Abstract:In order to study the possible influencing factors of the forecasting effect of stacking integrated model, this paper takes Andun River Basin as an example, selects support vector regression, multiple linear regression, long short-term memory neural network, feedforward neural network, gradient boosting regression tree, autoregressive integral moving average model and adaptive boosting algorithm as the base learners, and selects multiple linear regression, support vector regression and multilayer perceptron as meta learners to establish stacking ensemble models. The prediction effects of each ensemble model are compared by the average absolute error, correlation coefficient, root mean square error, average relative error, quality rate, Nash efficiency coefficient. The results show that the prediction effect of stacking ensemble model is not related to the number of base-learners, but positively related to the quality of base-learners. In addition, different selection of meta-learners will also affect the prediction effect of stacking ensemble model. This study can provide scientific guidance for runoff forecasting using stacking integrated model.