基于BP神经网络的氧化铝溶出苛性比值预测方法  

Precast method of digestion RSA based on BP neural network

作　　者：张羽飞[1] 李来时[1] 杨琦[1] Zhang Yufei Li Laishi Yang Qi(Shenyang Aluminum and Magnesium Engineering and Research Institute Co., Ltd., Shenyang 110001, China)

机构地区：[1]沈阳铝镁设计研究院有限公司,辽宁沈阳110001

出　　处：《轻金属》2017年第9期53-58,共6页Light Metals

摘　　要：针对拜耳法生产氧化铝的溶出工序中苛性比值难以实时获得的问题,提出了一种通过BP神经网络模型对溶出苛性比值进行提前预测的方法。根据溶出生产物料输入输出关系,提出了基于物料平衡的溶出苛性比值机理计算模型,并对计算模型的输入变量进行优化,使其能够满足BP神经网络输入的要求。最终设计了一种具有误差反传学习及历史数据训练功能的BP神经网络,经山西某铝厂实际数据测试,BP神经网络能够较好的预测溶出苛性比值。A method is proposed to forecast the RSA （ Ratio of Soda to Aluminate） of digestion by Back Propagation Neutral Network （BPNN） model as it is difficult to acquire real time RSA from digestion in Bayer Process. A calculation model, based on the mechanism of digestion RSA under mass bal- ance, is also proposed in line with the correlation between input and output of the production materials for digestion, and the input variable of the proposed calculation model is optimized so as to meet the input requirements of BPNN. A BPNN with the function of error back propagation study and historical data training is designed. Through a test conducted in an alumina refinery in Shanxi Province, BPNN is capable of forecasting the RSA of digestion.

关 键 词：氧化铝 溶出苛性比值 模型 BP神经网络 预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TQ133.1[自动化与计算机技术—控制科学与工程]