煤直接液化间歇反应器性能的人工神经网络模拟  

Simulation of Direct Coal Liquefaction in Batch Reactors by Artificial Neural Networks

作　　者：杨中凯[1] 刘辉[1] 李建伟[1] 李成岳[1] 

机构地区：[1]北京化工大学化工资源有效利用国家重点实验室,北京100029

出　　处：《煤化工》2007年第2期27-30,共4页Coal Chemical Industry

基　　金：国家重点基础研究发展计划(973计划);煤直接液化的反应工程学基础研究(No.2004CB217604)

摘　　要：鉴于煤直接液化反应的复杂性和机理模型难以建立、计算量大等,结合人工神经网络建模方法,建立了煤直接液化间歇反应器的BP神经网络模型,并利用统计极差分析方法考察了影响反应器煤转化率、油收率及综合评价3个指标的8个关键因素的敏感程度。结果表明,训练完成的BP神经网络模型不仅能很好地拟合已知实验结果,而且具有良好的预测能力,预测结果的平均相对误差为3.75%;当煤为烟煤或亚烟煤、溶剂分别为萘满(1,2,3,4-四氢化萘)或循环溶剂时,综合评价给出的影响因素的敏感性顺序为:温度>反应时间>剂煤比>煤粒度>氢分压和填充程度>煤类型>溶剂类型。This paper presents an artificial neural networks （ANN） reactor model of direct coal liquefaction operated in batch mode for simulating the performances of muhiphase reactors with complex reaction kinetics and transport mechanisms. A parameter sensitivity analysis of eight major influencing process factors, including reaction temperature and time, size of coal particles, gas pressure, etc., was made by statistic range analysis. It is demonstrated that the ANN model simulation results are consistent with the experimental data with good predictive capability. The average relative error of calculation outcomes is only about 3.75%. When bitumite or sub-bituminous coal is used as raw material and tetralin or recycle solvent is used as the solvent, the sensitivity Order of the influencing factors to synthesis grade is listed as follows： temperature 〉 reactive time 〉 the size of coal particles 〉 gas pressure and extent of reactor filling 〉 the type of coal liquefied 〉 the type of solvent.

关 键 词：煤直接液化 多相间歇反应器 BP 神经网络 模拟 

分 类 号：TQ529.1[化学工程—煤化学工程]