In view of the permeability coefficient prediction for gravel soil of a core wall only considering material sources parameters, while a few prediction methods only considering construction quality having the problems of large error and incomplete data characteristics, a combined permeability coefficient predicting model BPNN-WOA-SVM considering both construction quality and material sources parameters was proposed. Whales optimization algorithm (WOA) was introduced to solve the difficult problem of SVM parameter selection, and the maximum information entropy principle was adopted to synthesize the strong adaptive ability of BPNN, good regression performance of whale optimized support vector machine (WOA-SVM) algorithm, and the advantages of small sample size.Engineering application results show that compared with the single prediction method, the combined prediction method reduces the mean square error, mean absolute error and relative analysis error, improves the prediction accuracy and convergence rate, and has a strong advantage in predicting the permeability coefficient for gravel soil of a core wall.