智能家居利用不断发展的互联网、人工智能、传感器技术、无线通信等技术，将家庭设备与互联网连接起来，允许用户通过智能手机、电脑等设备控制和管理家庭设备，使人们的生活更加便利和智能化。智能家居已经成为物联网应用中最受关注的应用之一，但当前智能家居的发展仍然面临着诸多问题，如缺乏统一的设备标准化定义与互操作性、缺乏对于用户活动的准确认知与识别功能、主要依赖人为控制等问题，距离实现真正的主动智能尚有一定距离。本文聚焦于智能家居中的协作化智能关键技术，以打通设备与数据间的壁垒、加强互操作性的协作化体系为基础，对智能家居中的语义建模、用户活动识别、主动服务等重点工作进行了研究与改进。 对于语义建模任务，本文提出了一种智能家居的标准化语义建模方法，并融合时间信息、空间信息和用户信息等上下文信息，在通用规范的基础上实现了标准化的语义建模，以更好地理解用户日常生活的需求和活动习惯。该方法有效解决了非同源异构数据以及非同源设备的互操作性问题。 对于用户活动识别任务，本文提出了一种融合多环境传感器输入的基于Transformer的智能家居用户活动识别方法。提出的方法使用基于信息增益的活动驱动窗口分割传感器事件，提取多传感器输入特征，并创新性地使用了基于Transformer的识别模型，相较于已有的机器学习、深度学习方法，有效提高了智能家居用户活动识别的性能。 对于主动服务任务，本文提出了一种智能家居应用中基于规则匹配和生成候选服务的主动服务方案。利用互联网用户的群体智慧引入自动抓取的外部规则，构建了一个智能家居服务规则库，并设计了基于BERT预训练模型的双编码器规则匹配方法，用以根据用户当前活动状态和上下文信息生成相应的候选服务，为智能家居的主动智能提供解决方案。 本文对提出的各种方法逐一进行了实验验证和结果分析，充分验证了本研究的有效性和相较于已有工作的先进性。最后，本文对所有研究内容进行了总结，并立足实际应用场景对于未来研究工作的展望进行了探讨。
Smart home utilizes constantly evolving technologies such as the Internet, artificial intelligence, sensor technology, and wireless communication to connect home devices with the internet. This allows users to control and manage their home devices through smart devices like smartphones and computers, making their lives more convenient and intelligent. Smart home has become one of the most popular applications in the Internet of Things (IoT), but the current development of smart home still faces many issues, such as the lack of unified device standardization definition and interoperability, the lack of ac-curate recognition and identification of user activities, and heavy dependence on human control, which still has some distance from achieving true proactive intelligence. This article focuses on the key technologies of collaborative intelligence in smart homes, based on a collaborative system that enhances interoperability and breaks down barriers between devices and data. It investigates and improves the key work of semantic modeling, user activity recognition, and proactive services in smart homes. For the semantic modeling task, this thesis constructs a standardized semantic mod-eling method for smart home, and integrates contextual information such as time, space, and user information to achieve standardized semantic modeling on the basis of common specifications, which better understand the needs and activity habits of users. This method effectively solves the interoperability problem of non-same source heterogeneous data and non-same source devices. For the task of human activity recognition, this thesis proposes a transformer-based method for smart home human activity recognition with multiple environmental sensor inputs. The proposed method uses an information gain-based activity-driven window to segment sensor events, extract multi-sensor input features, and innovatively use a Transformer-based recognition model, which effectively improves the performance of smart home user activity recognition compared with the existing machine learning and deep learning methods. For the active service task, this thesis proposes an active service scheme based on rule matching and candidate services generation in smart home applications. By using the collective intelligence of Internet users to introduce the automatically captured rules, a smart home service rule base is constructed, and a dual encoder rule matching method based on BERT pre-training model is designed to generate the corresponding candidate services according to the user‘s current activity state and context information, so as to provide solutions for the active intelligence of smart home. In this thesis, various proposed methods are experimentally verified and the results are analyzed, which fully verifies the effectiveness of this research and the advanced compared with the previous work. Finally, the thesis concludes with a summary of the research and discusses the prospects for future research work based on practical application scenarios.
