Accurate prediction of reservoir porosity and permeability is of great significance to reservoir evaluation. For the calculation of reservoir parameters, the traditional empirical common method still has large errors. In order to improve the prediction accuracy of reservoir parameters and improve the generalization ability of the model, an integrated learning algorithm is proposed based on the improved Stacking fusion model. The differences in data observation and training angles between different algorithms are employed as the basic principle to fully leverage the advantages of the model. First, in terms of the traditional Stacking fusion learning model, the output results of the model for the base learner at the first layer are optimized. In view of the possible uneven data division that results in poor prediction, the weighted average of the prediction results is performed according to the test precision of the base model, and the results are obtained as the characteristics of the second layer. Secondly, the new combined training set may lose some of the information in the original training set, and the original data set is also used as a part of the training of the secondary learner, so that the meta-learner can learn the implicit relationship between the original training set and the new training set, thereby improving the model prediction effect. Finally, the models that are independent of each other are integrated through the Stacking fusion model to enhance model generalization. Compared with the traditional Stacking fusion learning model, the Root-Mean-Square-Error(RMSE) prediction of porosity and permeability in the improved model is reduced by 7.7% and 7.1%, respectively, which verifies that the model has good prediction.