基于CART-SVR模型的混凝土抗压强度预测研究

Prediction method of concrete compressive strength based on CART-SVR

为了提高混凝土抗压强度预测精度,根据影响因素与抗压强度的变化规律,提出了1种基于CART算法优化SVR回归算法的预测方法。SVR算法基于结构风险最小原理来提升泛化能力,由于惩罚因子人为设置,该因子设置过大会导致训练欠拟合,过小会导致预测精度不足。主要改进是通过CART决策树中的剪枝算法生成一系列惩罚因子,进而交叉验证生成最优决策树,进行属性约简,剔除冗余信息,实现数据降维。使用CART-SVR模型与AB、AB-KNN、RF、GBM模型对比试验,均方根误差(RMSE)至少减少了6.7%,平均减少了40.1%。研究结果表明,该模型具有生成最优惩罚因子、优化SVR参数的优点,为实现混凝土抗压强度的高精度预测提供了新的思路。

In order to improve the prediction accuracy of concrete compressive strength,based on the data of concrete influencing factors and compressive strength,the variation law of compressive strength was mined,and a concrete compressive strength prediction method based on classification and regression tree(CART)optimized support vector regression(SVR)was proposed.The SVR algorithm improves the generalization ability based on the principle of minimum structural risk.Due to the artificial setting of penalty factor,too high setting of this factor will lead to inadequate training fitting,and too small will lead to insufficient prediction accuracy.The main improvement is to generate a series of penalty factors through the pruning algorithm of CART decision tree,and then cross-verify to generate the optimal decision tree,so as to reduce attribute reduction,eliminate redundant information,and effectively reduce model complexity.Compared with AB,AB-KNN,RF and GBM prediction models,root-mean-square error(RMSE)was reduced by at least 6.7%and 40.1% on average.The results show that the model has the advantages of generating the optimal penalty factor and optimizing the SVR parameters,which provides a new idea for the high-precision prediction of concrete compressive strength.