磨浆过程输出纤维长度随机分布预测PDF控制

Predictive PDF Control of Output Fiber Length Stochastic Distribution in Refining Process

磨浆过程作为制浆和造纸工业最为重要的生产环节之一,其输出纤维长度随机分布(Fiber length stochastic distribution,FLSD)形状直接决定着后续造纸过程的能耗和纸品质量.针对传统的均值和方差难以描述输出FLSD特征,即具有非高斯分布特性,本文提出一种磨浆过程输出FLSD的预测概率密度函数(Probability density function,PDF)控制方法.首先,采用径向基函数(Radical basis function,RBF)神经网络逼近输出FLSD的PDF,在采用迭代学习方法完成基函数参数整定基础上对相应权值向量进行估计.其次,针对权值之间存在强耦合特点,利用随机权神经网络(Random vector functional-networks,RVFLNs)建立表征输入变量和权值之间关系的预测模型.最后,基于建立的输出FLSD模型设计预测PDF控制器,最终实现对期望输出PDF的跟踪控制.基于工业数据实验验证了所提方法的有效性,为磨浆过程优化运行和控制提供理论依据.

As one of the most important production links in the pulp and papermaking industry, the output fiber length stochastic distribution (FLSD) shaping of the refining process directly determines the energy consumption and paper quality of the subsequent papermaking processes. The traditional mean and variance are insufficient to describe the characteristics of the output FLSD, which displays non-Gaussian distributional properties. This paper proposes a predictive probability density function (PDF) control method for the output FLSD in the refining process. Firstly, use the radical basis function (RBF) neural network to approximate PDF of the output FLSD, the iterative learning technique is utilized to tune the parameters of basis functions, and the corresponding weights vector can be estimated. Secondly, in view of the strong coupling between these weighting vectors, the random vector functional link networks (RVFLNs) are employed to characterize the prediction model between the input variables and the weights. Finally, we design the predictive PDF controller based on the established output FLSD model, thus realizing the tracking control of the desired output PDF. Furthermore, the industrial data based experiment verifies the effectiveness of the proposed method, which provides theoretical basis for optimal operation and control of the refining process.