Swarm intelligence optimization methods have steadily gained popularity as a solution to multi-objective optimization issues in recent years. Their study has garnered a lot of attention since multi-objective optimization problems have a hard high-dimensional goal space. The black-winged kite optimization algorithm still suffers from the imbalance between global search and local development capabilities, and it is prone to local optimization even though it combines Cauchy mutation to enhance the algorithm’s optimization ability. The heuristic optimization algorithm of the black-winged kite fused with osprey (OCBKA), which initializes the population by logistic chaotic mapping and fuses the osprey optimization algorithm to improve the search performance of the algorithm, is proposed as a means of enhancing the search ability of the black-winged kite algorithm (BKA). By using numerical comparisons between the CEC2005 and CEC2021 benchmark functions, along with other swarm intelligence optimization methods and the solutions to three engineering optimization problems, the upgraded strategy’s efficacy is confirmed. Based on numerical experiment findings, the revised OCBKA is very competitive because it can handle complicated engineering optimization problems with a high convergence accuracy and quick convergence time when compared to other comparable algorithms.

Adaptive spiral flight and multi-strategy fusion are the foundations of a new FOX optimization algorithm that aims to address the drawbacks of the original method, including weak starting individual ergodicity, low diversity, and an easy way to slip into local optimum. In order to enhance the population, inertial weight is added along with Levy flight and variable spiral strategy once the population is initialized using a tent chaotic map. To begin the process of implementing the method, the fox population position is initialized using the created Tent chaotic map in order to provide more ergodic and varied individual beginning locations. To improve the quality of the solution, the inertial weight is added in the second place. The fox random walk mode is then updated using a variable spiral position updating approach. Subsequently, the algorithm’s global and local searches are balanced, and the Levy flying method and greedy approach are incorporated to update the fox location. The enhanced FOX optimization technique is then thoroughly contrasted with various swarm intelligence algorithms using engineering application optimization issues and the CEC2017 benchmark test functions. According to the simulation findings, there have been notable advancements in the convergence speed, accuracy, and stability, as well as the jumping out of the local optimum, of the upgraded FOX optimization algorithm.

TianQin is a geocentric space-based gravitational wave detecting mission that relies on three drag-free spacecraft with a mean inter-spacecraft distance of approximately 1.7 × 10⁵ km. Considering the efficiency requirement of orbit optimisation, we apply a gradient descent method based on variational gradient calculation and the damped Newton iterative method to find the most stable solution of the constellation near the given initial position and velocity of the spacecraft. The efficiency was improved and the number of orbital integrals required for gradient calculation was significantly reduced by over 83%. Based on this optimisation method, we analysed the influence of perturbation on the stability of the TianQin constellation, and Moon’s three-body perturbation was the main factor. In addition, the influence of the orbit elements on the optimisation results was studied. The influence of the initial mean anomaly can be ignored, and the stability of the optimised constellation decreases with an increase in the semi-major axis. Working in the ‘3 months on +3 months off’ observation scheme for 2 years, the optimisation result is the ideal direction when the normal direction of the orbital plane of the TianQin constellation points to RXJ0806.3 + 1527.

5G Mobile Adhoc Networks (5G-MANETs) are a popular and agile solution for data transmission in local contexts while maintaining communication with remote entities via 5G. These characteristics have established 5G-MANETs as versatile communication infrastructures for deploying contextual applications, leveraging physical proximity while exploiting the possibilities of the Internet. As a result, there is growing interest in exploring the potential of these networks and their performance in real-world scenarios. However, the management and monitoring of 5G-MANETs are challenging due to their inherent characteristics, such as highly variable topology, unstable connections, energy consumption of individual devices, message routing, and occasional inability to connect to 5G. Considering these challenges, the proposed work aims to address real-time monitoring of 5G-MANETs using a connection-aware Digital Twin (DT). The approach provides two main functions: offering a live virtual representation of the network, even in scenarios where multiple nodes lack 5G connectivity, and estimating the performance of the infrastructure, enabling the specification of customized conditions. To achieve this, a communication architecture is proposed, analyzing its components and defining the involved processes. The DT is implemented and evaluated in a laboratory setting, assessing its accuracy in representing the physical network under varying conditions of topology and Internet availability. The results show 100% accuracy for the DT in fully connected topologies, with ultra-low latency averaging under 80 ms, and suitable performance in partially connected contexts, with latency averages below 3000 ms.

5G Mobile Adhoc Networks (5G-MANETs) are a popular and agile solution for data transmission in local contexts, while maintaining communication with remote entities via 5G. These characteristics have established 5G-MANETs as versatile communication infrastructures for deploying contextual applications, leveraging physical proximity while exploiting the possibilities of the Internet. As a result, there is growing interest in exploring the potential of these networks and their performance in real-world scenarios. However, the management and monitoring of 5G-MANETs is challenging due to their inherent characteristics, such as highly variable topology, unstable connections, energy consumption of individual devices, message routing, and occasional inability to connect to 5G. Considering these challenges, the proposed work aims to address real-time monitoring of 5G-MANETs using a connection-aware Digital Twin (DT). The approach provides two main functions: offering a live virtual representation of the network, even in scenarios where multiple nodes lack 5G connectivity, and estimating the performance of the infrastructure, enabling the specification of customized conditions. To achieve this, a communication architecture is proposed, analyzing its components and defining the involved processes. The DT is implemented and evaluated in a laboratory setting, assessing its accuracy in representing the physical network under varying conditions of topology and Internet availability. The results show 100% accuracy for the DT in fully connected topologies, with ultra-low latency averaging under 80 ms, and suitable performance in partially connected contexts, with latency averages below 3000 ms.

Parameter-efficient fine-tuning methods such as Low-Rank Adaptation (LoRA) have become indispensable for adapting large language models under resource constraints. However, standard LoRA applies a uniform rank across all layers, overlooking the heterogeneous causal contributions of different transformer components to downstream task performance. While differentiated rank allocation offers a promising remedy, existing importance metrics rely on correlational signals that may not reliably guide architectural decisions, and principled causal intervention techniques are typically computationally prohibitive for iterative layer-wise analysis. This paper introduces LoRA-Cau, a low-overhead causal intervention framework that enables efficient, causality-driven rank allocation and learning rate configuration. Our method quantifies layer-wise causal importance by applying minimal perturbations to LoRA matrices and measuring the resulting shift in prediction confidence. A three-fold optimization strategy—representative downsampling, single-pass activation caching, and localized forward reuse—reduces the measurement overhead by over 60%. The normalized causal scores are then used to warm-start fine-tuning with differentiated ranks and inversely scaled learning rates. Extensive experiments on RoBERTabase (GLUE benchmark) and Llama-2-7B (MT-Bench and GSM8K) demonstrate that LoRA-Cau achieves strong and competitive performance across tasks. On MT-Bench, it surpasses standard LoRA and several existing variants; on the challenging GSM8K mathematical reasoning task, it delivers a notable +10.12-point improvement over standard LoRA, recovering a considerable portion of the full fine-tuning performance. Ablation studies and gradient flow analyses further reveal that causally-informed rank allocation enhances gradient signals in critical layers, while differentiated learning rates stabilize optimization dynamics. Our findings underscore the value of efficient causal measurement in guiding resource allocation for parameter-efficient adaptation.