With the development of new-type power systems, source-load uncertainty and their mutual coupling have become increasingly prominent. Power system analysis must therefore fully account for the impact of uncertain factors on probabilistic power flow. To address the issue that traditional stochastic response surface method (SRSM) rely heavily on a large number of samples to achieve high modeling accuracy in probabilistic power flow calculations, a sample selection strategy based on the stochastic reduced order method (SROM) is proposed to extract representative samples from the full ensemble for constructing the surrogate model, thereby ensuring high computational accuracy. Furthermore, to better capture the spatial correlation and nonlinear relationships among input variables, the Kriging method is integrated with SRSM to develop an enhanced probabilistic power flow model. In addition, the global sensitivity analysis method is used to establish the sensitivity indices, calculate the sensitivity of the output variable of the probabilistic power flow to the input random variables, and quantify the impact on the operation state variables of the distribution network. Finally, numerical simulations demonstrate the feasibility and effectiveness of the proposed SROM-Kriging enhanced SRSM framework for accurate and efficient probabilistic power flow modeling.