【摘要】 近年来，智能家居应用的普及使得用户的家庭生活变得更加舒适与便捷。现有的智能家居应用涵盖了包括自动控制、安全识别与人机交互等多个方面。其中，主动智能控制在智能家居应用场景中扮演着至关重要的角色。主动智能控制系统旨在基于感知信息、专家经验与用户行为等多方面因素，挖掘全屋设备控制逻辑，为用户提供全天候自动化的家居控制服务，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年来在多方向上取得了突破，其在应用方面仍然面临许多挑战。受制于边缘资源有限与硬件多样适配难的问题，主流主动智能控制算法大多采用云计算模式为用户提供服务。然而相较边缘实现，云端部署算法存在难以克服的网络延迟、稳定性、隐私安全等问题。于是提出了一种用于智能家居主动智能控制的边缘集成框架，针对资源有限与硬件适配问题，上述框架提供了统一的模型压缩方法，可以按需针对不同算力压缩要求进行适配，灵活的模型伸缩性有助于快速应用于资源有限的各类物联网设备或不同平台，有效地解决了资源有限和硬件能力差异大的问题，提高主动智能控制算法的应用效率和可靠性。更多还原

【Abstract】 In recent years, the popularity of smart home applications has made users’ home life more comfortable and convenient. Existing smart home applications cover many aspects including automatic control, security identification and human-machine interaction. Among them, active intelligent control plays a crucial role in smart home application scenarios. The active intelligent control system aims to mine the whole house equipment control logic based on various factors such as sensory information, expert experience and user behavior, and provide users with all-weather automated home control services, thus providing a comfortable living environment and ensuring the health of occupants. Although active intelligent control algorithms have made significant breakthroughs in various directions in recent years, they still face many challenges in their applications. Most algorithms use cloud computing frameworks due to limited edge resources. However, cloud computing has inevitable issues such as latency, stability, privacy and security. In this paper, we propose an edge-integrated framework for automatic indoor environment control algorithms. The framework provides a unified model compression approach, enabling rapid deployment for various types of devices with limited resources. Flexible model resizing capability effectively addresses the problem of limited resources and hardware capability differences.更多还原