污染企业数字孪生体是指与污染企业物理实体完全等价的计算机信息模型，是实现在产污染企业场地管控智能化的基础。目前数字孪生体的构建方法在面对污染企业污染源种类多，源汇关系复杂及溯源难度大的情况时，难以实现对污染企业物理实体特征数据的完整表达。因此，本文提出一种构建污染企业数字孪生体的多粒度时空实体建模方法，通过对污染企业时空实体相关特征和关系的建模分析，以数据模型的形式实现对污染企业场地实体全面多维的信息描述与刻画。以广州市某橡胶厂为应用案例，进行数据模型的实例验证。结果表明，多粒度时空实体建模能高效组织和表达实体多维度的特征，实现污染企业数字孪生体模型的精准构建，为污染企业数字孪生应用服务提供有效方法和数据模型基础。

Digital twin entity refers to the information model that exists in the computer virtual space that is complete-ly equivalent to the physical entity, which is a multi-dimensional and multi-scale description and portrayal of the physical entity, and can achieve the real mapping and simulation analysis of the physical entity. In the field of ecological environment, there are many types of soil-groundwater pollution sources and pollutants in production enterprises, complex source-sink relationship, pollution early warning and traceability diffi-culties, etc. It is necessary to study the use of digital twinning technology to establish a real-life three-dimensional model of the plant area of the production enterprise and a digital twin model within the plant, and to incorporate real-time monitoring data into the simulation system of soil and groundwater twin-ning, and to carry out rapid data fusion and simulation derivation, and eventually form a complete digital twin system of a multi-granularity polluting enterprise plant. When facing the above problems, the current construction method of digital twin model mainly focuses on the construction of three-dimensional model, ignoring the complete expression and description of the enterprise entity's own geometrical parameters, at-tributes, relationships and other characteristic data, which is unable to meet the requirements of comprehen-sively and multi-dimensionally dynamic portrayal of the enterprise entity. In order to achieve the accurate construction of the twin entity model of the pollution enterprise, it is necessary to adopt a more efficient way to organise and manage the characteristic data of the enterprise entity, and select a suitable data model to model the enterprise scene. The traditional GIS data model uses conventional map data as a template for in-direct spatial modelling of entities, which limits the application of spatial data and makes it difficult to adapt to new expression needs such as real-time dynamic correlation in multimodal situations. To address this problem, Hua Yixin proposed a multi-granularity spatio-temporal object data model from the perspec-tive of data model. The model provides an all-round description of multi-granularity spatio-temporal entities from spatio-temporal reference, spatial location, attribute characteristics, spatial morphology, compositional structure, etc., with multi-dimensional and all-attribute characteristics. The model also increases the descrip-tion of characteristic information such as behavioural ability and association relationship, and the expres-sion content is more comprehensive and more in line with the actual entity expression. Therefore, this paper proposes a multi-granularity spatio-temporal entity modelling method for constructing a digital twin of a polluting enterprise, which achieves a comprehensive multi-dimensional information description and por-trayal of the site entity of a polluting enterprise in the form of a data model by modelling and analysing the relevant features and relations of the spatio-temporal entity of the polluting enterprise. A rubber factory in Guangzhou City is taken as a case study for instance validation of the data model. The results show that the multi-granularity spatio-temporal entity modelling can efficiently organize and express the mul-ti-dimensional features of the entity, achieve the accurate construction of the digital twin body model of the pollution enterprise, and provide an effective method and data model basis for the application service of the digital twin of the pollution enterprise.