Influence of the process and regeneration air conditions on the thermal performance of various desiccant air-conditioning systems

The need for desiccant cooling systems has grown in response to the need to protect the environment and reduce the costs associated with low-grade thermal energy. The desiccant air-conditioning (DAC) system is chosen based on a variety of factors such as air flow rate, regeneration temperature and relative humidity, process air temperature and relative humidity, among others. In order to overcome the limitations posed by the existing methods such as experimental and simulation, the aim of this research is to perform parameter sensitivity analysis and evaluate thermal performance of standalone and Maisotsenko (M−cycle) desiccant air-conditioning systems by using Bayesian optimized surrogate models coupled with deep learning. Furthermore, explainable artificial intelligence (XAI) is used to assess the individual impact of each input parameter on the individual outputs. In order to assess the impact of a single input on the output parameters of DAC systems, a parameter sensitivity analysis was conducted by varying the input while keeping the others constant. To gain a deeper understanding of the model, we employed XAI techniques using the Deep SHAP library to interpret the deep neural network (DNN) model's predictions. This approach helped to highlight the individual contributions of each input variable, specifically process and regeneration air conditions, in predicting the thermal performance of both standalone and M−cycle based DAC systems. It is found that the intake regeneration air temperature has the greatest influence on the thermal performance of DAC systems followed by the relative humidity of the inlet regeneration air as is evident by their R² values (0.96 and 0.97, respectively). This is also proven by the other metrics, such as MAE and MSE. The model considering all the input parameters demonstrated the best predictability for the DAC systems (R² = 0.996). The proposed method provides a simple and dependable way for assessing the thermal performance of both standalone and M−cycle based DAC systems.