【作者】 林碧华；

【导师】 韩璞；

【作者基本信息】 华北电力大学（河北） ， 热能工程， 2005， 博士

【摘要】 神经模糊系统是一种将模糊逻辑系统和神经元网络有机结合的新型的模糊推理系统结构,是模糊推理系统的神经网络实现,它具有以任何精度逼近任何线性或非线性函数的功能,又具有收敛速度快、误差小、所需训练样本少的特点。本文主要研究了神经模糊系统的学习能力、泛化能力,神经模糊系统输入变量的选取方法,过拟合问题、维数灾难问题的解决方法,最后将神经模糊系统应用于电厂的复杂系统——协调系统的建模和控制中,验证了神经模糊系统的有效性。1、分别研究了输入变量组合、隶属函数个数、训练次数、训练样本个数等对神经模糊系统的学习能力、泛化能力的影响以及神经模糊系统的学习能力和泛化能力之间的关系。2、提出了一种综合考虑了训练误差和检验误差的评价神经模糊模型性能的误差性能指标。3、提出了一种神经模糊系统输入变量的选取方法。该方法简单、快速,并且可以直接得出被建模系统的n、m、td 等,且输入变量被选中的先后顺序表明了该输入变量对被建模系统输出影响的重要程度。4、对于神经模糊系统的过拟合问题,本文提出了一种训练样本、检验样本的选择方法以及一种最佳训练次数的确定方法。5、对于复杂系统的神经模糊建模,(1)本文研究了多级神经模糊系统的结构及学习方法;(2)提出了一种基于多级神经模糊系统的输入变量选取方法;(3)提出了一种基于多级神经模糊系统的多输出系统建模方法。6、针对电厂协调系统中锅炉侧存在着很大纯迟延的问题,本文设计了两级神经模糊系统对电厂协调系统进行建模,分别建立其非线性模型、线性模型以及根据现场采集的数据建立的实际模型。仿真结果表明,应用神经模糊系统建立的模型具有较高的辨识精度和较小的预测误差。7、对于电厂协调系统,本文以神经模糊系统作为控制器,设计了两种控制方案:(1)基于LQR 控制的神经模糊协调控制系统;(2)基于PID-解耦控制的神经模糊-解耦协调控制系统。仿真结果表明,基于神经模糊系统的协调控制系统具有较好的控制性能和鲁棒性。更多还原

【Abstract】 Neural Fuzzy System (NFS) is a new fuzzy inference system structure that combined fuzzy logic system with neural networks. It is the neural network realization for Takagi and Sugeno fuzzy inference system. It can approach any linear and nonlinear function with any precision, but also quicken converging speed, decrease precise errors, and lesser training samples that are needed. This paper researched the learning ability, generalization ability, method of input selection, overfitting problem and dimension disaster of NFS. In the end, NFS was applied to the modeling and control of Coordinated Control System (CCS) that is complex system of power plant to validate the validity of NFS. 1、The effect that the combination of input, the number of membership function of input, training times and training sample affect the learning ability and generalization ability of NFS was researched respectively. The relation between the learning ability and generalization ability was studied. 2、A performance index of error was presented. This index is a kind of evaluation of neural fuzzy model performance and synthetically considered training error and checking error of NFS. 3、A new method of input selection for NFS was proposed. This method is easy, fast, and can directly obtain n, m and td of the object system. Another advantage is that precedence order of input selection indicates the importance grade of input influencing output. 4、As for the overfitting problem of NFS, this paper proposed a method of selecting training sample and checking sample and a method of selecting the most befitting training times. 5、As for the complex system modeling based on Neural Fuzzy System, This paper (1) researched the structure and learning algorithm for the Multistage Neural Fuzzy System (MNFS); (2) presented a method of input selection based on MNFS; (3) presented a method of building the model of the multi-output systems based on MNFS. 6、Because of the large time-delay of CCS of Boiler-turbine unit, this paper designed two-stage neural fuzzy system to build the models of CCS. NFS was respectively used to build the nonlinear model, the linear model and the NFS model built according to the field data of the CCS. The simulation results show that the models based on NFS have higher identification precision and less predictive error. 7、For the CCS of Boiler-turbine unit, using NFS as a controller, this paper designed two control methods: (1) the neural fuzzy CCS based on Linear Quadratic Regulator(LQR) control system; (2) the neural fuzzy CCS based on PID-decoupling control system. The simulation results show that the control systems based on NFS have better control performance and stronger robustness.更多还原