基于极限学习机的充填体强度预测

Prediction of Backfill Body Strength based on Extreme Learning Machine

针对充填体强度与影响因素之间复杂的非线性关系,利用极限学习机算法(ELM)建立了充填体强度预测的ELM模型,该模型以良山铁矿充填料浆配比试验数据作为训练和测试样本,选取料浆浓度、灰砂比、水泥含量及矿渣微粉掺入量4个影响因素作为输入因子,28天抗压强度作为输出因子。结果表明:当隐含层节点数目选取为17个,激活函数为Sigmoidal时网络模型具有良好的泛化能力和预测精度,模型训练值和预测值与实测值的均方误差(MSE)分别为0.39和0.36,平均相对误差控制在5%以内,对充填体的强度具有良好的预测能力。

Considering the complex nonlinear relations between backfill body strength and its influencing factors, this paper establishes a predicted model of backfill body strength with extreme learning machine （ELM） applying ratio test data of backfill slurry in Liangshan iron mine as the training and test samples Four influencing factors, including slurry density, cement-sand ratio, cement contents and slag powder are adopted as input variables, while 28 day-compressive-strength as output variable. The result shows that the network model has well-generalization capability and accurate quantitative when the numbers of hidden nodes is 17 and activation function is Sigmoidal. The training value and predicting value＇s mean square error （MSE） are 0.39 and 0.36 respectively. The average relative error is controlled within 5 %. It shows the ELM prediction model is favorable for predicting the backfill body strength.