The less-is-more approach applied to metaheuristic variable neighborhood search combines simplicity and effectiveness in a unique way. With a minimal volume of source code, one can quickly obtain very good solutions. However, the time spent on algorithm implementation may grow significantly on attempts to fit its elementary components, such as, for example, selection of an initial solution, local search moves, or perturbation moves. This study aims to support such a process through a comparison of several possibilities of element composition, including new ideas for elementary moves. The systematic comparative analysis includes 36 versions of this metaheuristic tested on instances of two problems: the traveling salesman one, since it is a model problem commonly used in real-world applications, and that of DNA sequencing, which is a basic and well-studied problem in bioinformatics. The results give basis for conclusions about the compositions of procedures that perform better than others, in the sense of solution quality and computation time.
For multi-agent systems with missing iteration batches and randomly varying lengths, a data-driven adaptive iterative learning control strategy based on neural networks is proposed. Considering the situation where the iteration batches of the system vary randomly with the number of iterations, first, an iterative batch missing fault model that follows a random probability distribution is designed. Secondly, based on the radial basis function neural network method, a novel data-driven adaptive iterative learning consistency control algorithm is proposed. The stability analysis of the system is strictly performed using the Lyapunov function, and the sufficient conditions for consistency of multi-agent systems are derived through relevant mathematical derivations. This enables multi-agent systems to achieve state consistency even when there are missing faults in the iterative batches. Finally, the feasibility of the algorithm is verified through numerical simulations in a multi-agent system.
Aiming at the parallel task scheduling problem (P |sizej|Cmax), which requires multiple machines to work collaboratively, this paper proposes an improved tabu search algorithm to overcome the shortcoming of traditional methods being prone to falling into local optima. Firstly, a hybrid neighborhood structure combining swap and insertion operations is designed to enhance search diversity and local optimization capabilities. Secondly, a dynamic tabu length adjustment mechanism is introduced to adaptively change the cycle of tabu objects according to the search process, thereby achieving a balance between breadth and depth of search. Finally, a simplified aspiration criterion is proposed to avoid the loss of high-quality solutions caused by excessive tabu. Extensive computational experiments on randomly generated test instances show that the designed algorithm achieves highly competitive performance compared with the original tabu search algorithm and the OR-Tools solver, while significantly improving computational efficiency. In particular, the proposed method reduces the average Cmax by approximately 10.6% on extra-large-scale instances, demonstrating superior solution quality and robustness.
The design of modern high-tech equipment (pumps, compressors, turbines) is inevitably related to the need to ensure vibration reliability, especially at high shaft rotation speeds, which is particularly relevant for flexible rotors. Therefore, the research aims to provide vibration reliability of energy-intensive rotary machines by developing a hybrid parametric identification approach that leverages analytical dependencies in rotor dynamics, numerical simulations to study rotor behavior, and artificial intelligence systems. While implementing the proposed solution, the Monte Carlo method and Moore-Penrose pseudoinversion are also applied. As a result, modified linear and nonlinear parameter identification approaches are developed. They are based on an improved finite element model that accounts for the analytical dependence of bearing stiffness on rotor speed, with the parameters evaluated numerically. The corresponding algorithms were implemented in the PTC MathCAD 14 computer algebra system and the Python 3.13 programming language, using the authors' program code and an artificial neural network. The developed methods and algorithms were proven in the case study of a natural gas compressor station's multistage centrifugal compressor, S325 GC2-650/6-56M12. As a result, stiffness parameters of bearing supports were identified with a 0.2-0.7% accuracy for the first two critical frequencies compared with the available data from the accelerating-balancing stand "Schenck" with a vacuum chamber. The proposed parameter identification approach can also be extended to evaluate stiffness parameters for high-precision bearing supports with speed-dependent stiffness in multistage centrifugal pumps, turbochargers, turbines, and turbopump units.
Anomaly detection using machine learning (ML) has become a critical research focus point in the modern digital era. Although recent ML-based models demonstrate strong detection capabilities, their performance is often limited by the vast volume and high dimensionality of data, leading to reduced accuracy, lower detection rates, and increased learning complexity. To address these challenges, this study proposes a novel ensemble learning framework that combines random forest (RF), extreme gradient boosting (XGBoost), adaptive boosting (AdaBoost), category boosting (CatBoost), and a light gradient boosting machine (LGBM) with a hybrid feature selection strategy. An improved particle swarm optimization (IPSO) algorithm enhanced with elimination and opposition-based learning is developed and integrated with the chi-square method (chi-IPSO) to optimize feature selection. Additionally, the synthetic minority oversampling technique (SMOTE) is employed to handle class imbalance in the datasets. The proposed framework is evaluated using two benchmark datasets, UNSW-NB15 and CICIDS 2017. Experimental results reveal that the RF algorithm with chi-IPSO achieves superior performance, attaining an accuracy of 94.58% on UNSW-NB15 and 99.70% on CICIDS 2017. These results illustrate the effectiveness of the proposed model, showing notable betterment over existing state-of-the-art anomaly detection approaches.
