【作者】 刘代飞；

【导师】 李劼； 丁凤其；

【作者基本信息】 中南大学 ， 有色金属冶金， 2008， 博士

【摘要】 目前,我国铝工业迅速发展,氧化铝产量已达1900万吨/年。围绕节能减排,开展氧化铝行业技术创新的需求日益迫切。氧化铝焙烧是对氧化铝产、质量和生产能耗有重大影响的工序之一,目前该工序已普遍采用气态悬浮焙烧工艺。众多气态悬浮焙烧生产表明,该工艺在设备配置、操作调节和过程控制等方面仍有很大改进潜力。对焙烧过程开展设备、操作和控制的优化研究有利于实现焙烧生产的增产、节能和降耗。本文在国家自然科学基金的资助下,以年产5万吨的气态悬浮焙烧炉为试验对象,集成应用FLUENT、人工神经网络、遗传优化、模糊控制、专家系统等技术,对氧化铝焙烧过程开展设备、控制和指导的整体优化。研究成果主要有:（1）针对焙烧燃烧系统缺少配置依据,开展炉体燃烧优化的仿真研究。采用FLUENT对主炉P04仿真研究得出:某燃料的最佳空燃比值（A/F）以及低氧完全燃烧对应的最佳操作条件;最佳下料区域为Ⅳ部炉体,最佳V08预热烧咀布置区域为Ⅱ部炉体;保持V08烧咀小比例投入燃料有利减少NO生成;提高空气预热温度节能效果明显。仿真得到NO_x、CO、CO₂等废气生成量,为生产操作提供重要参考。（2）针对焙烧旋风器工况分析的不足,开展气固分离研究。对预热旋风器P01采用雷诺应力输运模型求解气场,拉格朗日坐标求解颗粒运动轨迹。计算不同的工况风速、温度、漏风率和物理结构下旋风器分离效率,探讨了P01环流式旋风器和收尘锁气设备改造方案,为操作提供优化参考。（3）针对现有描述焙烧过程模型的缺乏,提出采用神经网络（ANN）、遗传算法（GA）、灰色模型（GM）优化建模,建立温度预测、废气软测量评价和产能评估三大过程模型。温度预测模型由GM（1,1）与ANN组合优化实现,绝对误差±5℃评价模型,预报命中率达90%以上,可以指导生产调节。废气软测量模型结构为ANN{3-5-4},用绝对误差小于1评价模型,预测准确率达88.6%。基于FLUENT仿真结果对新工况排废的预测,具二次仿真性。产能评估模型结构为ANN{3-9-1},用相对误差小于1%评价模型,预报准确率达96%。产能ANN模型比回归模型更能揭示系统关系。（4）针对焙烧过程常规、单一PID控制方式的不足,提出并建立了焙烧过程模糊专家控制系统。设计了一种Complex-PID控制器和空燃比专家调节器,并提出了一种焙烧过程分段调节控制策略。其中,控制器由FNN单元、PID单元和阈值调节单元组成,采用模糊方法、神经网络和遗传算法对PID进行调整,保证具有最优或次优控制参数。调节器综合数值模拟、视频监控和烟气氧量等反馈信息寻优调节。分段调节控制策略实现了不同工况下温度的优化控制,精度达±5℃,稳定了炉况。（5）针对焙烧生产和管理工作的不完善,提出并架构了焙烧过程ANNES指导系统。采用产生式规则表示过程显式知识,ANN模型表示隐式知识,两类知识由隶属函数实现转化。建立风机故障、燃烧调节和状态分析知识库,实现了燃烧和过程的分析和监测;建立GA-ANNES优化模型库,实现了过程能耗分析,解决了高产低耗参数优化问题;建立旋风器操作指导知识库,实现了旋风分离ANNES分析诊断和操作优化。（6）开发了基于PLC的SCADA系统和基于VC++、Matlab的集成优化系统。两系统间的通讯采用OPC技术、自定协议和DeviceNet总线方式实现。PLC系统实现基础控制,优化系统集成神经网络、遗传算法、专家系统实现过程的优化和控制。本文开发的集成优化系统在年产能5万吨气态悬浮焙烧炉工业试验中取得很好的优化效果:热耗降低了14.3%,达到了3.09MJ/kg;主炉温度降低了8.8%,控制在1040±5℃;含氧量降低了75%,控制在1～2%;NO排量降低了53.9%,控制在53ppm。更多还原

【Abstract】 At present,the scale of China’s alumina industry is developing rapidly and the productivity of alumina has been up to nineteen million tons per year.In order to achieve energy saving and emission reduction, it is imperative to carry out technological innovation.In the roasting procedure that affects the quality,yield and energy consumption,the gas suspension calcination process（G.S.C） which stands for the main developing trend in the fluidized calcinations has been widely adopted. Many applications indicate that it still has much improving space in the equipment configuration,manual adjustments and process control for the G.S.C.The optimization studies of equipment,operation and control can improve production and reduce energy consumption.This dissertation is supported by the National Natural Science Foundation of China and subjected in G.S.C with the productivity of fifty thousand tons per year,in which an integrated optimization strategy combining FLUENT,artificial neural networks,genetic optimization, fuzzy control and expert system technology is presented for equipment, operation and control optimization.The original research achievements of this paper are listed as follows:（1）In order to solve the problem of the scarcity in configuration information for combustion system,a three-dimension G.S.C model is studied by using a CFD software FLUENT.The obtained optimal results include:For a certain fuel,the optimal ratio of air to fuel（A/F） and optimal operation condition for low oxygen complete combustion are obtained;The optimal feeding inlet position is on partⅣand optimal fixing position of V08 igniter is on partⅡof G.S.C;By setting the ratio of V08 to V19,NO emission declines; With increased preheated air temperature,considerable energy can be saved.The quantity of NO_x,CO and CO₂ emission estimated through simulation can serve as a very good reference to the production.（2）As the operation of cyclone separator is not fully explored,a three-dimensional P01 Grid model obtained by hexahedral approach in Gambit was applied in the simulation.With Reynolds stress turbulent viscous model,the physical field distributions and the particle tracks were computed through the gas phase coupled with the discrete phase.The separation efficiency in different inlet winds velocity,operation temperature,air leakage rate and physical structure are obtained and the reconstruct of P01 adopted circumfluent cyclone separator and certain dust collection equipments are discussed,which provides important reference for the optimization of operation.（3）In order to extend the existing process description modes,artificial neural networks（ANN）,genetic algorithms and grey theory model（GM） were adopted to optimize G.S.C modeling.A temperature combinatorial optimal prediction model,emission soft measurement model and productivity estimation model are obtained. The temperature model was established based on GM（1,1） and ANN. The forecast accuracy of combinatorial model is over 90%as the absolute error±5℃is adopted,which meets the requirements of production adjustment.The structure of emission soft measurement model is ANN{3-5-4} and its forecast accuracy is 88.6%as the absolute error less than 1 is adopted.The prediction of new operation conditions deduced by the emission ANN model based on FLUNET simulation results can be taken as secondary simulation.The productivity estimation model is ANN{3-9-1} and its forecast accuracy is 96%as the relative error less than 1 is adopted,which is more advanced in process expression than the existing regression model.（4）To improve the conventional PID control,a roasting process fuzzy-expert control system was established.A Complex-PID controller and an expert regulator for A/F are designed.The Complex-PID controller is composed of fuzzy neural network（FNN）, PID and threshold switch unit,by which the optimal or suboptimal control parameters can be obtained with fuzzy rules,neural networks and genetic algorithms.With the expert regulator for A/F,the optimal A/F value can be deduced by combining numerical simulation results, image analysis and oxygen content feedback information.A set of roasting process subsection adjustment control strategy is presented to achieve various conditions optimization,of which temperature error is within±5℃so as to create stable furnace conditions.（5）In order to extend the existing production and management modes,a roasting process guidance system based on ANNES was built.In this system,the explicit information is expressed by production rules and the implicit information by ANN model.And the transform between the two kinds of information is achieved by Membership function.A fault information base for centrifugal blower and Roots blower and a state regulation information base for combustion and process analysis are established,to which process analysis and monitoring can be achieved.A GA-ANNES optimization model was built to achieve energy analysis and to provide a good solution for higher yield with lower consumption.A guiding information base for cyclones separator operation is created and cyclones diagnosis and analysis are performed by ANNES.（6）A calcination SCADA system based on PLC and an integrated optimization system based on VC++ and Matlab are developed.The communications between the two systems were performed by three kinds of modes:OPC technology,consumer defined protocol and DeviceNet Bus.The PLC system performs as basic control and optimization system integrated neural networks,genetic algorithms and expert knowledge achieves process control and optimization.In this paper,the integrated optimal control system has been applied in G.S.C with the productivity of fifty thousand tons per year and the industrial tests have achieved satisfactory results:the heat consumption ups to 3.09MJ/kg,a decrease of 14.3%;the furnace temperature is controlled within 1040±5℃,a decrease of 8.8%;oxygen content is controlled within 1-2%,a decrease of 75%;NO emission is around 53ppm,a decrease of 53.9%.更多还原