Reinventing domestic refrigeration: How machine learning is powering energy savings

A recent body of research has focused on the optimisation of domestic refrigerators, using for instance door sensors and thermal load to implement fuzzy controllers that reduce energy consumption. Although these strategies improve energy efficiency, their setting still largely depend on empirical heuristics and expert knowledge.

In a recent study published in IIR’s International Journal of Refrigeration, a team of researchers from Zhejiang University and HangZhou Kangbei Motor Co., Ltd in China developed a control strategy for domestic refrigerators based on behaviour prediction and deep reinforcement learning.

Reinforcement learning (RL) is a type of machine learning process in which the system interacts with its environment through trial and error, utilising feedback mechanisms to learn optimal actions that maximise cumulative rewards.

In their study, Hao-Ran Li and colleagues proposed an algorithm which incorporates predictive information on user door-opening behaviour to improve the refrigerator's temperature regulation and energy efficiency. Their model predicts user door-opening behaviour with 85% improved accuracy compared to existing research.

Experiments show that their algorithm achieved up to 28.7% energy savings over one week compared to conventional control methods, while maintaining stable cabinet temperatures.

The researchers believe that with the advancement of IoT technology, future improvements in energy efficiency could be realised by incorporating a broader range of data to better capture user activity patterns.

The complete study is available in the International Journal of Refrigeration.