Abstract: Aiming at energy optimal scheduling problems in newly established microgrids（MGs） with the data scarcity and the uncertainty of source and load, this paper proposes a dual-data-model-driven distributionally robust optimization（DRO） framework for MGs. Firstly, the accuracy and robustness of the scenario generation using historical meteorological data are enhanced by the integration of neural networks with photovoltaic physical generation models to address the problem of data scarcity. Secondly, by the introduction of the DRO strategy and linear decision rules based on the Wasserstein distance, the energy optimization scheduling problem of MGs considering the uncertainty of source and load is transformed from a complex semi-infinite programming（SIP） problem to a mixed-integer linear programming（MILP） problem that is easy to be solved. The proposed DRObased energy scheduling framework can realize the balance between low operation costs and high reliability, and can adapt to the real-time changes in photovoltaic generation power and other factors. Finally, the experimental comparison results under three typical weather conditions verify the effectiveness of the proposed method.