Abstract: The accurate prediction of the SO₂ concentration at the desulfurization outlet is of great significance to realize the economic operation of the desulfurization system.Aiming at the problem that SO₂ concentration at desulfurization outlet is difficult to predict accurately due to many influencing factors, we proposed a prediction model based on Runge Kutta optimization kernel limit learning machine（KELM） and improved AdaBoost integration algorithm. Firstly, the kernel extreme learning machine was used as the weak predictor, and the strong predictor was constructed by using the AdaBoost ensemble algorithm. By adjusting the operating data weight of the desulfurization system under different working conditions, we established a prediction model of outlet SO₂ concentration based on the AdaBoost ensemble algorithm. In order to further improve the learning performance and prediction accuracy of the model, the loss function of the AdaBoost algorithm was improved by introducing penalty coefficients and prior knowledge parameters, and the Runge-Kutta algorithm was used to optimize the regularity coefficient C and kernel parameter S of KELM to overcome the influence of the initial parameter setting on the model stability and prediction accuracy. Finally, the simulation experiment was carried out by using the power plant operation data and the results show that the established integrated model of outlet SO₂ concentration has higher prediction performance and accuracy, and can provide technical support for the on-site optimal control of the desulfurization system.