Addressing the uncertainties of renewable energy and load within isolated microgrids, a robust economic optimization approach for microgrids is proposed based on scenario probability distribution uncertainty and probabilistic combined scenario performance. The K-means clustering method is employed to preprocess extensive historical data, constructing a fuzzy set of data-driven scenario probability distributions. In the day-ahead planning phase, the binary expansion concept is introduced to discretize the probabilistic combination coefficients in continuous variable form, simplifying and effectively parameterizing the intensity and search interval of the worst-case scenario search. This extends the search range of the worst-case scenario effectively from the boundary of the uncertainty set to its interior, enabling the search for the worst probabilistic combined scenario. By optimizing the performance of the worst probabilistic combined scenario, the day-ahead optimal solution for microgrid operation is calculated. Subsequently, in the real-time scheduling phase, real-time measurement data of renewable energy and load are utilized to perform secondary optimization adjustments on part of the day-ahead planning optimization solutions, enhancing the economic efficiency and robustness of the microgrid control scheme. Simulation examples demonstrate the effectiveness of the proposed method.