针对混凝土坝变形预测模型中环境量与效应量之间复杂的非线性问题，以及单支持向量机(SVM)模型预测精度不高的问题，提出一种AdaBoost-SVM的混凝土坝变形预测模型，该模型采用结构风险最小化的原则，并借鉴提升算法强化学习的思想，从而提高模型的学习性能，达到增强模型泛化能力和预测精度的目的。结合实例，经过AdaBoost-SVM预测模型对混凝土坝位移原型监测数据进行训练及预测，并将预测结果与单支持向量机模型的预测结果进行对比，结果显示：基于AdaBoostSVM预测模型得到的均方差为0.5565，平均误差绝对值为0.40，预测精度比单支持向量机模型高出一个数量级；而且相较于单支持向量机预测模型，强化后的模型在预测时段表现出更好的稳定性。该模型综合了提升算法与支持向量机各自的优势，可作为混凝土坝变形预测的一种有效方法。

Aiming of the complex non-linearity problem between environment and effect in concrete dam deformation prediction model and the low prediction accuracy of single support vector machine (SVM) model,an AdaBoost-SVM model for concrete dam deformation prediction was proposed.The model adopted the principle of minimizing structural risk and used the idea of learning algorithm for reference to improve the learning performance of the model for enhancing the generalization ability and prediction accuracy of the model.The prototype monitoring data of concrete dam displacement were trained and predicted by the AdaBoost-SVM prediction model,and the prediction results were compared with those of the single support vector machine model.The results showed that the mean square deviation of the prediction model based on AdaBoost-SVM was 0.5565, and the absolute average error was 0.40,while the prediction accuracy was one number higher than that of the single support vector machine model.In addition,compared with the single support vector machine prediction model,the enhanced model showed better stability in the prediction period.The model combined the advantages of lifting algorithm and support vector machine,and can be used as an effective method for deformation prediction of concrete dams.

国家重点研发计划（2016YFC0401601）；国家自然科学基金项目（51739003;51779086;51609074）