Abstract In this study, an artificial neural network observer based sliding mode control strategy is proposed for load frequency regulation problem in multi-area power system. In this study both unmatched disturbance estimation and its rejection is done by using artificial neural network (ANN) observer. A three layer feed forward neural network is considered for ANN observer and its weights are trained using new modified adaptive training rule. The ANN observer guarantees precise estimation of the actual variables and this leads to convergence of estimation error to zero. In order to neglect chattering in control signal, the estimated unmatched unknown disturbance via ANN observer is utilized to select switching surface boundary limits. The ANN observer based controller improves closed loop system time response when compared with well known existing GESO based NSMC and two layer active disturbance rejection control (ADRC) schemes at random unmatched and unknown disturbances. It also rejects the effects of unmatched unknown disturbances and unknown bounded power integration in the system. The ANN observer based controller is also validated on IEEE 39 bus system. The robustness of the proposed ANN observer based controller in terms of stability and effectiveness when subjected to unmatched unknown disturbance and unknown power integration is established by the simulation results.

中文翻译：

---

基于神经网络观测器的电力系统控制器调频

摘要 在本研究中，针对多区域电力系统中的负载频率调节问题，提出了一种基于人工神经网络观测器的滑模控制策略。在这项研究中，通过使用人工神经网络 (ANN) 观察器来完成不匹配干扰估计及其拒绝。ANN 观察器考虑了三层前馈神经网络，其权重使用新的修改自适应训练规则进行训练。ANN 观测器保证对实际变量的精确估计，这导致估计误差收敛到零。为了忽略控制信号中的颤动，通过人工神经网络观察器估计的不匹配未知干扰被用来选择切换表面边界限制。与众所周知的现有基于 GESO 的 NSMC 和两层自抗扰控制 (ADRC) 方案相比，在随机不匹配和未知干扰下，基于 ANN 观测器的控制器改进了闭环系统时间响应。它还拒绝了系统中不匹配的未知干扰和未知有界功率集成的影响。基于 ANN 观察器的控制器也在 IEEE 39 总线系统上得到验证。模拟结果确定了所提出的基于 ANN 观察器的控制器在遭受不匹配的未知干扰和未知功率积分时在稳定性和有效性方面的鲁棒性。它还拒绝了系统中不匹配的未知干扰和未知有界功率集成的影响。基于 ANN 观察器的控制器也在 IEEE 39 总线系统上得到验证。模拟结果确定了所提出的基于 ANN 观察器的控制器在遭受不匹配的未知干扰和未知功率积分时在稳定性和有效性方面的鲁棒性。它还拒绝了系统中不匹配的未知干扰和未知有界功率集成的影响。基于 ANN 观察器的控制器也在 IEEE 39 总线系统上得到验证。当受到不匹配的未知干扰和未知功率积分时，所提出的基于人工神经网络观测器的控制器在稳定性和有效性方面的鲁棒性由仿真结果确定。