To achieve real-time power balance between uncertain resource and demand consumptions in locally standalone power systems with a high penetration of new energy, it is promising to build concentrated solar power (CSP) plants through the establishment of power generation and flexible reserve provision. To address that, a refined maximum mean discrepancy-distributionally robust optimization (MMD-DRO) based economic dispatching model considering flexible reserve provision from CSP plants is designed in this work. Remarkably, a maximum mean discrepancy (MMD)-based ambiguity set representing new energy uncertain outputs is carried out. This proposed MMD-based ambiguity set mathematically keeps the shape of an ellipsoid through kernel mean embedding and related reformulation, and its center is equivalated as a group of empirical distributions extracted from historical data. On this basis, the impact of CSP and relevant flexible reserve capabilities in combination with conventional thermal power units is analyzed, and the joint up-forward/down-forward reserve provision produced from thermal power units and CSP is derived. Furthermore, a two-stage economic dispatching model considering flexible reserve configurations is constructed, which can be solved by the column & constraint generation algorithm. Some results on a modified IEEE 30-bus test system have demonstrated the effectiveness of the proposed model.