In order to effectively solve coal mine water penetration accident and improve emergency rescue capability and evaluation accuracy, a emergency rescue capability evaluation model of coal mine water penetration accident is proposed, which combines an improved combination weighting method with the marine predator algorithm (MPA) optimizing BP neural network (BPNN). First of all, the evaluation index system of emergency rescue capability of coal mine water penetration accident is constructed, including four primary indicators, emergency rescue prevention capability, emergency rescue preparation capability, emergency rescue response capability, rehabilitation recovery capability, and sixteen secondary indicators. Secondly, the subjective and objective weighting of the evaluation indicators are determined by the best worst method (BWM) and the criteria importance though intercrieria correlation (CRITIC) method. Lagrange function is introduced to build a decision model, and combination weighting are obtained by coupling the subjective and objective weighting through the euclidean distance function. Thirdly, the combined weight value is used as the input of the MPA-BPNN model, and the expected value as the output for linear regression prediction. Finally, the model is applied in a coal mine in Shanxi and compared with the BPNN model, GA-BPNN model, and PSO-BPNN model. The results show that mean absolute error of MPA-BPNN model has decreased by 6.5%, 4.3% and 3.5% respectively compared with other models, which proves the effectiveness and accuracy of the model. Therefore, MPA-BPNN is applicable to the evaluation of emergency rescue capability for coal mine water penetration accident.