A two-stage robust optimization model of power system considering source-load uncertainty is proposed，to address the serious lack of system scheduling flexibility caused by the source-load uncertainty in new energy power systems. According to the characteristics of source-load uncertainty，the K-means method and robust optimization theory are combined to quantify the flexibility demand of the power system at multiple time scales. Firstly，the robust dispatch model is established，and the flexible regulation potentials of thermal power units，pumped storage and other resources are fully exploited.The flexible transformation of thermal power units and pumped storage pumping status are included in the first stage of the model，and the output of the flexible resources is taken as the second stage of the decision variables. The optimization objective of the model is to minimize the cost of retrofitting，carbon emission and operating costs. The two-stage robust model is transformed into relatively independent main problems and sub-problems，and the column constraint generation （C&CG） algorithm and strong dyadic theory are adopted to iterate repeatedly to approximate the optimal solution. Finally，the proposed optimal scheduling strategy is verified through examples，so that the proposed optimal scheduling strategy can integrate all kinds of resources based on meeting the demand for flexibility，which achieve the balance of economy，environmental protection，and flexibility in the system，and improve the ability to resist the risk of uncertainty in the source load.