When the single soft sensor model is used for the dynamic liquid level prediction, there are many shortcomings such aspoor generalization ability, weak adaptive ability, etc. In order to solve these problems, a soft sensor modeling method based on AdaBoost ensemble learning algorithm is proposed in this paper. The proposed method focuses on effects of the prediction error to the weight of the modeling samples and weak learning machine, therefore which is more suitable for the regression model prediction.In practical production,dynamic and changing working conditions during operations may lead to failure of the soft sensor model. In order to solve this problem,a small amount of patrolmeasuring data of the dynamic liquid levelis used to evaluate the original model, and then the similarity principle is used to add new data on the basis of the original model. And on this basis the weight of the new data is used to update the weak learning machine to become the strong learning machine model to dynamically adapt to the new production conditions.The simulation results using the real operation data of the oil well show that the proposed method has strong adaptive ability for fluctuation in production and can improve the generalization ability and the prediction accuracy of the soft sensor model.