摘要
随着人工智能算法近年的不断突破,全屋智能化技术迎来了快速发展,而家用空调作为全屋家电中具备室内环境调节能力的子系统,是全屋智能化中至关重要的一环。在全屋智能的多类算法发展方向中,空调器的主动智能控制又是如今智能化浪潮中最重要的一步。其致力于基于用户使用行为,综合多类感知信息,挖掘空调控制逻辑,为用户提供全天候的智能化控制,从而提供舒适的居住环境,保障居住者的健康。尽管主动智能控制算法近些年在很多方向产生了突破,但目前其在算法应用方面,仍需要面对建模方法不统一与算力资源有限的挑战。为了解决上述问题,提出了结合多任务架构与任务自平衡方法的自适应多任务的主动智能控制边缘计算集成框架Auto-adaptive Multi-task Edge Computing Framework (AMEC)。该框架能够统一集成多类任务,提高家电尤其是空调主动智能控制算法的算力算效。在实验中,AMEC通过集成并优化现有主动智能调控系统中多个子任务,在核心算法指标不产生明显衰减的基础上可带来20%至50%的参数缩减,大大提高了算法的计算性能,减少了其应用时的资源开销。
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.
作者
唐善玄
刘猛
王少华
张浩
樊其锋
TANG Shanxuan;LIU Meng;WANG Shaohua;ZHANG Hao;FAN Qifeng(Guangdong Midea Refrigeration Equipment Limited Company,Foshan 528000;Chongqing University,Chongqing 400045)
出处
《家电科技》
2023年第6期22-26,共5页
Journal of Appliance Science & Technology
关键词
边缘计算
物联网
多任务框架
智能控制
Edge computing
Internet of Things
Multi-task
Smart control
