مدل‌سازي سيستم مرطوب ساز غشايي با استفاده ازتكنيك‌هاي هوش مصنوعي

This thesis investigates the intelligent modeling of insulated membrane humidifier systems (parallel an‎d spiral configurations) in proton exchange membrane fuel cells (PEMFC). Focusing on the challenge of membrane humidity management to enhance efficiency, stability, an‎d lifespan, it employs machine learning techniques including linear models (Linear Regression, Ridge, Lasso, an‎d their polynomial extensions) an‎d nonlinear models (SVR, GPR, MLP, an‎d Ran‎dom Forest). The study starts with limited real data (11 samples per system) an‎d progresses using synthetic data generated via the K-Nearest Neighbors (KNN) method. Single-output an‎d multi-output strategies are compared across four scenarios. Results indicate that the combination of synthetic data with the multi-output approach yields the best performance, with low RMSE (0.2 to 2.5), R² near 1, an‎d negligible overfitting. Top-performing models include Ran‎dom Forest, MLP, an‎d SVR, while regularized models like Ridge_Poly2 an‎d Lasso_Poly2 are more stable with limited data. The primary objective is to develop data-driven models to reduce simulation time an‎d optimize humidifier performance for sustainable energy applications

5

159258