The architecture for IoT is the primary foundation for designing and implementing the System of Internet of things. This paper discusses the theory, method, tools and practice of modeling and reasoning the architecture of the Internet of Things system from the dimension of semantic ontology. This paper breaks the way of static ontology modeling, and proposes an implementation framework for real-time and dynamic ontology modeling of the IoT system from the running system. According to the actual needs of the health cabin IoT system and the combination of theory and practice, the system architecture model of the semantic ontology dimension of IoT is built. Then, based on the reasoning rules of the ontology model, the model is reasoned by Pellet reasoning engine which injects the atom of the custom reasoning built-ins into the source code. In this way we have realized the automatic classification and attribute improvement of resources and behaviors of the IoT system, the real-time working state detection and fault diagnosis of the IoT system, and the automatic control of the IoT system and resources.