Abstract
The control of the battery-thermal-management-system (BTMS) is key to prevent catastrophic events and to ensure long lifespans of the batteries. Nonetheless, to achieve a high-quality control of BTMS, several technical challenges must be faced: safe and homogeneous control in a multi element system with just one actuator, limited computational resources, and energy consumption restrictions. To address those challenges and restrictions, we propose a surrogate BTMS control model consisting of a classification machine-learning model that defines the optimum cooling-heating power of the actuator according to several temperature measurements. The la-belled-data required to build the control model is generated from a simulation environment that integrates model-predictive-control and linear optimization concepts. As a result, a controller that optimally controls the actuator with multi-input temperature signals in a multi-objective optimization problem is constructed. This paper benchmarks the response of the proposal using different classification machine-learning models and compares them with the responses of a state diagram controller and a PID controller. The results show that the proposed surrogate model has 35% less energy consumption than the evaluated state diagram, and 60% less energy consumption than a traditional PID controller, while dealing with multi-input and multi-objective systems.