基于BP神经网络的中心提升管内循环流化床颗粒循环流率预测  

PREDICTION OF SOLID CIRCULATION RATE IN INTERNALLY CIRCULATING FLUIDIZED BED WITH DRAFT TUBE BASED ON BP NEURAL NETWORK

作　　者：陈鸿伟[1] 史洋[1] 尹萍[1] 危日光[1] 高建强[1] 

机构地区：[1]电站设备状态监测与控制教育部重点实验室(华北电力大学),保定071003

出　　处：《太阳能学报》2012年第10期1738-1742,共5页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(50876030);高校博士点基金(20090036110008)

摘　　要：基于BP人工神经网络原理,利用MATLAB神经网络工具箱,以试验得到的243组数据作为样本,建立一个以提升管风速、鼓泡床风速、鼓泡床物料静床高、床料平均粒径为输入变量,以颗粒循环流率为输出变量,用于预测中心提升管内循环流化床颗粒循环流率的BP神经网络模型。对模型的隐含层层数和隐含层节点数对预测结果的影响进行分析,发现在隐含层层数为1,隐含层节点数为15时,模型预测结果误诊率最小,预测相对误差在±8%以内,总体平均偏离度为3.09%,网络性能最优,从而为中心提升管内循环流化床装置的设计和运行提供指导。A BP Neural Network model was built to predict the solids circulation rate of internally circulating fluid- ized bed with draft tube based on MATLAB neural network toolbox. Four input variables, i.e. gas velocities to the draft tube or annulus section, static bed height, particle size and one output variable were selected. 243 experimen- tal data were taken as training and checking samples, the effects of number and nodes of hidden layer on predicted results were studied. It is found that the misdiagnosis rate will reach the minimum value with a relative error within _＋ 8%, the overall mean diversion extent being no more than 3.09%, the property of network achieving optimum condition when the number of hidden layers is one and nodes of hidden layer are 15. The model will guide the de- sign and operation of the internally circulating fluidized bed with draft tube.

关 键 词：提升管 内循环流化床 颗粒循环流率 BP神经网络 

分 类 号：TK229[动力工程及工程热物理—动力机械及工程]