Hybrid Approach Mixing Mathematical Programming and Machine Learning Techniques for Thermal Grid Systems

We focus on thermal grid systems which aim to control a number of air conditioning systems as a unit for achieving highly energy-efficient buildings. In order to realize real-time control of them, we propose hybrid systems in which artificial neural networks are involved and it is trained by using mathematical programming techniques determining the outputs of the air conditioning machines in multi-periods. Through some numerical experiments with the actual settings of the thermal grid systems, the potential of our proposed method is examined.