Modeling of zeolite-based desiccant wheel with machine learning

The objective of this study is to model the desiccant wheel, which is frequently used in desiccant air-conditioning systems, using various machine learning and linear regression methods. The desiccant wheel with LT3 material was analyzed under 32,975 different input conditions using the manufacturer's software, considering 8 independent variables. The dataset generated was utilized to develop a total of 96 models, comprising 7 multiple linear regression models, 45 decision tree models, 12 support vector machine models, and 32 multilayer perceptron models. The models were evaluated based on various statistical metrics. Among the multiple linear regression models, the 3rd Degree Cross model demonstrated the best performance. It achieved RMSE values of 0.6018 ℃ and 0.1774 g/kg for process air outlet temperature and humidity ratio, with R2 values of 0.9934 and 0.9972, respectively. The DT-28 model, the top-performing decision tree model, achieved RMSE values of 0.7196 ℃ and 0.1903 g/kg for temperature and humidity ratio, with R2 values of 0.9977 and 0.9992, respectively. The SVM-5 model, the best among support vector machine models, achieved RMSE values of 0.3780 ℃ and 0.1082 g/kg for temperature and humidity ratio, with R2 values of 0.9974 and 0.9989, respectively. The MLP-8 model, which provided the best results among the developed models, achieved RMSE values of 0.2059 ℃ and 0.0693 g/kg for temperature and humidity ratio, with R2 values of 0.9994 and 0.9997, respectively.