To reduce the cost and cycle time of iterative design experiments for composite solid propellant （CSP） formulations， this study combines machine learning with virtual formulation generation algorithms to conduct high-throughput virtual screening of nitrate ester plasticized polyether （NEPE） propellant burning rate performance and formulations. First， a dataset was constructed using 85 experimental data samples of NEPE propellants， with formulation composition， theoretical molar mass （M_T）， and working pressure （P） as inputs. A gradient boosting regression tree （GBRT） model was developed to predict the burning rate of NEPE propellants， and the Shapley Additive Explanations （SHAP） method was employed to calculate feature importance and identify key factors influencing burning rate. Subsequently， a constrained grid search algorithm was used to generate a large number of virtual formulations， and the GBRT model was applied to calculate the burning rates of all virtual formulations at multiple working pressures and the burning rate pressure exponents across various pressure ranges. Finally， virtual formulations were filtered and ranked according to different screening criteria. The results show that the GBRT model achieved a coefficient of determination of 0.980， mean absolute error of 0.427 mm·s^-1， root mean square error of 0.574 mm·s⁻¹， and symmetric mean absolute percentage error of 6.972% on the test set. P， M_T， and the mass percentage of Φ-Pb were identified as the three most important features. Using the constrained grid search algorithm， 377，127 virtual formulations were generated. After four rounds of screening （criteria： burning rate at 6 MPa within （10.00 ± 0.10） mm·s^-1， and burning rate pressure exponents below 0.5 across the ranges of 4-10， 6-10， and 4-6 MPa）， 637 virtual formulations met the requirements. Finally， the top 10 virtual formulations were selected based on proximity to the target burning rate （10.00 mm·s^-1） and the magnitude of burning rate pressure exponents， respectively. The formulation screening framework proposed in this study provides an efficient and feasible pathway for achieving intelligent design of CSP formulations.