磨浆过程作为造纸工业最为重要的环节之一，主要为后续造纸环节提供满足相应物理特性的纸浆。纸浆是整个造纸工业重要的生产原料，其质量决定了最终纸品的诸多性能，如纸品伸缩率、抗张强度及撕裂度等。实际工业生产中，磨浆过程纸浆质量主要取决于纤维长度和游离度(CSF)。若纤维长度过短或游离度偏低，则磨浆过程能耗高、脱水率低，并且由于纤维含水量过高，也会造成后续造纸环节纸品烘干过程的能耗显著升高；而若纤维长度过长或纸浆游离度太高，虽然有利于提高纸品生产的脱水效率和降低烘干过程的能耗，但过大的孔隙率将会严重影响均匀度、强度及平滑度等纸品质量指标。因此，磨浆过程尤其是纤维长度与游离度的控制直接关系到整个造纸的纸品质量和运行能耗。然而，实际磨浆过程中由于纤维之间的相互揉搓、挤压、分丝、切断帚化以及设备老化等不确定性因素的影响，使得纸浆纤维长度分布形状具有较强的非高斯随机分布动态特性，而纤维长度的均值并不足以完全描述纤维长度分布的统计信息，因此，传统纤维长度的均值控制方法难以有效控制纤维长度分布形状。为了提高纸浆质量、降低运行能耗和生产成本，需要研究针对具有非高斯随机分布动态特性的磨浆过程控制方法。 针对当前磨浆过程存在的问题和控制方法研究的不足，依托国家自然科学基金重点项目“面向节能降耗和纤维形态分布的制浆过程运行优化控制(61333007)”，以当前造纸工业中广泛的化学热磨机械制浆方法下典型的磨浆过程为研究对象，开展基于随机分布控制的磨浆过程控制方法研究，主要工作如下： 1)针对基于纤维长度均值的传统控制方法难以有效控制具有非高斯随机分布动态特征的磨浆过程纤维长度分布形状的问题，将数据驱动建模与控制和随机分布控制(SDC)相结合，研究基于SDC的磨浆过程纤维长度分布形状控制的系列方法，包括几何分析双闭环迭代学习控制(ILC)方法和数据驱动预测概率密度函数(PDF)控制方法。 A.针对难以通过机理分析方法建立磨浆过程纤维长度分布形状数学模型，进而实现纤维长度分布形状控制的难题，将数据驱动建模和控制相结合，集成数据驱动子空间参数辨识、几何分析双闭环迭代学习控制与随机分布控制(SDC)等理论方法，提出基于几何分析双闭环ILC的磨浆过程纤维长度分布形状控制方法。所提方法针对具有非高斯随机分布动态特性的磨浆过程特有的时空特征设计双闭环控制结构，分别基于ILC原理通过跟踪误差实现输出PDF模型和控制量的迭代更新。在内回路，基于ILC原理构建磨浆过程纤维长度分布形状的PDF模型和参数自适应调节机制，包括权值向量的线性子空间参数辨识和基于迭代学习的基函数参数自适应调整；在外回路，通过引入基于几何分析的ILC方法实现磨盘间隙、稀释水流量等控制量的更新，从而提升闭环控制系统的收敛速度等控制性能。基于工业数据仿真实验结果表明所提方法通过磨浆过程空间变量(基函数形状)和时域变量(权值向量)的混合动态控制，实现了磨浆过程纤维长度分布形状控制。 B 针对磨浆过程运行工况时变，采用传统线性子空间辨识建立纤维长度分布形状动态PDF模型存在泛化能力弱、精度不高的问题，集成智能建模、数据驱动预测控制以及随机分布控制，提出基于数据驱动预测PDF控制的磨浆过程纤维长度分布形状控制方法。首先，利用智能建模方法构建表征控制输入与权值向量之间动态关系的非线性模型；然后，通过引入基函数参数的迭代学习更新机制，根据动态PDF模型误差对基函数参数进行自适应调节；最后，在构建纤维长度分布形状PDF模型基础上，综合随机分布控制和数据驱动预测控制，将控制器设计转化为求解有约束的最优化问题，基于工业数据仿真实验验证了所提方法的有效性。 2)针对当前磨浆过程除了要获得满足特定要求的纸浆纤维(如纤维长度分布形状、游离度)，还需要尽可能降低运行能耗和生产成本的问题，综合随机分布控制与多目标预测优化控制，研究面向纸浆质量和运行能耗优化的磨浆过程多目标优化控制方法。 A.针对当前磨浆过程仅关注于单一的表征纸浆质量指标滤水性能参数(即纸浆CSF)控制，而忽略了运行能耗及生产成本的问题，提出面向纸浆CSF和运行能耗优化的磨浆过程多目标优化控制方法。所提方法在纸浆游离度与运行能耗之间的非线性机理模型基础上，通过最小二乘参数估计方法构建磨浆负荷、产量等与控制量之间的动态关系模型，从而建立磨浆过程CSF机理与数据相结合的非线性混合动态模型。在此基础上，提出了面向纸浆CSF和运行能耗优化的磨浆过程多目标优化控制方法。基于工业数据实验结果表明所提方法不但使得磨浆过程输出纸浆CSF获得满意的设定跟踪性能，而且能够有效降低过程运行能耗，实现面向输出纸浆CSF和运行能耗优化的磨浆过程多目标优化控制。 B.针对以纸浆CSF或者纤维长度的均值为单一控制目标的磨浆过程控制模式难以全面实现纸浆质量指标有效控制的问题，进一步提出面向纸浆CSF和纤维长度分布形状的磨浆过程纸浆质量指标多目标优化控制方法。一方面，将控制纤维长度分布形状取代传统纤维长度的均值进行控制，从而克服纤维长度的均值作为衡量纸浆质量指标的不足。另一方面，综合数据驱动控制、随机分布控制和多目标预测控制，在构建的纸浆CSF和纤维长度分布形状纸浆质量指标的混合动态预测模型基础上，提出了面向纸浆CSF和纤维长度分布形状的磨浆过程纸浆质量指标多目标优化控制，实现具有时空特性的纤维长度分布形状优化控制以及纯时域特性的纸浆CSF优化控制。基于数据仿真实验验证了所提方法的有效性。 关键词：磨浆过程；纸浆质量；随机分布控制；纤维长度分布形状；概率密度函数；迭代学习控制；数据驱动控制；预测控制；多目标预测优化控制。

 As one of the most important steps in the papermaking industry, the refining process mainly provides pulps meeting the corresponding physical properties for the subsequent papermaking process, and is also a prerequisite for ensuring the final paper quality. Pulp is the key raw material for the entire paper industry, and its quality plays a decisive role in the final paper properties, such as the stretch ability, tensile strength and tear resistance of the paper. The pulp quality in the refining process is mainly characterized by fiber length and water filtration performance (CSF). Specifically, the short fiber length or the low pulp freeness makes the refining process display high energy consumption and low dehydration rate. When the water volume contained in the fiber is too high, the energy consumption of the paper drying process in the subsequent papermaking process also high. On the contrary, if the fiber length is too long or the freeness is high, it is advantageous to form large porosity in the paper sheet, thus increasing the dewatering efficiency in the paper production process and reducing the energy consumption of the drying process. However, the excessive porosity will seriously affect the quality indexes of the paper products, such as the uniformity, the strength and the smoothness. Therefore, the performance of the refining process, especially the control of fiber length and pulp freeness, is not only directly related to both the energy consumption and the paper quality in the subsequent papermaking process, but also directly affects the uniformity and stability of the pulp quality. However, due to the existence of uncertainties such as the mutual entanglement, extrusion, friction, fibrillation between fibers and the process equipment aging, the fiber length distribution shape characterizing the pulp quality shows strong non-Gaussian stochastic distribution dynamics. Using the mean and variance of fiber lengths are not sufficient to fully describe all the statistical information of the fiber length distribution shape. In order to improve the pulp quality, reduce the operating energy consumption and production cost, it is necessary to study control methods for the refining process with non-Gaussian stochastic distribution dynamic characteristics. In view of the shortcomings and the existing problems in the current refining process control methods, this research relies on the National Natural Science Foundation's key project “Optimal Operation and Control of Pulping Process for Energy Saving and Fiber Morphology Distribution (61333007)”. Targeting at the typical refining process under the current popular Chemi-Thermo-Mechanical Pulping (CTMP) method in paper industry, we conduct research on the control methods of the refining process based on stochastic distribution control (SDC) method. The contribution of this thesis is summarized in the following aspects: 1)Since controlling the mean of the fiber length is difficult to effectively control the shape of the fiber length distribution in the refining process with non-Gaussian stochastic distribution characteristics. We combine the SDC with the data-driven control methods, and then study a series of control methods for Probability Density Function (PDF) shaping based on the SDC for fiber length distribution in the refining process. It includes the geometric analysis double closed-loop iterative learning control (ILC) method and the data-driven prediction PDF control method. A .In order to address the challenge that the mathematical model of refining processes is impossible, a data-driven modeling and control method for maintaining the shape of the fiber length distribution of the refining process is proposed, consisting of data-driven subspace parameter identification, geometric analysis-based double closed-loop ILC and SDC. Different from the double closed-loop control structure in the traditional sense, it is a double closed-loop control structure designed with the two-dimensional feature of spatial-domain and time-dimension for the refining process with non-Gaussian stochastic distribution dynamic. The double closed-loop is based on the ILC principle, tracking the error to realize the update of the output PDF model and the control variables. We construct the PDF model of the fiber length distribution shaping in the refining process based on the ILC principle in the inner-loop, including the linear subspace parameters identification of weight vector and the adaptive adjustment of the basis function parameters based on the iterative learning law. In the outer-loop, the control effects (such as the disc gap, the dilution water flow rate) are updated by introducing the ILC method based on the geometric analysis, which further improves the convergence rate of the closed-loop system. The proposed method combines the data-driven modeling method for the output PDF and ILC, and realizes the dynamic control of the PDF shaping of fiber length distribution in the refining process through the hybrid dynamic control of the spatial-domain variable (shape of basis function ) and the time-domain variable (weight vector). The simulation experimental results using data show the effectiveness of the proposed method. B .Concerning the time-varying operation condition in the refining process, the dynamic PDF model of the fiber length distribution shaping using the traditional linear subspace identification often has problems such as weak generalization ability and low precision. Based on the intelligent modeling method and the data-driven predictive control as well as the SDC method, the control method for the shape of fiber length distribution in the refining process is proposed based on the data-driven predictive PDF control. Firstly, utilizing the data-driven intelligent modeling method, the nonlinear model characterizing the dynamic relationship between the control input and the weight vector is constructed. Secondly, the basis function parameters are adjusted adaptively based on the dynamic PDF model error by introducing the iterative learning update mechanism of the basis function parameters. Finally, based on the dynamic PDF model of the fiber length distribution in the refining process, the controller design is transformed into solving the optimization problem with constraints so as to achieve the optimal control of the shape of fiber length distribution in the refining process. The simulation experimental results using data show the effectiveness of the proposed method. 2)In view of the problem that pulp fibers satisfying specific requirements (such as shape of fiber length distribution and pulp CSF) need to be obtained in the current refining process, the operation energy consumption and production cost need to be reduced as much as possible. By Combining the SDC with multiobjective optimal control methods, the multiobjective optimal control methods for the refining process oriented to the pulp quality and the operation energy consumption optimization are studied. A .In order to address the problem that current refining process only focuses on the control of the filtration performance parameters (CSF) of the pulp quality and ignores the operating energy consumption and the production cost, a multiobjective optimal control method for the refining process based on the pulp CSF and the operation energy consumption optimization is proposed. Based on the nonlinear mechanism model between the pulp CSF and the operating energy consumption, the dynamic model between the refiner load, throughput and the control variables is constructed by the least squares parameter estimation, and a nonlinear hybrid dynamic model combining the mechanism of the pulp CSF with data-driven is established. On the basis of the constructed hybrid dynamic model, the multiobjective optimal control method for the refining process aiming at optimizing the pulp CSF and the operating energy consumption is proposed. The simulation experimental results using data show that the proposed method not only makes the output pulp CSF obtain satisfactory setpoint tracking performance, but also can effectively reduce the operating energy consumption, thus achieving the multiobjective optimal control of the refining process for the output CSF and the operation energy optimization. B. To handle with the difficulties existed in fully realizing effective control of the pulp quality indexes through individually controlling the CSF or the shape of the fiber length distribution in the refining process, the multiobjective optimal control method for the pulp quality indexes of the CSF and the fiber length distribution is proposed. By controlling the shape of fiber length distribution instead of the traditional average of the fiber length, we overcome the shortcomings of using the average of the fiber length as a measure of pulp quality index. On the other hand, based on the hybrid dynamic prediction model of the pulp CSF and the PDF shaping of the fiber length distribution utilizing the data-driven modeling method, we propose the nonlinear multiobjective predictive optimal control method for the pulp quality indexes in the refining process. It includes the optimal control of PDF shaping of the fiber length distribution with spatiotemporal features and the optimal control of pulp CSF with pure time-domain. The simulation experimental results using data show the effectiveness of the proposed method. Key words: Refining process; pulp quality; stochastic distribution control (SDC); fiber length distribution shaping; probability density function (PDF); iterative learning control (ILC); data-driven control; predictive control; multiobjective predictive optimal control.