SVM模型预测充填体强度

SVM Model to Redict Strength of Filling of Cemented Tailings Backfill

为了准确评价和预测充填体强度,提出基于支持向量机(Support vector machine SVM)的强度预测模型,建立了充填体强度与影响因素之间的SVM模型(建立胶凝材料掺量与胶砂比与充填体28d单轴抗压强度),首先分别采用不同的寻优算法来计算模型关键参数即核函数的参数g和惩罚系数c,分析每种算法的结果参数与预测结论影响,从中选出最优对充填体预测强度的参数寻优算法SVM回归模型。通过分析:PSO算法、GA寻优算法、网格寻优的训练集回归参数各为0.88931、0.9727、0.88605。PSO算法、网格寻优、GA寻优算法的测试集回归系数分别为0.9842、0.97868、0.86683,三种算法中GA-SVM的泛化能力较差,PSO算法的推广性较强。改进的GA-SVM模型的泛化能力较强,预测得到的相关系数R=0.9877,高于传统GA寻优算法。SVM为充填体强度预测提供了新的技术手段。

In order to evaluate and predict the cemented tailings backfill strength accurately,strength prediction model based on support vector machine(support vector machine SVM)is proposed.By use of 18 groups backfill laboratory uniaxial compressive strength test results,the models of a cemented tailings backfill strength and influencing factors SVM(input is more than mortar and various cementitious materials content,the output is uniaxial compressive strength backfill of 28 d)is established,respectively,using a grid parameter optimization,genetic algorithm(GA)parameter optimization,particle group(PSO)optimization algorithm to calculate parameters of SVM model,comparative analysis parameters obtained three algorithms to determine the most suitable parameter optimization algorithm cemented tailings backfill strength prediction of SVM regression model,results showed,GA-SVM model for restorations the intensity has good predictive capability,network model is not only fast convergence and high precision training,real data correlation coefficient of 0.9877 and test data provides a new method for the prediction of the cemented tailings backfill strength.