【作者】 王淑青；

【导师】 李朝晖；

【作者基本信息】 华中科技大学 ， 系统分析与集成， 2006， 博士

【摘要】 水电机组系统是具有时变、非线性、非最小相位等特性的复杂系统,其控制的可靠性是水电厂安全运行的关键。为了提高系统的控制性能,设计出控制效果更好的水轮机调节器,在全面总结现有水电机组系统控制策略和辨识方法成果的基础上,结合水电机组控制系统特性,引入智能控制系统,对水电机组系统辨识和控制方法做了深入研究,提出了具体的改进措施和智能控制器的设计方法,并给出了仿真结果。在分析水电机组经典数学建模和常规人工神经网络建模方法基础上,提出了基于Takagi-Sugeno型的ANFIS网络进行水电机组辨识方法,并针对该方法在误差较小时训练网络参数收敛速度慢的问题,采用拟Newton算法和梯度下降混合学习算法进行参数的训练学习,使ANFIS辨识网络具有很好的实时性,对水电机组进行辨识提供了一种新途径。针对常规PID控制存在的问题,详细分析了现有的PID控制优化方法,并对这些优化方式进行了设计和仿真;结合仿真实验结果,分析了其优缺点,为选择较好的优化PID方法提供了依据,在此基础上提出了采用经过模糊整定的PID控制与模糊控制并联揉合构成的水电机组控制器,仿真结果验证了其有效性。为了提高系统控制性能,设计了水电机组模糊神经网络控制器,详细介绍了其控制器结构、模型和学习方法。为了克服基于误差反传的模糊神经网络控制器学习过程容易产生振荡和收敛速度慢的缺点,提出了采用自适应学习训练算法。为确保模糊神经网络控制器学习过程的稳定和收敛,采用Lyapunov理论对模糊神经网络学习参数进行了优化,仿真结果验证了其有效性。针对模糊神经网络设计和仿真过程中的结构参数选择具有一定的主观性和试探性的问题,提出将软计算方法应用于水电机组控制器设计中,设计了基于模糊推理系统、神经网络、遗传算法相融合的水电机组控制器,给出了控制器的结构和设计方法,针对遗传算法优化中存在的早熟等问题,提出了采用改进遗传算法进行优化的方法。为了防止水电机组大扰动时模糊神经网络学习速度慢、易陷入局部最优引起控制效果不佳或引起不稳定现象的发生,控制器中并联了监督控制器,给出了仿真结果。全面系统地总结了本文的工作和研究成果,并指出了有待改进的地方和需进一步开展的工作。更多还原

【Abstract】 The hydroelectric generating system of water power plant is a high-order, non-linear, with time-variable and non-minimal phase properties system. The reliable control of hydroelectric generating system is a key of water power plant safe running. In order to research better intelligent hydro turbine governor to improve control performance, the dissertation summarizes the existing control strategies and identifying methods. Combining the character of complicated hydroelectric generating, intelligent control strategies and identifying methods have been researched and some concrete measures of improvement and designing methods of intelligent governor are put forward. The corresponding application simulating results are given. After analyzing general math modeling and conventionality artificial neural network modeling of hydroelectric generating, the ANFIS network based on Takagi-Sugeno is put forward to identify the character of hydroelectric generating. To settle its slow convergence problem in turning network parameters when error is less,analogy Newton algorithm and gradient-descent mix algorithm are adopted to turning network parameters , which bright excellent real-time character and provided a brand new way to the identification of hydroelectric generating.According to the problem existing in the regular PID control, some optimizing methods of PID control are further analyzed and their relevant design and simulating results are given. Then the advantages and disadvantages are discussed based on simulating experiment results, which provides convincing proves for better choice PID optimizing methods. Based on all above work, mix control strategy based on PID control with fuzzy optimizing and fuzzy control is put forward. The simulating results prove its validity.To improve the controllability of bigger water power hydroelectric generating system,a new hydroelectric generating controller is designed based on fuzzy neural network. The structure, model and study approaches are introduced in detail in the dissertation as well. To handle the problem that slow convergence speed or shake brought by using back-propagation method only in learning,the self-adapt learning method is used to turning parameters. To ensure the stability and convergence of fuzzy neural network controller during the learning process,network parameters are chosen based on Lyapunov theory. The simulating results proved its validity.As for choosing parameters of the fuzzy neural network controller having some subjectivity and probing, soft-computing method based on fuzzy reference system, neural network and genetic algorithm is put forward to control hydroelectric generating. The structure and design of controllers are given. The improved optimizing method of genetic algorithm is put forward to solve the problem that existed in the optimizing of genetic algorithm. Control monitor is used to avoid brought the problem bad performance or shake when fuzzy neural network optimizing in bigger work condition change. The simulating results of designed intelligent controller are given.Finally, the dissertation summarizes all the works and results achieved in this dissertation. The further research works to be developed are also put forward.更多还原