基于粗糙集(RS)和支持向量机(SVM)的混凝土性能预测实证研究

Prediction of concrete properties based on rough sets and support vector machine method

针对多元胶凝体系混凝土的性能预测问题,以各种原材料用量及其质量指标为预测自变量,以混凝土初始坍落度和7天、28天抗压强度为预测因变量,先采用粗糙集(RS)方法约简自变量组合,后采用支持向量机回归(SVR)模型建立混凝土性能预测模型。通过单掺粉煤灰集169组配合比、复掺粉煤灰和矿粉集135组配合比的2/3样本拟合、1/3样本预测实验,表明该模型预测精度明显优于线性回归,稳定性优于BP人工神经网络。进一步的自变量敏感性分析表明,RS-SVM模型能准确地反映混凝土性能对原材料变化的依赖性,总体性能最优。

This study develops a new prediction model of concrete properties that uses independent variables of all dosages of raw materials and their qualities and dependent variables of initial slump and cubic compressive strengths at the age of 7,28 days.In this model,a rough sets（RS） method is adopted for reduction of the independent variables,and a method of support vector machine regression（SVR） for prediction of concrete properties.Two groups of sample are collected and tested,one containing 169 sets of mixture with fly ash as admixture and the other containing 135 sets of mixture with fly ash and slag as admixture.Among these groups,one third of the mixtures are randomly selected for prediction verification.The test results show a better prediction accuracy of the new model than the linear regression method and more stable performances than the BP artificial network model.Through a sensitivity analysis it is revealed that the RS-SVM model is able to reflect the dependencies of concrete properties.