利用神经网络模型预测球团矿的冷压强度(英文)

建立了一个神经网络模型来预测球团矿的冷压强度,该网络模型采用三层前向BP神经网络,网络结构为12-12-1,12个输入变量分别为给料率、料层高度、焙烧温度、干透点温度、COREX煤气单耗、膨润土的添加量、生球水分、生球碳含量以及成品球的FeO、MgO、Al2O3含量和碱度;隐层含有12个神经元;输出为成品球团冷压强度;神经元激活函数选择双曲正切函数;神经网络学习算法使用的是带惯量项的误差反向传播学习算法(BP学习算法)。选取353组数据来训练和测试神经网络,其中247组数据用于训练网络,其余数据用于测试网络。测试结果表明,该网络的预测结果与实际结果的误差在3%以内,同时通过敏感性分析得出以下结论:①膨润土添加量、生球碳含量以及成品球的FeO、MgO、Al2O3含量和碱度对球团矿的冷压强度有重要影响;②增加膨润土添加量、成品球碱度、MgO含量、焙烧温度、干透点温度、COREX煤气单耗有助于改善球团矿的冷压强度;③增加FeO含量、生球碳含量、Al2O3含量、料层高度、给料率将使球团矿的冷压强度迅速下降;④增加生球水分会降低冷压强度;⑤提高球团矿冷压强度的参数设置(膨润土的添加量:0.86%～0.92%;wFeO<0.5%;生球碳含量:1.00%～1.10%;MgO含量:0.39%～0.44%);⑥在0.3～0.7范围内增加碱度不能显著改善球团矿的冷压强度。

Application of neural network model to predict cold crushing strength of iron ore pellets

JSW Steel Limited is a 10 Mt/a integrated steel plant having two Corex and four blast furnaces (BF) iron making units,two pellet plants and four sinter plants.The major iron bearing feed to Corex units is pellet,whereas,in the BF,the iron bearing charge consists of pellet/sinter/lump ore in the ratio of 30∶ 50∶ 20.The raw material requirement for Corex 1 & 2 and BF 1 and 2 is met by a 4.2 Mt/a pelletising plant and 2.3 Mt/a sinter plant.Hence pellet quality plays a major role in the performance of both Corex and blast furnace 1 & 2.Cold crushing strength (CCS) of the pellet is an important physical property of the pellet and pellet CCS depends on the raw material composition and machine parameters.To get desired CCS of the pellet it is necessary to operate induration machine with optimum parameters.To study the effect of input variables on CCS property,an Artificial Neural Network model has been developed to predict the CCS of the pellets produced by a straight grate induration machine from 12 input variables,namely feed rate,bed height,firing temperature,burn-through temperature,specific Corex gas consumption,bentonite,green ball moisture,green pellet carbon content and FeO,MgO,Al2O3 and basicity (CaO/SiO2) of fired pellets.The variables to which CCS of the pellets was most sensitive were bentonite,FeO,green pellet carbon content,MgO,Al2O3,basicity and green pellet moisture content.CCS of the pellet was influenced by pellet porosity which was itself determined by the firing temperature and green pellet carbon content.The predicted results were in good agreement with the actual data with less than 3% error.