【作者】 唐飞；

【导师】 潘海天；

【作者基本信息】 浙江工业大学 ， 化学工程， 2007， 硕士

【摘要】 造纸过程具有大滞后、大惯性、强非线性、时变以及多变量耦合等特点，这给造纸生产过程控制带来了极大困难，各种智能技术的飞速发展为解决此类问题提供了一条比较可行的途径。首先就智能控制技术在制浆造纸生产过程中的应用现状进行了详细综述，其中包括神经网络、模糊逻辑和专家系统以及集成控制技术等，并对智能控制技术的应用研究发展前景进行了评价，指出支持向量机将在制浆造纸生产过程中具有良好的应用潜力。造纸主要包含两大过程——制浆过程和抄纸过程，制浆过程包括蒸煮、洗涤、筛选、漂白工序。抄纸过程包括纸料制备、纸浆流送等过程组成。其中白水系统工艺复杂，白水浓度难以测量。神经网络具有结构可变性、容错性、非线性、自学习性和自组织性等特点，但是基于神经网络的建模技术都必须先选择模型结构，不能动态调整结构，易陷入局部最优解，也不能保证最优性。另外制约上述技术向实用化推广的一个根本原因就在于这些技术都基于经验风险最小化，无法保证充分的样本数量。因此提出支持向量机算法，一种新的机器学习方法，它以统计学习理论为基础，从结构风险最小化原则出发，这与传统学习方法基于经验风险最小化有本质的区别。克服了神经网络训练陷于局部极值的问题；小样本学习使支持向量机具有很强的泛化及推广能力，不必过分依赖样本的数量和质量；支持向量机有别于神经网络还在于它不需要人为设计模型结构，避免了建模过程中人为因素的影响。通过仿真实验，表明了支持向量机在回归问题上比神经网络具有更好的预测性能。到目前为止，支持向量机的参数主要还是依靠经验选择，并没有坚实的理论支持，因此提出了遗传算法进行参数选择。针对造纸工艺过程的复杂性，提出最小二乘支持向量机，由于最小二乘支持向量机失去了稀疏性，因此提出一种改进的最小二乘支持向量机回归方法，简化了模型复杂程度。本文采用改进的最小二乘支持向量机建立白水浓度软测量模型，研究影响白水浓度的因素，深入分析主要影响因素，选择实际生产可采集数据的主要因素，根据采集到主要因素的数据(纸浆流量、纸浆浓度和上网浓度)与白水浓度建立对应关系，实现造纸白水浓度在线估计。实验结果显示，这种方法具有很高的预测精度，预测效果很好，能够优化造纸生产白水浓度工艺条件。充分运用智能控制技术进行造纸生产过程建模和优化研究，不仅为白水浓度研究提供了一种有效的研究手段，还有助于进一步开展造纸过程先进控制、在线优化和故障诊断等方面的研究。本研究工作指出智能技术合将在造纸生产过程中具有良好的应用潜力，为智能技术在化工过程中应用提供一种良好的示范。更多还原

【Abstract】 Due to the fact that large-lag, large inertia, strong nonlinearity, time-varying and multivariable coupling existed in papermaking process.It is difficult to model, control and optimize papermaking process. Thedevelopment of all kinds of intelligent technologies provide a feasibleapproach to solve the problem.This paper describes the application status of intelligent controltechniques, such as artificial neural network, fuzzy logic, expert systemand its integration. Then its applications and research developmentprospects are evaluated, and pointed out that support vector machine willexhibit great potentialities in pulping and papermaking process.Papermaking is made of two process—pulping process andpapermaking process, pulping process include cooking, washing, screening and bleaching process, papermaking process include Preparation of stock, Approach of stock and so on. White water system iscomplex, white water consistency is measured difficulty.The nerve network has structure variability, fault-tolerant, non-linearity, self-organizing and self-learning etc, but all the technologybased on artificial neural network must first choose the model structure, easily getting a local optimum solution, also cannot guarantee optimality.Moreover the basic reason lie in these technologies based on theexperience risk, is unable to guarantee the full sample quantity. Thereforeproposed the support vector machine, a new kind of machine learningmethod. This machine learning method is basis of the statistical theoriesand the structure risk minimum principle, it is difference in essential withthe traditional study method based on the experience risk. Overcame thenerve network training to sink into the local optimum solution; The smallsample study may enable the support vector machines having favorablecapabilities of universality and generalization, does not need excessivelyto rely on the quantity and the quality of sample. The vector machine liein dose not need the artificial design model structure, is different from thenerve network, avoids artificial influence in the modeling process. It canbe seen from the simulation results that the support vector machine canmake better predictions of regression than the nerve network.But so far, parameter of support vector machine choose was mainlydepended on their experience, without a theoretical basis, Therefore proposed Genetic Algorithm carried on the parameter choice. In view ofthe complex papermaking technological process, Least square supportvector machines was proposed. It lost its original sparseness, so animproved Least square support vector machines regression method wasproposed simplified complex degree of model. This article use the leastsquares support vector machine to establish soft sensing model of whitewater consistency, Study Influential factors of white water consistency, In-depth Analysis on main influential factors, choose main influentialfactors which could acquire data in actual production, Establishingcorresponding relationship between data acquisition of main influentialfactors (flow of pulp, consistency of pulp and stock consistency on wire)and white water consistency, realization white water consistency on-lineestimate in paper making process. The test show that the methodpossesses high prediction precision and excellent forecasting effect, it canoptimize parameter of white water consistency in the papermakingprocess.An effective research approach about modeling and optimization ofthe papermaking process by using intelligent technology is provided. Theresults might be helpful to correlative study (advanced control, on-lineoptimization, diagnosis, etc.) in papermaking process. The achievementspointed out that intelligent technology will exhibit great potentialities inpapermaking process, and provide an instantiation for the application of intelligent technology in chemical process.更多还原