一类连续化工生产过程的模型辨识及非线性预测控制研究-化工过程机械专业论文.docx

预测控制。分别针对不同的模型延迟分析高斯过程的动态建模效果，比较其计算时间和预测性能，选择最为合适的延迟建立最终的高斯模型，输出其模型预测结果。将辨识得到的高斯过程模型作为预测模型，构建非线性预测控制算法，并用于MMA聚合过程数均分子量的跟踪控制。5.针对基本粒子群算法容易出现“早熟”收敛的缺点，提出一种改进的混合遗传粒子群算法，在基本算法的迭代中引入淘汰机制，将满足淘汰条件的粒子与适应度最优的粒子进行多后代择优交叉和一定概率的变异，以期得到适应度更优的新粒子代替被淘汰粒子，从而提高算法的全局寻优能力。将混合遗传粒子群算法用于有约束线性广义预测控制的控制律求解中，仿真验证其有效性。在此基础上，提出了一种基于NARIMAX模型的有约束非线性广义预测控制方法，给出了非线性广义预测控制的参数在线辨识方法和基于混合遗传粒子群算法的控制量求解步骤，并用于开环不稳定生物反应器的控制。关键词：连续化工过程；模型辨识；非线性预测控制；粒子群算法；AbstractModelidentificationandcontrolofchemicalprocessesisaresearchhotspotinthefieldofprocesscontrol.Chemicalprocessesusuallyhavetime-varyingandhighnonlinearcharacteristics,soitisdifficulttodevelopfristprinciplemodeloftheprocesses.Controlperformanceoftemperature,productconcentrationandaveragemolecularweightsaffectthequalityandquantityoftheproducts,soitissignificanttoresearchthechemicalprocessmodelidentificationandcontrol.Modelpredictivecontrol(MPC)methods,whichdevelopedinpractice,havebeenbroadlyappliedinmanyfields.Thetraditionalpredictivecontrolmethodsareusuallybasedonthelinearmodel.SatisfactoryperformancesaredifficulttobeobtainedusinglinearMPCalgorithmsbecauseofthetime-varyingandhighnonlinearcharacteristicsofthechemicalprocesses,sononlinearmodelpredictivecontrolshouldbedeveloped.Typicalcontinuouschemicalprocessesshuchascontinuousstrirredtankreactor,pHneutralizationprocessandMMApolymerizationreactionareusedinthedissertation.Input-outputdataoftheprocessesareusedtoidentifythemodeltorepresentthesystemdynamiccharacteristics,andthenthenonlinearmodelpredictivecontrollersaredevelopedbasedontheidentifiedmodel.Also,particleswarmoptimizationalgorithmisusedtosolvethepredictivecontrollawoftheconstrainedMPC.Detailsofthecontentsarelistedasfollows:Step-responsesandstabilitycharacteristicsofcontinuousstrirredtankreactorareanalysised.MechanismmodeloftheCSTRisidentifiedtoaLSSVM-ARXHammersteinmodel,andtheidentificationresultsarecomparedwiththetraditionalpolynomialbasedHammersteinmodel.AnonlinearmodelpredictivecontrollerisdesignedbasedontheLSSVM-ARXHammersteinmodel,optimiz