摘要
为了得到最优的絮凝沉降参数,以絮凝沉降正交试验数据为训练样本和检验样本建立BP神经网络预测模型。絮凝剂单耗、料浆浓度及絮凝剂浓度作为输入因子,沉降速度和极限浓度作为输出因子。对比隐含层节点数对模型训练过程及预测精度的影响,选取最佳预测模型节点数为9。将絮凝沉降参数细化输入到预测模型中,从而搜索出优选样本,优选参数絮凝剂单耗为4.5 g/t,絮凝剂浓度为0.11%,料浆浓度为15%。经实验对比,该模型对絮凝沉降参数预测结果的相对误差能控制在5%左右,精确度较高,可以作为絮凝沉降参数优选的一种新思路。
Back-propagation neural network was used to optimize the flocculating sedimentation parameters. To get the best network mode, some learning and training samples were established by the numbered orthogonal blasting tests. In the process of establishing the network mode, the tailings concentration, flocculant consumption and flocculant concentration were used as the input data, the sedimentation speed and limiting concentration were confirmed to be the synthesized output data. Comparison of the influences of hidden layer nodes on model training process and prediction accuracy indicates that the optimal hidden layer node was 9. By entering the refined flocculating sedimentation parameters into the prediction model, optimal samples are searched and the optimal parameters show that the flocculating agent consumption is 4.5 g/t, flocculating concentration is 0.11% and tailings concentration is 15%.Compared with that of the experimental results, the relative error of the prediction results can be controlled at about 5%. The application indicates this mode has relatively high accuracy, providing a new method to optimize the flocculating sedimentation parameters.
出处
《科技导报》
CAS
CSCD
北大核心
2014年第17期23-28,共6页
Science & Technology Review
基金
"十二五"国家科技支撑计划项目(2012BAC09B02)
关键词
BP神经网络
全尾砂
絮凝沉降
动态放砂
back-propagation neural network
unclassified tailings
flocculating sedimentation
dynamic sand release
