【作者】 王勇；

【导师】 刘吉臻；

【作者基本信息】 华北电力大学（北京） ， 热能工程， 2008， 博士

【摘要】 工业过程控制是通过对生产过程的描述、仿真、设计、控制和管理等方面问题的研究,以改善工艺操作、提高自动化水平、优化生产过程、加强生产管理为手段,最终达到提高生产效益、降低能源损耗、控制有害物排放等目标。目前,随着过程控制技术、计算机技术和其它相关技术的飞速发展,使我们更广泛地获得了过程中产生的大量数据。如果能有效地分析这些信息,就能为我们进一步控制目标的实现提供有益帮助。同时也存在过程变量众多、且具有高度非线性和时变性等难以克服的困难。本文以多变量统计过程控制和数据挖掘为理论,深入地研究了基于部分最小二乘(PLS)和相关方法,并将该方法引入到热工过程控制软测量建模及自适应控制中。论文的主要工作内容和研究成果包括:将PLS算法和多元回归以及主元回归方法进行比较,验证了当数据存在相关性时,PLS算法和主元回归能够很好地解决相关性问题,然后通过一个简单的例子进行了验证。并进一步概要地推导出PLS算法具有更好的拟合性能的结论。研究了递推PLS及其改进算法。以高斯核函数来处理数据的非线性,用聚类方法来自动选择高斯函数的参数,采用递推PLS来调整和更新模型。这样得到的模型即具有较好的非线性处理能力,又能随着工况的改变更新模型以适应过程的变化。针对火电厂辐射受热面污染形成机理不明晰、非线性强且有较大的时间间隔等难题,应用所提出的算法建立电站辐射受热面污染预测模型,并且通过实验检验了模型的有效性。在递推PLS的基础上引入折息因子,得到了折息递推最小二乘算法(DRPLS)。并进一步与动态矩阵预测控制算法相结合,提出基于折息递推最小二乘的自适应控制算法。因为受各种因素影响,模型参数是时变的,通过DRPLS,对模型进行修正可以提高其精度。并应用该方法对某电厂主汽温控制进行了仿真试验,取得了较好的效果。基于Mercer理论对核偏PLS引进了支持向量机来处理内部非线性问题。这样不仅考虑了数据外部的非线性关系,也考虑了输入和输出之间内部的非线性关系,从而使所提出的算法较之一般的非线性PLS算法有更强的非线性处理能力。将该方法应用于火电厂烟气飞灰含碳量软测量模型的建立,并通过实验验证了该算法具有较好的实用价值。更多还原

【Abstract】 Industry process control is to mend techniques manipulation, improve automation degree, optimize production process and enhance production manage through studying on description, stimulation, design, control and manage of production process. Finally, it can achieve the object which improve production efficiency, reduce source consume and control deleterious substance release. Now, following the fast improving of process control and computer technique and so on, we can get more data come from production. If the information can be full extracted and analyzed, it may give us useful help to more control production. However, there are many difficult problems such as a great deal of process variable and nonlinear and time-varying of process. The dissertation deeply studied partial Least Squares method and applied it into soft measure modeling and adaptive control of thermal process control. The main contributions of this dissertation can be summarized as followings:By comparing PLS and mult-irecursive least-square with principal component recursive algorithm, the validity which PLS and principal component recursive algorithm effectively deals with relativity problems when data has relativity is tested. Experiments results show that PLS possess better fitting property.Recursive PLS and its improved methods are investigated. A new recursive Nonlinear PLS is proposed which deal with nonlinear problem with Guess kernel function and update model by recursive PLS. The model obtained by this method have better nonlinear manage ability and update model to fit process variety following condition. Radiation heat-receiving side and pollute side of thermal power plant have difficulty problems of dimness mechanism, strongly nonlinear and big time space. Power station forecasting model is building using proposed algorithm. Experiments verify its effect.Discount recursive PLS (DRPLS) is obtained by using discount factor based on recursive PLS. Moreover, combining DRPLS with dynamic matrix forecasting control algorithm, an adaptive control algorithm based on discount recursive least-square is proposed. Because of factor effect, model parameter is time-varying which can be improved precision by update model with DRPLS. The simulation experiments on superheated steam temperature control of power plant verify its effect convenient.Kernel nonlinear partial least square algorithm based on Mercer theory namely KPLS-LSSVM is proposed. Considering the nonlinear relation is not only being data outside but also being inside data between input and output, the method has more strong process ability than normal nonlinear PLS. The method was applied in building carbon content in fly ash soft measure model of thermal power plant. Experiment verify its effect and convenient.更多还原