A machine learning and Particle Swarm Optimization approach for desiccant wheel modeling and performance prediction

Abstract

Accurate modeling of desiccant wheels (DWs) is critical for the design and optimization of energy-efficient dehumidification systems. This study presents a novel approach for predicting DW performance by coupling machine learning (ML) models with Particle Swarm Optimization (PSO) for hyperparameter tuning. To validate the effectiveness of this metaheuristic approach, the performance of the PSO-optimized models was rigorously benchmarked against counterparts tuned using conventional Bayesian Optimization (BO). Four distinct ML models, including Artificial Neural Network (ANN), k-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Regressor (SVR), were developed to predict the process air outlet temperature (Tp,out) and humidity ratio (ωp,out). The models were trained and validated on a comprehensive dataset, uniquely expanded to include experimental data from low-humidity and low-temperature deep dehumidification conditions. The results demonstrate that the PSO-optimized Artificial Neural Network (PSO-ANN) model provides superior predictive accuracy. For the process outlet temperature, the PSO-ANN model achieved a Coefficient of Determination (R2) of 0.9985 and a Root Mean Square Error (RMSE) of 0.3204 °C. For the outlet humidity ratio, it achieved an R2 of 0.9984 and a RMSE of 0.1497 g/kg. Furthermore, a SHAP (SHapley Additive exPlanations) analysis confirmed that the model’s predictions are physically consistent and interpretable. The developed high-fidelity model serves as a robust and reliable tool for the advanced analysis and design of desiccant air conditioning systems across a wide range of operational scenarios