Abstract: Smart home technology has developed rapidly. In the development of smart home technology, air conditioner plays a vital role. It improves indoor thermal comfort and air quality, providing a comfortable living environment and protecting the health of the occupants. For smart home algorithms, active smart control of residential air conditioning is the most important component. It aims at providing occupants with all-day auto-control based on usage behavior and sensor information. However, active smart control needs to face the challenges of limited edge resources and lacking of unified modelling architecture. Proposing a framework called Auto-adaptive Multi-task Edge Computing Framework (AMEC), which combines multi-task architecture and auto-adaptive training strategy for active smart control. The framework can unify and integrate smart control algorithms, improving edge computing efficiency of ensemble active smart control algorithms for air conditioners. In the experiment, by integrating and optimizing modelling tasks in a mainstream active smart control system, AMEC brings about 20% to 50% parameter reduction without significant performance degradation, greatly improving the computational performance of the algorithm and reducing the edge resource overhead of its application.

Key words: Edge computing, Internet of Things, Multi-task, Smart control