【作者】 李超顺；

【导师】 周建中；

【作者基本信息】 华中科技大学 ， 水利水电工程， 2010， 博士

【摘要】 水电机组控制系统与机组的稳定、安全、高效运行密切相关,对其进行精确建模研究是系统动态过程仿真、自适应控制、机组及互联电网稳定性分析和机组故障诊断的基础。水电机组控制系统是具有时变、非最小相位、强非线性等特点的复杂系统,精确建模一直是相关研究的难点；水电机组日趋大型化、复杂化导致机组故障风险日益突出,机组故障诊断研究一直备受学界关注。因此,深入研究水电机组控制系统辨识理论与方法,获得机组控制系统的精确模型描述,研究机组故障诊断策略,对实现水电机组的安全、可靠和高效运行,具有十分重要的理论意义和工程应用价值。在水电机组系统辨识研究中,传统方法多集中在线性系统辨识领域,辨识的模型多为忽略了非线性环节的简单线性模型,缺乏对非线性系统辨识的研究,制约了高精度水电机组对象模型的获取,亟需进一步发展水电机组非线性辨识理论并系统地建立完善水电机组系统辨识方法体系。为此,在全面分析水电机组特性基础上,凝练出机组控制系统辨识及机组故障诊断所面临的科学问题,结合模糊理论及先进智能优化方法,对水电机组控制系统参数辨识方法、模型整体辨识策略进行了系统深入的研究,进一步开展了基于系统辨识的机组故障诊断研究,提出了基于模糊聚类理论的故障模式识别方法体系。论文的主要工作及创新性成果如下：(1)针对水电机组对象特性及系统辨识研究需求,研究并建立了水轮机调节系统各环节数学模型,对其中的非线性特性进行了重点解析,探讨了不同工况下水电机组控制对象模型表现形式,建立了水轮机调节系统典型线性模型及非线性模型的SIMULINK仿真平台,为系统辨识研究打下了基础。(2)考虑水电机组参数辨识的特殊性,研究并推导了基于微分变换和积分变换的连续系统参数辨识方法,直接辨识对象物理参数。构建了基于Harley变换的连续系统辨识模型,实现了水轮机调速器控制参数的精确辨识,研究了具有结构简单、计算速度快的Haar类正交变换,成功应用于水轮机调节对象的参数辨识。(3)水电机组控制系统是复杂的非线性系统,在不对模型进行简化的情况下进行非线性系统参数辨识。引入引力搜索算法,结合粒子群算法的优点提出了改进引力搜索算法,使其在保留引力搜索的前提下增加了信息共享及记忆能力,进一步提高了搜索能力,在此基础上研究了基于智能优化的非线性系统辨识方法,构造了基于智能优化方法的水轮机调节系统辨识框架,实现了复杂工况下水轮机调节系统非线性模型参数的精确辨识。(4)进一步研究了水电机组复杂非线性系统的整体模型辨识,提出用模糊模型来精确描述水轮机调节系统。在传统T-S模糊模型的基础上,研究并提出用变尺度混沌优化方法来优化T-S模糊的结构与参数,实现结构参数的一体化辨识,为提高模糊空间划分的合理性,提出了基于线性回归原型及超平面原型的模糊聚类方法,实现了T-S模糊模型的高精度辨识,最后验证其在水轮机调节系统辨识中的效果。(5)研究了基于数学模型的动态系统故障诊断策略,为进一步开展该类问题研究打下了基础。进而研究了基于模糊聚类模式识别方法的机组故障诊断,提出了加权混合模糊聚类方法(WCOFCM)和一种加权核聚类算法(WFKC),结合混沌变量的全局搜索能力与梯度算子的局部寻优能力,通过核函数非线性映射及样本特征加权,有效区分了重要和非重要的样本特征,突出了敏感样本特征在聚类中的主导作用,有效实现了机组故障模式的准确识别。更多还原

【Abstract】 The control system of hydropower generating unit is tightly associated with the stability, safety and efficient operation of hydropower generating unit (HGU), and the precise modeling of this system is foundation of dynamic process simulation, adaptive control, stability analysis of HGU and power system. The control system of hydropower generating unit is a non-minimum phase and time-varying nonlinear system, precise modeling of which is a difficult problem. In addition, the trend of HGU to be large and complicated brings significant risk of fault, thus problem of fault diagnosis of HGU is focused by researches widely.In researches of HGU system identification, traditional methods focus on linear system identification based on models which have almost been simplified by omitting nonlinearity sections, and the deficiency of research on nonlinear system identification of HGU restricts the precise modeling of HGU. Thus development of theory of nonlinear system identification and building the system of system identification of HGU are necessary. In this paper, based on comprehensive analysis of models of HGU, scientific problems in system identification of control system of hydropower generating unit (CSHGU) are proposed. Based on fuzzy theory and intelligent optimization methods, parameter identification and system identification of CSHGU are researched, furthermore fault diagnosis strategy based on model and system identification, fault pattern recognition are researched. The main contents and innovative results are listed as follows:(1) In considering of characteristics and system identification of HGU, models of all sections of governing system of HGU are researched and nonlinear sections are analyzed with emphasis. Models in different operating condition of CSHGU are discussed, and SIMULINK simulation platforms of linear and nonlinear models of CSHGU are built, providing foundation of system identification research.According to the special requirements of parameter identification, continuous system identification methods are studied and deduced to identify physical parameters directly, based on differential transform and integral transform. Continuous system identification strategy based on Hartley transform is researched and applied in control parameter identification of control system of HGU, then Haar transform with simple structure and high computing efficiency is studied and applied in parameter identification of objects in CSHGU.(3) In fact control system of HGU is a complicated nonlinear system. System identification of control system of HGU is studied under the condition of keeping all nonlinear sections of model of CSHGU. Gravity search algorithm (GSA) is introduced and improved by combining merits of particle swarm optimization, the search ability of improved GSA is enhanced by combination of gravity search, information sharing and ability of memory. The improved GSA is applied in nonlinear parameter identification of CSHGU and the identification strategy based on intelligent optimization method is proposed, realizing the precise identification of governing system of HGU under complicated operating conditions.(4) In addition, system identification of CSHGU based on nonlinear model is researched, focusing precise modeling using fuzzy model. On the basis of T-S fuzzy model, mutative scale chaos optimization is used to optimize the structure and parameters of T-S fuzzy model. In order to improve the linearity of fuzzy partition, fuzzy clustering methods based on linear regressive model and hyperplane are proposed, realizing precise identification of T-S fuzzy model, finally the effectiveness of proposed methods are verified in system identification of CSHGU.(5) In the end, fault diagnosis of dynamic system based on model and system identification is studied, building the foundation for further research. Fault diagnosis based pattern recognition based on fuzzy clustering analysis is emphasized, while weighted chaos optimization based fuzzy clustering method (WCOFCM) and weighted fuzzy kernel clustering (WFKC) algorithm are proposed. In WCOFCM, the global search ability of chaos optimization and local search ability of gradient operator are combined to improve the ability of obtaining more excellent partition solutions. In WFKC, samples in original space are mapped to high-dimension feature space by mercer kernel, and then a similarity based weighting method is used to assign weight to features of the transferred samples, and finally weighted fuzzy clustering in feature space is realized. WCOFCM and WFKC are applied in fault pattern recognition of HGU, the results show that the accuracy of fault pattern recognition are significantly improved.更多还原