogen production system under three operating conditions, such as hydrogen system startup, wind power fluctuation, and electrolyzer partial failure are simulated and analyzed, revealing the ability of the alkaline electrolytic water hydrogen production system to respond to the fluctuation of wind and solar power.
Recently, deep learning-based fingerprint localization has attracted significant interest due to its simplicity in implementation and effectiveness in complex multipath environments, especially for the Internet of Things (IoT) devices in multiple-input multiple-output (MIMO)-orthogonal frequency-division multiplexing (OFDM) system. However, the huge amount of training data collection has become a challenge, which increases the labor burden of fingerprint localization heavily and hinders its large-scale implementation. In this paper, we propose a novel fingerprint localization system, termed as SiamResNet, which can be trained only on the radio map by contrastive self-supervised learning without the need for any other additional data. To be more specific, we first model the fingerprint localization problem as a dictionary look-up task. Subsequently, a channel fingerprint capturing the multipath angle and delay of wireless propagation is introduced, which exhibits excellent uniqueness, stability, and distinguishability. Meanwhile, we propose the corresponding data augmentation strategy to ensure data diversity when generating the training data from the radio map. Thus, the cost of data collection for training can be significantly reduced. Lastly, the Siamese architecture-based SiamResNet is applied for location estimation, which can comprehensively extract the features of fingerprints and accurately compare the similarity of any fingerprint to the radio map in the representation space. The performance of the proposed localization method is validated through extensive simulations with a ray-tracing channel model, which demonstrates promising localization accuracy for our SiamResNet with reduced training costs.
The present paper attempts to design an adaptive multi-model predictive control strategy for strongly nonlinear or switched systems with various operating points. The proposed control system guarantees the feasibility and the asymptotic stability of the closed-loop system, considering various challenges such as inherent uncertainties in the local models constituting the model bank, limited prediction/control horizons, and set point changes. To this end, four fundamental challenges in this area, namely guaranteeing feasibility throughout the region assigned to each subspace, ensuring asymptotic stability in each subspace considering the inherent uncertainties of the local models, guaranteeing feasibility and asymptotic stability during changes in the set point and switching between the subspaces, are addressed. By introducing transferring mode concept, this paper presents a novel method for guaranteeing the feasibility and stability of the switched systems without the need for increasing the prediction/control horizons or decreasing the size of the feasibility region. The proposed control structure uses a supervisor algorithm along with a soft-switching technique. The supervisor algorithm is responsible for determining the suitable local model/controller pair, determining the operational mode of the control system, managing the soft switching, and specifying the control objectives in accordance with the defined set point. The efficiency of the proposed control strategy is demonstrated by simulating a Continuous Stirred Tank Reactor (CSTR) as the controlled system. Based on the results, the proposed controller is able to guarantee the feasibility and stability of highly nonlinear and switched systems in a wide operating region under set point changes and uncertainties in the local models.
