We present a modelling approach for an adaptive cloud infrastructure consisting of secure pools of virtualized resources, in order to facilitate automated management tasks and interaction with the system by a human administrator, or programmatically by a higher level service. The topology of such a system is rapidly changing as, for example, it has the abilities to create, modify or destroy pools of virtual resources according to customer demand, as well as dynamically modify the mapping of virtual to physical resources. It is also highly distributed and management data needs to be compiled from disparate sources. Our modelling approach, based on the semantic web, allows us to represent complex topologies, model incomplete or erroneous systems and perform operations such as query, while still allowing validation of the models against system invariants and policies. It also supports distributed modelling, allowing sub-models to be combined, data merging, and shared vocabularies.