Abstract: Compared with single model, ensemble learning has better prediction accuracy and generalization ability, which is widely used in industrial process quality prediction. The diversity between basic learners and the accuracy of basic learners have a great impact on the generalization ability of the ensemble model. To improve the generalization ability of ensemble model, a selective ensemble modeling method based on online extreme learning machine and game theory was proposed in this paper. Online sequential extreme learning machine(OSELM) was used as base learner, and the accuracy of the base learner and the contribution rate of the base learner to the diversity of the ensemble model were regarded as the two sides of the game. Game theory was used to deal with the optimal selection scheme for the accuracy and diversity of the ensemble model, and the accuracy and diversity of the ensemble model were optimized. After the model prediction was completed, the weights of the base learner were updated online by integrating current errors and historical errors, therefore, the ensemble model was dynamically adaptive. The rationality and effectiveness of the proposed algorithm were verified by using open data sets and real industrial data.