中文摘要：

为提高汽轮发电机组励磁与汽门系统机端电压和功角的控制性能，提出了基于在线学习和自抗扰控制（ADRC）的神经网络逆鲁棒控制方法．首先，将神经网络逆（ANN）与被控励磁汽门系统组成的复合伪线性系统等效为含有扰动的线性系统；然后，基于ADRC，设计了用于在线估计复合伪线性系统状态和扰动的ESO，解决了神经网络逆在线学习时训练样本获取的难题，并在设计的伪控制量中对扰动进行补偿，基于线性系统理论证明了ESO的收敛性并针对励磁子系统和汽门子系统与神经网络逆系统组成的伪线性复合系统分别设计整数阶PID控制器和分数阶PID控制器以实现闭环控制；同时，在离线训练的基础上设计了基于在线梯度方法的神经网络逆在线学习算法，利用李雅普诺夫稳定性理论证明了神经网络逆在线学习的收敛性．最后，以典型的两区域四机系统为例进行数值仿真，与传统的AVR／PSS和基于离线训练的神经网络逆控制方法的比较结果表明所提方法明显提升了电力系统的暂态性能．

英文摘要：

To improve the performance online learning control scheme and active disturbance of the terminal voltage and power angle rejection control （ADRC） based ANN was proposed. Firstly, the composite pseudo linear system, of the turbogenerator, an -inversion （ANNI） robust which is composed of the ANNI system and the controlled excitation and valve system, is equivalent to a linear system with disturbance. Then, an ESO was designed based on the ADRC method to estimate the states and the disturbance of the composite pseudo linear system online, in order to resolve the difficulty of online acquisition of training samples for the online learning of ANN inversion, and the pseudo control input with disturbance compensation was designed for the composite pseudo-linear system. Furthermore, the convergence of the ESO was proved by the linear system theory and an integral order PID controller, and a fractional order PID controller was designed for the the composite pseudo linear excitation and valve system. Meanwhile, an online learning algorithm of the ANNI was proposed with the online gradient descent method based on offline training, and the convergence of the online learning algorithm of the ANNI was proved according to the Lyapunov stability principles. Finally, a case study was conducted on a typical two-area four- machine power system and, compared with the conventional AVR/PSS and the offline trained based ANNI control scheme, the results show that the proposed control scheme can greatly improve transient performance.