【作者】 吴雪松；

【导师】 李振然；

【作者基本信息】 广西大学 ， 农业电气化与自动化， 2004， 硕士

【摘要】 本文对线性最优、非线性最优和直接反馈线性化等励磁控制方法进行了深入分析。发现它们有一个共同点就是如何把状态方程A、B矩阵线性化，再利用线性最优控制理论求出最优控制规律。 本文提出一种线性自适应励磁控制器，它不谋求将A、B矩阵线性化，而是使A、B矩阵非线性元素随运行方式而变化。对某一运行方式而言，A、B矩阵是已知常数矩阵，可利用线性最优控制理论求出最优控制规律，但根据运行工况的变化在线自适应求解最优控制规律是困难的。本文将神经网络和线性最优控制方法结合起来试图解决这个难题。利用各种运行方式下有功功率(P_e)、无功功率(Q_e)和发电机端电压(U_G)的组合作为神经网络输入；针对特定的网络结构和运行方式求解线性最优反馈增益作为神经网络的教师值。利用大量样本离线训练神经网络，使用中只要输入当前运行方式P_e、Q_e和U_G值，神经网络便输出相应的最优反馈增益。同时考虑到线性最优控制在一定范围内的阻尼特性和连续调节增益带来的干扰，设计中检测状态量的变化大小决定是否启动神经网络控制器。仿真结果表明，这种新型励磁控制器与LOEC 广西大学硕士学位论文 相比，在稳态电压调节精度和系统动态特性等方面有明显的提高。 此外，本文对即神经网络结构、各种算法及收敛性作了深入 研究，对如何利用MATLA日的5 1 mpowersystem进行仿真进行了深 入探讨。更多还原

【Abstract】 This paper gives a thorough research on different excitation control methods such as linear or nonlinear optimal excitation control and direct feedback linear excitation control. A common ground was found: to linearize the matrix of A?B in the state equation in different ways and seek for the optional control laws through linear optimal control theory.This paper designed out a linear adaptive excitation controller. It doesn’t linearize the matrix of A and B, but make the nonlinear elements change with the operating condition. As for a given condition, the matrix is constant. Although the control laws can be worked out from linear optimal control theory, it is difficult in the real time. This paper combines ANN with linear optimal control theory to solve this problem. Active power Pe, reactive power Qe and the end voltage of the generatorU are used as the inputs of ANN; the optimal feedback gains from the Riccati equation are used as teachers. After trained offline, the ANN can output the optimal gains according to the operating condition, the simulation result shows, this controller has better voltage regulation precision and dynamic characteristic than LOEC.Besides, this paper makes some researches on the structure? training algorithm and convergence of BP neural network and also gives a embedded discuss about how to simulate the power system with the software of MATLAB/SimPowerSystem.更多还原