Image Super-Resolution (ISR) is employed to generate high-resolution images from low-resolution inputs. However, most current techniques for ISR encounter important challenges such as: (i) the assumption of sufficient training data availability, and (ii) the presumption that target image regions are complete without missing data. To address these practically important challenges, this study applies a lightweight approach termed Fuzzy Rough Feature Selection-based ANFIS Interpolation for ISR, especially on Martian imagery. Feature extraction algorithms are first applied to capture potentially significant features, and population-based search mechanisms are then utilised to perform effective feature selection (via extending the popular fuzzy-rough feature selection mechanism). The selected feature set is subsequently fed into an ANFIS interpolation model to perform the ISR task. Particularly, to handle the issue of sparse and incomplete data in dealing with Mars images, two adjacent ANFIS models are trained on nearby regions with sufficient data, positioning the model for the sparse region in between. Experimental studies conducted on Martian image datasets under both sufficient and sparse data conditions validate the effectiveness of the proposed approach, in overcoming the specific challenges faced by the task of ISR in extraterrestrial imaging scenarios.
Mission abort policy (MAP) has been widely studied in systems subject to random shocks. Most existing models assume either individual shocks degrading a single component or common shocks simultaneously impacting multiple components. A few recent studies address both types of shock processes, but are limited by restrictions on the timing of abort decisions. In this paper, we relax these timing restrictions by allowing aborting decisions to be made at any time during a mission performed by an asynchronous system with two heterogeneous components, offering more responsive decision-making. A new probabilistic modeling procedure is proposed for deriving the mission success probability and the expected cost of component losses, which are further used for calculating the normalized expected damage (NED). The optimal MAP that minimizes NED is then determined. A bi-sensor monitoring system is analyzed to illustrate the proposed model accommodating both individual and common shocks as well as flexible abort decision timing.
Wind affects both energy consumption and routing decisions in UAV delivery, making launch site selection and route planning strongly interdependent. This paper investigates their joint optimization under wind effects. Under a quasi-steady two-dimensional uniform wind-field assumption, wind speed, wind direction, payload status, battery capacity, and time windows are incorporated into the problem. A mixed-integer linear programing model is developed to minimize total system energy consumption by jointly determining launch site selection, task assignment, and flight routes. To solve the problem efficiently, an adaptive large neighborhood search (ALNS) algorithm is proposed. For 9-node and 20-node instances, ALNS obtains the same results as Gurobi. For the 50-node instance, it finds a feasible solution in about 105 s. Results further show that wind conditions influence energy consumption, delivery time, route structure, and launch site choice. These findings support the joint optimization of launch site selection and routing in UAV delivery.
Given a number of imprecise probability models, we aim at aggregating them into a joint one using an aggregation rule such as the conjunction, disjunction, convex mixture, Pareto, conjunction-disjunction or maximal consistent subsets rules. We focus on the problem of analysing if these operators are closed, in the sense that the output belongs to the same family as the inputs. Specifically, we analyse this problem for the family of comparative probabilities, 2-monotone capacities, probability intervals, belief functions, p-boxes and minitive measures.
This paper introduces a novel gradient-projection fixed-point iteration method for consistent approximation of pairwise comparison matrices in the Analytic Hierarchy Process (AHP), addressing challenges in multi-criteria decision-making such as inconsistent judgments and incomplete data. Leveraging Brouwer's fixed-point theorem, the method ensures existence of solutions, achieving convergence in 30-102 iterations with Frobenius norm errors of 0.2833-3.2964 for matrices of size n = 3,4,5, and demonstrating stable convergence for larger synthetic matrices up to n=50. Compared to logarithmic least squares method (LLSM) and eigenvector method (EM), it provides higher accuracy for both complete inconsistent and incomplete matrices, with runtimes of 0.04-0.13 s, as demonstrated in numerical tests and real-world applications. The approach's novelty lies in its robustness to initial vectors and inconsistencies, making it suitable for applications prioritizing precision. Limitations include higher iteration counts compared to LLSM and EM, which future work will address through adaptive step-size strategies.
The Paradox of Orthogenesis refers to the apparent tendency of biological and artificial systems to evolve toward greater complexity despite the lack of a clear mechanism to explain this directional growth. To investigate this question, we propose an evolutionary predator-prey simulation based on neural-network agents that learn through interaction and evolve via inheritance and mutation. The framework adopts a minimal setting, relying on local interactions and individual learning without reputation, trust, or intergenerational knowledge transfer. Results show that agents improve decision-making efficiency over time, but neural complexity does not increase indefinitely. Instead, network size and the number of active attributes converge to stable intermediate values, while decision times decrease and stabilize. No clear punctuated evolutionary dynamics are observed. These findings suggest that evolutionary pressure favours compact and efficient representations rather than continuous complexity growth, and that additional mechanisms may be required to sustain long-term increases in complexity.
Recommendation in tourism is an active research field aimed at enhancing traveler experiences through personalized suggestions of points of interest (POIs). While existing models have achieved promising results, many still struggle to capture the dynamic evolution of user preferences and to fully integrate contextual and practical travel constraints within the recommendation process. This paper proposes two main contributions. First, we introduce a hybrid EGCN-ODE recommendation model that combines a Graph Convolutional Network (GCN) with a Multi-Layer Perceptron (MLP) encoder to learn expressive user and item embeddings from raw attributes. To model the continuous evolution of preferences over time, we integrate an Ordinary Differential Equation (ODE) module that adapts to changing interests and contextual factors during a trip. Second, we present a routing and optimization component based on a Multi-Objective Ant Colony Optimization (MO-ANT) algorithm to solve the multi-objective route planning problem. This bio-inspired method efficiently generates feasible itineraries under constraints such as travel time and budget. Experimental results across multiple datasets demonstrate the effectiveness of our approach, yielding significant improvements over state-of-the-art baselines.
For the problem of efficient attribute reduction in fuzzy covering information systems under dynamic environments, traditional static methods face efficiency bottlenecks as they require re-executing global computations after data changes. To address this, this paper proposes an incremental attribute reduction method based on a hypergraph model. Firstly, a theoretical framework for attribute reduction based on the hypergraph model is presented. On this basis, the incremental update mechanism of relevant vertex and hyperedge metrics in the hypergraph model is systematically explored when the object set undergoes dynamic evolution through object addition. Building upon this mechanism, an efficient incremental attribute reduction algorithm is designed, which can perform local updates only on the changed parts, thereby avoiding global repetitive computations. Through comparative experiments on multiple public datasets, the effectiveness of the proposed algorithm is verified. The results demonstrate that, compared to classic static attribute reduction algorithms, the algorithm proposed in this paper significantly reduces the computation time while maintaining identical reduction results and preserving the classification ability, thereby exhibiting superior computational efficiency.
To further expand market share, some ride-hailing platforms with relatively high volume (H-platforms) may choose to introduce ride-hailing platforms with relatively low volume (L-platforms), thereby intensifying competition as more consumers are aware of L-platforms. To attract more consumers, L-platforms may choose to access H-platforms. This incurs additional commission costs. This paper develops a game theory model to explore whether the H-platform should introduce the L-platform and whether the L-platform should access the H-platform when considering the consumer's reference price. The results show that the perceived reference price level (PRPL) influences the operation mode choice of H-platform when the L-platform's service quality is low and the commission ratio is high. Furthermore, the PRPL influences the L-platform's operation mode choice when the L-platform's service quality is moderate (low) and the commission ratio is small (moderate). Additionally, our findings suggest that the H-platform introducing the L-platform always benefits consumers, but may hurt drivers.
Integrated energy systems (IES) research requires both domain depth and scientific logic, which current large language models lack. To bridge this gap, this study develops EnerAgentic, an IES domain research assistant. Using a Generator-Validator agent pipeline, we automatically built a high-quality Supervised Fine-Tuning (SFT) dataset of approximately 56,000 samples. Based on this, we propose a multi-agent framework that integrates reasoning, retrieval, and tool agents to autonomously handle complex interdisciplinary tasks. Evaluation results show that EnerAgentic comprehensively outperforms open-source models on general benchmarks. Crucially, in the domain-specific evaluation, EnerAgentic-RAG achieved an accuracy of 78.50%, significantly outperforming both its base model (45.00%) and showing highly competitive performance that approaches the GPT-5 (82.50%). This validates EnerAgentic's core capability to provide end-to-end support for complex data analysis, knowledge retrieval, and simulation modeling in the IES field.
Based on the fractal characteristics of Julia sets, this paper proposes a deep reinforcement learning framework for reference parameters adaptive synchronization of the Julia sets of models. This framework considers the unique characteristics of the fractal structure and designs a reward function that can not only quantify the global convergence error but also describes the fine differences in parameter matching. At the same time, the basic deep reinforcement learning algorithm is optimized by combining the fractals and parameter errors with the policy supervision reset mechanism. The feasibility of the method is demonstrated through multiple random training results from repeated experiments. The method demonstrates good robustness and stability in the synchronization process of the Julia sets of the classical complex quadratic function and the two-sector economic growth model. It has good generalization ability and provides a new possibility for the synchronization of complex dynamic systems.
This study explores dynamic pair trading strategies in the volatile cryptocurrency market, leveraging Hamilton-Jacobi-Bellman (HJB) optimal control frameworks alongside reinforcement learning (RL) techniques. The HJB Mean Reverting strategy achieved a cumulative return of 20.28%, providing intuitive, rule-based allocations albeit with moderate volatility. The Error Correction HJB variant, restricted to spread components without market index exposure, demonstrated exceptional stability - with a very low annualized volatility of 0.5% - and consistent, though modest, annualized returns of 3.65%. RL agents, constrained to allocate within spread components and enhanced through mean reversion and error correction mechanisms, delivered competitive cumulative returns ranging from 34% to 50%, with Sharpe ratios consistently above 1.2 and controlled drawdowns around 5-10%, illustrating robust adaptability to complex market dynamics. Together, these results highlight the complementary strengths of theoretical optimal control and data-driven reinforcement learning in achieving profitable and risk-managed trading strategies in rapidly evolving cryptocurrency markets.
This paper presents new fixed point results for xi-contractions in M-complete fuzzy metric spaces, extending conventional analytical tools that can be applied to nonlinear systems in fuzzy environments. An example is built to illustrate the applicability of the theoretical outcome. We also examine the usefulness of the theorem by analysing the development of the Chua attractor (in a fuzzy context) in a parametric way with a detailed numerical example. An additional application to RLC circuit model, expressed as a second-order differential equation, illustrates how such physical systems can be recast as fixed-point problems.
With rapid urbanization, traditional ground logistics systems are faced with growing congestion and emissions predicaments. Underground logistics system (ULS), as an emerging, efficient and low-carbon freight solution, has gained increasing attention. However, the absence of reliable demand continues to constrain its large-scale development and deployment. To accurately forecast ULS demand, this study develops an innovative framework (GRA-BP-LG) that integrates grey relational analysis (GRA), GM(1,1)-back propagation (BP) neural networks, and an improved logit-gravity model (LG). The proposed approach quantifies the impact of transport distance on allocation rate of ULS using an improved logit model, establishing a street-level matrix for freight flow computation. Finally, the results of the three cases achieve an average error of only 3% - 4% in Chongqing, Beijing and Shanghai, demonstrating the effectiveness and generalizability of the approach. This research provides a robust analytical tool for ULS demand forecasting and offers valuable insights for ULS planning and decision-making.
This paper addresses the challenges of Direct Recursive Generalized Predictive Control (DR-GPC) in Space Robot Teleoperation (SRT) under complex, uncertain and large time delays. Firstly, a rigorous DR-GPC mathematical model is developed, incorporating time delays into the control term and revising its original formulation. By adjusting the forward delay to match the control output's round-trip delay, DR-GPC is adapted to SRT systems with asymmetric forward and backward delays, whose rationality is rigorously explained via block diagram transformation in an event-driven framework. Furthermore, mathematical induction is employed to analyze DR-GPC's prediction accuracy, theoretically demonstrating that enhanced prediction efficiency does not compromise accuracy-filling a theoretical gap and elucidating DR-GPC's recursive prediction mechanism, which also lays a foundation for future improvements. Finally, simulations confirm DR-GPC's effectiveness in SRT system control.
The Dial-a-Ride Problem (DARP) focuses on designing vehicle routes and schedules to transport passengers between specified origins and destinations, with particular relevance to accessible transit systems. This study introduces a novel variant, termed the Heterogeneous DARP with Energy Replenishment Options (H-DARP-ERO), which simultaneously considers heterogeneous user demands with varying resource requirements, a heterogeneous electric vehicle fleet, and two energy replenishment options, namely recharging and battery swapping. A mixed-integer linear programming model is developed to jointly optimize passenger-vehicle assignment, routing, and energy replenishment decisions. To solve the problem, a Population-Guided Hybrid Genetic Algorithm (PG-HGA) is proposed, integrating customized encoding and decoding schemes, local search procedures, and adaptive population regulation mechanism to effectively balance exploration and exploitation. Experiments and a real-world case study demonstrate the efficiency and superiority of the proposed approach, and reveal how replenishment strategies and key parameters affect system performance, providing managerial insights for planning electric heterogeneous Dial-a-Ride systems.
