A reinforcement learning based energy optimization approach for household fridges

Abstract

The use of machine learning algorithms for the control of schedulable loads like Heating, ventilation, and air conditioning (HVAC), illumination, dryers and irrigation systems to optimize the use of RES and increase energy saving has obtained remarkable results in the last years. However, in the residential sector of tropical countries where HVAC systems are not necessary, these loads represent only a small percentage of the total energy consumption. In order to achieve a significant impact on energy savings and promote the use of RES, other residential loads must be taken into account in tropical countries. In the case of Colombia, for example, fridges account for 24% of residential energy consumption. This research proposes the use of RL for the development of a fridge energy management system capable of minimizing energy consumption and optimizing the use of RES for cooling. The fridge energy management system is based on an RL agent to control the fridge, and an artificial neural network to model the environment and assess the impact of its actions. Compared to the original fridge control, the RL-based control successfully reduced the total energy usage by 23% while also increasing the use of RES energy.