基于机器学习的超高性能混凝土性能预测

Performance prediction of ultra-high performance concrete based on machine learning

基于改进的Andreasen-Andersen(MAA)颗粒堆积模型设计了93组UHPC配合比,根据93组样本的抗压强度与流动度建立了数据库,采用线性回归模型、二次项回归模型、人工神经网络(ANN)模型预测了UHPC的流动度和28 d抗压强度,并使用遗传算法(GA)对ANN模型进行优化得到了GA-ANN模型,最后对比了四种模型的预测精度和稳定性。结果表明:四种模型对UHPC流动度和28 d抗压强度预测精度的高低顺序为GA-ANN模型>ANN模型>二次项回归模型>线性回归模型,其中,GA-ANN模型的R2分别达到了0.979和0.982。

Based on the modified Andreasen-Andersen(MAA)particle accumulation model,93 groups of UHPC proportions were designed,and a database was established based on the compressive strength and fluidity of the 93 groups of samples.Then the linear regression model,quadratic term regression model,and artificial neural network(ANN)model were used to predict the fluidity and the 28 d compressive strength of UHPC.And the genetic algorithm(GA)was used to optimize the ANN model to obtain the GA-ANN model.Finally,the prediction accuracy and stability of the four models were compared.The results show that,the order of accuracy of the four models in predicting the fluidity and 28 d compressive strength of UHPC is as follows:GA-ANN model>ANN model>quadratic term regression model>linear regression model,in which the R2 of the GA-ANN model reaches 0.979 and 0.982,respectively.