内循环流化床生物质气化过程的神经网络模型  

BP NEURAL NETWORK SIMULATION OF BIOMASS GASIFICATION IN AN INNER CIRCULATING FLUIDIZED BED

作　　者：马隆龙[1,2] 陈平[2] 原晓华[2] 阴秀丽[2] 吴创之[2] 颜涌捷[1] 

机构地区：[1]华东理工大学 [2]中国科学院广州能源研究所,中国科学院可再生能源与天然气水合物重点实验室,广东省新能源和可再生能源研究开发与应用重点实验室,广州510640

出　　处：《太阳能学报》2007年第12期1354-1359,共6页Acta Energiae Solaris Sinica

基　　金：国家高技术研究发展计划(863)项目(2003AA514010);广东省自然科学基金团队项目(003045);广东省科技计划项目(2004A11007002)

摘　　要：基于BP人工神经网络原理,利用MATLAB神经网络工具箱,以实验得到的57组气化实验数据作为样本,建立了一个以加料量和送风量为输入变量,以燃气热值、产气率、碳转化率和气化效率为输出变量,用于描述连续稳定气化过程的内循环流化床生物质气化模型。对模型的隐层节点数和训练周期改变对模拟结果的影响进行了分析,发现当隐层节点数为20,训练步骤为50步,模型的4个输出变量的模拟结果与实验结果相关系数均超过0.95;同时对该模型的预测能力进行了考察,模型预测结果与实验结果吻合良好,证明了该模型具有较强的泛化能力,为生物质内循环流化床气化系统的优化设计和自动控制提供新思路。A three layers back propagation （BP） Neural Network model was built to simulate the biomass gasification process in an inner circulating fluidized bed. Two input variables, i.e. feeding rates, air flow rate, and four output variables, i.e. gas heating value, gas productivity, carbon conversion rate, and gasification efficiency were selected. 57 experimental data were taken as training and checking samples, the effects of nodes of hidden layer and training echoes on simulation results were investigated. The results showed that correlation coefficient of the four output variables between simulation results and experimental data exceeded 0.95, when the nodes of hidden layer were 20 and training echoes was 50. Model-predicted results were in agreement with the experimental data, showing good generalization capacity. This model will be the basis of automatic control of biomass gasification process in fluidized bed.

关 键 词：生物质气化 内循环流化床 BP神经网络 

分 类 号：TK6[动力工程及工程热物理—生物能]