شناسايي سيستم‌هاي غيرخطي با استفاده از شبكه موجك المان فازي مبتني بر الگوريتم بهينه‌سازي ازدحام ذرات بهبوديافته

Identification of nonlinear systems is a fundamental topic in engineering, as having an accurate model of system behavior is essential for analysis, prediction, controller design, fault diagnosis, an‎d performance optimization. Since many real-world systems exhibit nonlinear dynamics, complex temporal dependencies, an‎d strong coupling among variables, employing an intelligent structure capable of modeling such complexities is of great importance. In this research, a smart modeling framework called the Fuzzy Element Wavelet Network (FEWN) is utilized for nonlinear system identification. By combining the strengths of fuzzy systems, element-based neural networks, an‎d wavelet functions, the FEWN structure provides a flexible an‎d powerful tool for representing highly nonlinear behaviors. To optimize the parameters of the FEWN model, the Particle Swarm Optimization (PSO) algorithm is employed in both its classical an‎d improved forms. In the enhanced version, dynamic inertia weighting, Latin Hypercube Sampling (LHS)–based initialization, local search around the global best solution, an‎d a partial particle reinitialization mechanism are incorporated. These strategies help maintain population diversity, prevent premature convergence, an‎d improve the accuracy of parameter tuning. To eva‎luate the performance of the FEWN model, several benchmark applications are considered, including nonlinear function approximation, prediction of the Mackey–Glass time series, identification of real Box-Jenkins gas furnace data, an‎d modeling of a nonlinear dynamic system. The results demonstrate that the FEWN model provides accurate approximation, prediction, an‎d identification capabilities, an‎d in most cases, the improved PSO algorithm significantly reduces modeling error an‎d enhances convergence stability. Furthermore, the capability of the FEWN model in identifying the Twin-Rotor Multi-Input Multi-Output nonlinear system is investigated. Due to the unavailability of real hardware, input–output data are generated using the nonlinear model of the system in Simulink, excited by a PRBS signal. Subsequently, for each system output—namely the pitch angle an‎d yaw angle—a multi-input single-output FEWN model is trained separately. Simulation results indicate that the FEWN model approach successfully identifies the dynamic behavior of both axes with acceptable accuracy. Additionally, the improved PSO algorithm leads to further error reduction an‎d enhanced identification performance. Overall, integrating the FEWN model with the improved PSO algorithm provides an effective an‎d efficient methodology for nonlinear system identification, particularly for complex systems such as the twin-rotor setup.

فريد شيخ الاسلام

مريم ذكري

ايمان ايزدي نجف آبادي , حامد جلالي بيدگلي