【摘要】 缺失值处理是大坝安全监测数据分析的重要基础，传统大坝缺失值处理方法多采用单一基学习器进行预测插补，未有效融合多类型机器学习模型的优势。本文在集成学习框架下，将多种经典机器学习及深度学习算法集成为强学习器，针对各模型权重分配问题，提出一种新的CRITIC-Stacking(CS)权重分配方法，从而建立基于CS集成学习的大坝监测数据插补混合模型。研究结果表明，相较于单一基学习器和传统的Stacking集成模型，其RMSE指标平均降低了72.7%和58%，说明集成学习可以充分发挥多种机器学习模型的预测优势，并且权重分配方法的改进同样可以提升集成学习模型的预测精度，进而为大坝监测数据缺失值处理及预测模型构建提供了一种新的建模思路。更多还原

【Abstract】 Missing value processing is an important foundation for analysis of dam safety monitoring data. Traditional methods for handling the missing values of a dam often use a single type of machine learning models for prediction and interpolation, ineffective in integrating the advantages of multiple types of machine learning models. This article integrates multiple classic machine learning and deep learning algorithms into a strong learner within the framework of ensemble learning. To address the issue of weight allocation to each model, we develop a new critic stacking(CS) weight allocation method so that we can construct a dam monitoring data interpolation hybrid model based on CS ensemble learning. The results show that compared to single base learners and traditional Stacking ensemble models, this CRITICStacking ensemble learning method reduces the RMSE index by an average of 72.7% and 58%. This indicates that the method can fully leverage the predictive advantages of various machine learning models, and the improvement of weight allocation can also improve the predictive accuracy of ensemble learning models, thus providing a new solution for handling missing values in dam monitoring data and constructing prediction models.更多还原