【作者】 刘铮；

【导师】 王新定；

【作者基本信息】 东南大学 ， 桥梁与隧道工程， 2025， 硕士

【摘要】 玄武岩纤维混凝土(Basalt Fiber Reinforced Concrete,BFRC)凭借其优异的力学性能在土木工程领域展现出广阔应用前景,然而,由于其复杂的材料组成,导致力学性能与配合比参数之间的关联机制并不清晰,而传统的试错法效率低下、成本高昂,严重制约工程应用发展。为突破这一瓶颈,本文利用机器学习技术对玄武岩纤维混凝土力学性能(抗压强度、抗折强度)进行研究,并结合多目标优化算法进行配合比优化设计。本文的主要研究内容如下:(1)分析了玄武岩纤维混凝土配合比参数对其力学性能的影响,确定了后续模型的输入、输出变量,通过相关文献收集BFRC配合比用量与力学性能数据,采用箱线图与局部离群因子法进行数据清洗,构建BFRC力学性能试验数据库,分析各特征变量的统计分布规律,计算两两变量之间的皮尔逊相关系数以初步分析二者之间的关联规律,确定数据集划分方法并建立评价指标体系,为机器学习建模奠定基础。(2)根据数据集自身特性以及力学性能预测需求,筛选出反向传播神经网络(BPNN)、支持向量机回归(SVR)和极端梯度提升(XGBoost)三种模型,并分析三者的适用性。结合模型的超参数空间特征,选择贝叶斯优化(BO)和遗传算法(GA)作为优化算法,构建超参数优化框架。(3)建立默认参数设置的基准模型初步评估预测性能,分析超参数设置对模型的影响并确定调优范围。基于贝叶斯优化与遗传算法进行模型优化,确定优化后模型的最佳超参数,并对比两种优化算法在计算效率与精度方面的差异。结合预测点的分布以及评价指标对比优化前后模型的拟合效果。提出一个综合评价指标并以此确定BO-XGB模型为BFRC力学性能预测的最佳模型,通过SHAP法对BFRC力学性能预测模型的输入参数进行分析,确定影响程度较大的变量。(4)基于非支配排序遗传算法(NSGA-Ⅱ),以BFRC的抗压强度、抗折强度及原材料经济成本作为目标函数,考虑材料配比约束,构建多目标优化模型。对全局范围内同时满足强度最高、成本最低的配合比进行寻优。分析获得的Pareto前沿在搜索空间里的分布,并对相应的配合比进行统计分析。利用熵权-TOPSIS法从Pareoto前沿中筛选出三类典型配合比(经济优先、强度优先、平衡性)作为推荐配合比,为不同工程偏好下的BFRC配合比优化设计提供了参考。更多还原

【Abstract】 Basalt Fiber Reinforced Concrete(BFRC)has shown broad application prospects in the field of civil engineering due to its excellent mechanical properties.However,due to its complex material composition,the correlation mechanism between mechanical properties and mix proportion parameters is not clear,and the traditional method is inefficient and costly,seriously restricting the development of engineering applications.To overcome this bottleneck,this thesis uses machine learning to study the mechanical properties(compressive strength and flexural strength)of basalt fiber reinforced concrete,and combines multi-objective optimization algorithms for mix proportion optimization design.The main research content of this thesis is as follows:(1)The influence of the mix proportion parameters of basalt fiber reinforced concrete on its mechanical properties was analyzed,and the input and output variables of the subsequent model were determined.BFRC mix proportion and mechanical performance data were collected through relevant literature.Box plot and LOF local outlier factor method were used for data cleaning,then a BFRC mechanical performance test database was constructed.The statistical distribution law of each characteristic variable was analyzed,and the Pearson correlation coefficient between two variables was calculated to preliminarily analyze the correlation law.The dataset partitioning method was determined and an evaluation index system was established,laying the foundation for machine learning modeling.(2)Based on the characteristics of the dataset itself and the demand for predicting mechanical properties,three models were selected:backpropagation neural network(BPNN),support vector machine regression(SVR),and extreme gradient boosting(XGBoost),and their applicability was analyzed.Based on the hyperparameter space characteristics of the model,Bayesian optimization(BO)and genetic algorithm(GA)are selected as optimization algorithms to construct a hyperparameter optimization framework.(3)Establish a benchmark model with default parameter settings to preliminarily evaluate predictive performance,analyze the impact of hyperparameter settings on the model,and determine the tuning range.Based on BO and GA for model optimization,determine the optimal hyperparameters of the optimized model,and compare the differences in computational efficiency and accuracy between the two optimization algorithms.Compare the fitting effect of the model before and after optimization based on the distribution of predicted points and evaluation indicators.Propose a comprehensive evaluation index and use it to determine the BO-XGB model as the optimal model for predicting the mechanical properties of BFRC.Analyze the input parameters of the BFRC mechanical properties prediction model using SHAP method to identify the variables with significant impact.(4)Based on the Non-dominated sorting genetic algorithmⅡ(NSGA-II),a multi-objective optimization model is constructed with the compressive strength,flexural strength,and raw material economic cost of BFRC as objective functions,considering material ratio constraints.Optimize the mix proportion that simultaneously satisfies the highest intensity and lowest cost on a global scale.Analyze the distribution of Pareto frontiers obtained in the search space and perform statistical analysis on the corresponding mix proportions.Using the entropy weight TOPSIS method,three typical mix proportions(economic priority,strength priority,balance)were selected from the Pareto frontier as recommended mix proportions,providing a new method for optimizing the design of BFRC mix proportions.更多还原