【作者】 肖祁南；

【导师】 焦楚杰；

【作者基本信息】 广州大学 ， 结构工程， 2022， 硕士

【摘要】 常规的混凝土配合比设计方法是通过大量的试配工作,才能得到满足目标的混凝土配合比,试配过程中耗费大量的人力、物力和时间资源。原因在于:混凝土不仅需要力学性能和工作性能上满足需求,同时还需要满足经济合理的需求,而上述目标之间往往存在一定的冲突性。二十一世纪以来,人工智能技术迅速发展并被广泛应用于多个领域的研究中。因此,本文使用人工智能技术建立混凝土性能预测模型,将其代入智能优化算法构建混凝土配合比多目标优化模型,以减少试配时的资源耗费和提高混凝土的性价比,并设计便于使用的用户图形界面,实现混凝土配合比设计的智能化、精准化。本文的主要工作内容:(1)建立混凝土性能预测模型:以混凝土配合比中各个原材料的用量为输入变量,利用反向传播神经网络(BPNN)、支持向量机(SVM)和随机森林(RF)三种机器学习模型分别建立抗压强度和坍落度预测模型,建立过程中均使用粒子群优化算法(PSO)结合交叉验证调整其超参数。训练建模完成后,比较三种模型的预测准确度,选择最优模型代入优化的目标函数。结果表明:在本文所选的三种机器学习模型中,BPNN对于混凝土的抗压强度有着更好的预测效果,其在验证集样本中预测输出与实际输出的相关系数R=0.9531,均方根误差RMSE=4.2568,绝对平均误差MAE=2.6627;RF对于混凝土的坍落度有着更好的预测效果,其在验证集样本中预测输出与实际输出的R=0.8986,RMSE=9.4906,MAE=5.5034。(2)建立混凝土配合比多目标优化模型:将BPNN混凝土强度预测模型和RF混凝土坍落度预测模型代入目标函数,并以混凝土配合比设计的相关规范对水胶比、胶材用量、砂率和设计容重的等进行约束,约束条件以惩罚函数的形式实现。将构建的目标函数代入PSO,并在PSO中加入遗传算法的进行随机变异操作,来提高模型的全局搜索能力,最终建立混凝土配合比多目标优化模型。(3)试验验证模型的优化效果:选择C30～C60的混凝土配合比,保持强度目标和坍落度目标不变,价格目标按照原材料成本、降低5元和降低10元划分为3组,分别代入混凝土配合比多目标优化模型,得到混凝土配合比和对应的期望性能。按照标准测试其抗压强度和坍落度,并对比其期望性能和实际性能之间的误差。结果表明:强度相关性R=0.9546,RMSE=3.4555,MAE=3.0583;坍落度相关性R=0.8046,RMSE=6.0902,MAE=5,模型输出配合比的抗压强度和坍落度的实际值与预期值之间均有着较强的相关性、误差较小;随着单方成本目标发生变化,强度预测准确度和坍落度预测准确度均有着不同程度的降低,与原价格相比降低5元与10元后,强度预测值与实际值之间的RMSE分别升高了12.1%和24.6%;坍落度预测值与实际值之间的RMSE分别升高了20%和40%;(4)编写混凝土配合比智能优化软件:将底层代码和用户图形界面相结合,搭建混凝土配合比多目标优化软件,并介绍了混凝土配合比多目标优化系统的运作流程。更多还原

【Abstract】 The conventional concrete mix proportion design method is to obtain the concrete mix proportion that meets the target through a lot of trial mixing work.The trial mixing process consumes a lot of manpower,material resources and time resources.The reason is that concrete not only needs to meet the needs of mechanical properties and working performance,but also needs to meet the economic and reasonable needs,and there is often a certain conflict between the above objectives.Since the 21 st century,artificial intelligence technology has developed rapidly and has been widely used in many fields.Therefore,this paper uses artificial intelligence technology to establish the concrete performance prediction model,and substitutes it into the intelligent optimization algorithm to construct the multi-objective optimization model of concrete mix proportion,so as to reduce the resource consumption during trial mixing and improve the cost performance of concrete,and design an easy-to-use user graphical interface to realize the intelligence and accuracy of concrete mix proportion design.The main contents of this paper:(1)Establishment of concrete performance prediction model: Taking the amount of each raw material in the concrete mix proportion as the input variable,the compressive strength and slump prediction models are established respectively by using three machine learning models: back propagation neural network(BPNN),support vector machine(SVM)and random forest(RF).In the establishment process,particle swarm optimization algorithm(PSO)combined with cross validation is used to adjust its hyperparameters.After the training modeling is completed,the prediction accuracy of the three models is compared,and the optimal model is selected to substitute into the optimized objective function.The results show that among the three machine learning models selected in this paper,BPNN has a better prediction effect on the compressive strength of concrete.In the validation set,the correlation coefficient between the predicted output and the actual output is R = 0.9531,the root mean square error RMSE = 4.2568,and the absolute average error MAE = 2.6627;RF has a better prediction effect on the slump of concrete.In the verification set,the predicted output and actual output are R= 0.8986,RMSE = 9.4906,MAE = 5.5034.(2)Establish the multi-objective optimization model of concrete mix proportion: replace the BPNN concrete strength prediction model and RF concrete slump prediction model into the objective function,and restrict the water binder ratio,binder dosage,sand ratio and design unit weight according to the relevant specifications of concrete mix proportion design,and the constraint conditions are realized in the form of penalty function.The constructed objective function is substituted into PSO,and the random mutation operation of genetic algorithm is added into PSO to improve the global search ability of the model,and finally the multi-objective optimization model of concrete mix proportion is established.(3)Test to verify the optimization effect of the model: select the concrete mix proportion of C30 ～ C60,keep the strength target and slump target unchanged,and divide the price target into three groups according to the raw material cost,5 yuan reduction and 10 yuan reduction.Replace them into the multi-objective optimization model of concrete mix proportion respectively,and obtain the concrete mix proportion and corresponding expected performance.The compressive strength and slump of the product are tested according to the standard,and the error between the expected performance and the actual performance is compared.The results show that the strength correlation R = 0.9546,RMSE = 3.4555,MAE = 3.0583;Slump correlation R = 0.8046,RMSE = 6.0902,MAE = 5.There is a strong correlation between the actual and expected values of compressive strength and slump of the output mix proportion of the model,and the error is small;With the change of unilateral cost target,the accuracy of strength prediction and slump prediction decreased in varying degrees.After reducing 5 yuan and 10 yuan compared with the original price,the RMSE between strength prediction value and actual value increased by 12.1% and 24.6% respectively;The RMSE between the predicted value and the actual value of slump increased by 20% and 40% respectively;(4)Compile the intelligent optimization software of concrete mix proportion: combine the bottom code with the user graphical interface,build the multi-objective optimization software of concrete mix proportion,and introduce the operation process of the multi-objective optimization system of concrete mix proportion.更多还原