The self-adaptation to the environmental changes is one of the key issues of organization-based multi-agent systems. Dynamic reorganization of organizational structures provides an effective approach for multi-agent systems to realize organizational objectives flexibly. Based on the structural characteristics of agent organizations, this paper presents a single-rooted hierarchical graph model describing social structure, role enactment and agent coordination of the organizational structures. This model decreases effectively the complexity of reorganization for large-scale agent organizations by maintaining their structural elements based on the single- rooted and hierarchical graph approach. It formalizes the reorganization process of agent organizational structures by extending the algebraic graph transformation with the DPO (double-pushout) approach. In this formal specification, the single-rooted hierarchical graphs characterize different states of organizational structures and the derivation sequences of transformation rules formulate the transition process of organizational structures. Finally, the experimental results on reorganization simulation and matching algorithm of organization transformation rules indicate that this hierarchical graph transformation approach defines formally the reorganization process of agent organizations, and supports the graph-based design of organizational elements during the reorganization process and the reorganization computation of large-scale agent organizations.