超大能力超细全尾砂长距离自流输送临界流速ELM预测  

ELM prediction of critical flow velocity in large-capacity long self-flowing transportation of super fine tailings slurry

作　　者：王新民[1] 张国庆[1] 张钦礼[1] 李帅[1] 

机构地区：[1]中南大学资源与安全工程学院,长沙410083

出　　处：《科技导报》2015年第15期27-31,共5页Science & Technology Review

基　　金：国家科技支撑计划项目(2008BAB32B03)

摘　　要：为准确预测司家营铁矿超大能力超细全尾砂浆体长距离管道自流输送的临界流速,对比传统的BP神经网络、支持向量机(SVM),建立了以管道直径、物料平均粒径、浆体体重和体积浓度为输入因子,临界流速为输出因子的极限学习机(ELM)预测新模型。研究结果表明,ELM模型与SVM模型的相对误差均控制在5%以内,远低于BP神经网络模型的9.56%。由于隐层节点参数均随机选取且无需调节,使得ELM算法在隐层节点数为110和200时,训练时间仅为0.02 s和0.05 s,远少于同节点状态SVM模型的0.04 s和0.095 s,且隐含节点数越多,训练时间差距越大,运算效率越高。To accurately predict the critical flow velocity of Sijiaying＇s large-capacity super fine tailings slurry in long self-flowing transportation, a new ELM prediction model is developed. The ELM model takes pipe diameter, grain diameter, slurry density and volume concentration as input factors, and critical flow velocity as output factor. By comparing it with traditional BP neural networks and support vector machines （SVMs）, the superiority of ELM in improving precision and efficiency is demonstrated. It is revealed that ELM model＇s relative error is blow 5%, which is lower than BP model＇s 9.56%. With the hidden node number being 110 and 200, the training times of ELM are 0.02 s and 0.05 s, respectively, which both are far below the corresponding SVM＇s 0.04 s and 0.095 s. The random choice and good adaptability of hidden node number makes the new ELM model superior in improving precision and efficiency.

关 键 词：超大能力 临界流速 极限学习机 预测精度 运算效率 

分 类 号：TD853.343[矿业工程—金属矿开采]