Inter-area Oscillation Damping Control using Machine Learning

Abstract

Inter-area oscillations are one of the major threats to the stability of an interconnected power system due to the fact that these oscillations involve generators from different areas of a power system and hence may result in higher oscillation in the tie-line by adding up effects from each of the participating generators. The success of oscillation

damping within a given duration, such as less than 15 seconds, is substantially corre- lated with the safe operation of a contemporary power system. Since power systems are

highly non-linear in nature, with time varying parameters, and a specific control design

based on the linearized model may not ensure satisfactory performance under varied op- erating scenarios. Therefore, a non-linear self-tuning controller is required that damps

inter-area oscillations in interconnected power systems under varied post-disturbance

operating conditions without requiring manual adjustment or re-tuning of controller pa- rameters. An effective controller for power oscillation damping has been designed in this

research. Moreover, a powerful online batch training method for neural networks known as the Levenberg-Marquardt (LM) algorithm is modified for computationally efficient

online estimation of power system’s dynamic behaviour and is referred as Computa- tionally Efficient Levenberg-Marquardt (CELM) algorithm. A unique form of neural

network, referred as Computationally Efficient Neural Network (CENN), is proposed that is compatible with the Feedback Linearizable Controller (FBLC), is used to ensure non-linear self-tuning control. It has been demonstrated that using the modified version

of LM algorithm i.e. CELM algorithm for successive disturbance leads to better accu- racy, faster convergence and offer less computational time than using the classical LM

algorithm.