Hybrid model based refrigerant charge fault estimation for the data centre air conditioning system

Abstract

Accurate refrigerant charge fault estimation is important to ensure the efficient operation of air conditioning systems. This paper presents a novel hybrid model based refrigerant charge estimation approach. Firstly, an improved gray box model is presented, which integrates the key characteristic variables of subcooling temperature, superheat temperature, quality and pressure drop. Secondly, three extra variables having highest maximal information coefficients with the prediction residual are used to extend the gray box, and the robust machine learning model is developed using the gradient boosting decision tree algorithm. Then, a hybrid model is presented by combining the improved gray box and machine learning models. Finally, the prediction and generalization capacities of the proposed models under various operation conditions are validated using the experimental data. The results show that the hybrid charge fault estimation model has the best performance. Its overall prediction and generalization MREs are 2.53% and 3.09%, respectively.