【摘要】 为了快速准确地判定锈蚀钢筋混凝土(Reinforced Concrete, RC)梁的抗弯承载力，利用集成学习研究锈蚀RC梁基于数据驱动的承载力预测模型。根据现有文献搜集并建立了锈蚀RC梁抗弯承载力试验数据库，基于数据集样本建立基于随机森林(RandomForest)、自适应增强(Adaboost)、梯度提升决策树(GBDT)、极限梯度提升算法(XGBoost)、轻量级梯度提升机算法(LightGBM)等5种集成学习算法的承载力预测模型，并借助网格搜索对模型进行超参数优化以提高其泛化性能。对比了不同集成学习算法的性能，即通过数据集分析了输入参数的特征重要性，对比分析了预测模型的平均绝对误差(MAE)、决定系数(R~2)、均方根误差(RMSE)以判定其合理性与精确性。分析结果表明：该预测模型可以高效地确定锈蚀RC梁抗弯承载力的关键影响因素，即钢筋配筋率和钢筋锈蚀率；基于RandomForest的模型表现最优，其次是基于XGBoost的预测模型，预测模型在训练集和测试集上的拟合度可以达到90%以上。更多还原

【Abstract】 To quickly and accurately determine the flexural capacity of corroded reinforced concrete(RC) beams, an ensemble learning-based data-driven bearing capacity prediction model for corroded RC beams was studied. A database of experimental tests on the flexural bearing capacity of corroded RC beams was established based on existing literature. Based on the dataset, five types of ensemble learning algorithms, namely Random Forest(Random Forest), Adaptive Boosting(Adaboost), Gradient Boosting Decision Tree(GBDT), Limit Gradient Boosting Algorithm(XGBoost) and Light Gradient Boosting Algorithm(LightGBM), were used to establish prediction models. Grid search was employed to optimize the hyperparameters of the models to improve their generalization performance. The performance of different ensemble learning algorithms was compared, and the feature importance of input parameters was analyzed through the dataset. The mean absolute error(MAE), determination coefficient(R~2) and root mean square error(RMSE) of the prediction models were compared to assess their rationality and accuracy. The analysis results indicated that the prediction model could effectively determine the key influencing factors of the flexural bearing capacity of corroded RC beams, namely the reinforcement ratio and the corrosion rate of the rebar. The model based on RandomForest performed the best, followed by the model based on XGBoost. The fitting degree of the prediction models on the training and test sets could reach over 90%.更多还原