Full autonomy is the ultimate goal of the building industry to achieve energy optimality. Buildings have non-stationary and complex dynamics, as the state space to describe their behavior is significantly large and intractable. Existing control approaches are not appropriate to meet the requirements of building automation since they are black-box methodologies which fail to provide a system-level optimal control solution for buildings. They are also static which makes them unable to adapt as the dynamics of the buildings vary. Moreover, model-free approaches are only valid for a specific building and lack the ability to extend to other buildings. This paper offers a Deep Digital Twin approach as a new artificial intelligence structure for building automation platforms. The Deep Digital Twin utilizes a physics-based ontology to provide a self-knowledge of the building operation, and the cause-and-effect interrelation between building subsystems which enables future-forward energy efficient control of the building at the system level.