A model-based approach for a control strategy of a charge air cooling concept in an ejector refrigeration cycle

An efficient thermal management in vehicles can reduce fuel consumption or improve the electrical range. Optimized control strategies adapting to various load cases can reduce the energy consumption of the cooling system and keep components in efficient operating temperature ranges. Current cooling control strategies use performance maps or rules, which are time- and cost-consuming to develop due to a high manual workload and the necessity of vehicle prototypes. In this paper, a highly automatized process is proposed to create control strategies with machine learning methods and simulation models. A new tool is introduced, which can couple Python code with Dymola to extend simulation models by calibration and optimization features. Simplified control models are created with the dataset of optimized control settings using machine learning implementations for a multivariant linear and polynomial regression as well as a decision tree and a random forest classification. The performance of the different control models is compared on a dynamic drive cycle in a co-simulation.