Search

• Methodology developed for the construction of information-rich training data. • A full data-driven machine learning model for small-signal stability assessment. • Bayesian optimisation-based generation rescheduling for preventive control. • Higher performance (52% improvement) compared to engineering-judgement scheduling. Due to the development of communication technologies, a large amount of data, particularly in power systems, can be measured through digital equipment. Efficient analysis of the collected data can bring many benefits to power system operators, particularly in stability assessment. This paper presents an Extreme Gradient Boosting (XGBoost) decision tree (DT) for predicting damping ratios of an inter-area oscillation. This machine learning technique uses system data to predict the damping of critical system oscillations which can be subsequently improved by optimisation-based generation rescheduling. As power systems experience increased modelling nonlinearity introduced by converter-interfaced sources, traditional techniques for eigenvalue computation become difficult to implement in real-time due to the need for an estimated linearised model. An XGBoost DT uses an ensemble technique which provides higher prediction accuracy compared to other DT techniques. The effectiveness of the XGBoost DT is illustrated by using it in conjunction with a Bayesian optimisation to perform active power rescheduling to improve the damping ratio of an inter-area mode. The results show an average relative improvement of 58.10% in the damping ratio of the inter-area oscillation across multiple test cases when the proposed optimisation-based generation rescheduling is implemented in conjunction with the XGBoost-based oscillation damping estimation model. [ABSTRACT FROM AUTHOR]