Experiments and modeling on thermal performance evaluation of standalone and M-cycle based desiccant air-conditioning systems

This study aims to evaluate the performance of standalone and Maisotsenko-cycle based desiccant airconditioning systems with silica gel through experimental and explainable artificial intelligence methods. In order to examine the desiccant systems' performance, deep neural networks have been constructed based on the optimal hyper-parameters provided by the Bayesian surrogacy. The influential factors, such as regeneration air temperature and relative humidity, primary air temperature and relative humidity, among other system features like regeneration temperature and humidity, are chosen as the inputs using feature engineering and correlation heat maps. The results indicated that with the right framework and training data, the proposed model can forecast the performance of the standalone and M-cycle-based desiccant air conditioning systems with a great accuracy (i.e. R2 = 0.99). In addition, regeneration temperature and humidity, and inlet regeneration air temperature are considered to be the primary parameters that affect the thermal performance of Maisotsenko-cycle and stand-alone desiccant systems. The regeneration conditions are shown to have a positive correlation with the primary and regeneration outlet temperatures while having a negative correlation with the primary and regeneration outlet relative humidity. The primary inlet relative humidity has the greatest impact on the primary outlet relative humidity while the regeneration inlet temperature has the highest impact on the regeneration outlet relative humidity of the system. In addition, the primary inlet relative humidity is found to have positive correlations with the outlet regeneration and primary relative humidity and outlet regeneration temperature. Also, the primary air inlet temperature is found to be directly associated with the outlet primary relative humidity and temperature for the considered systems.