Adaptive Methodology for Determining Frequency Control Law in Isolated Power Systems Based on Machine Learning Algorithms

Modern energy systems are undergoing significant changes due to the active implementation of renewable energy sources, control devices based on power electronics, the development of transient processes control systems and monitoring, as well as digitalization of production, transmission and distribution electricity processes. These changes contribute reduce the inertia of power systems and increase the speed of transient processes, as well as increase their stochasticity. Under these conditions, traditional power system emergency control devices can lead not only to non-optimal implementation of control actions, but also contribute to the development of cascading failures. The study presents a methodology for selecting a frequency control law in isolated power systems based on machine learning algorithms. This class of algorithms has significant speed and adaptability, which satisfies the conditions of modern power systems. For the case study, the standard IEEE39 digital power system model was used, for which a dataset was generated. Next, the task of feature selection were solved based on the Boruta algorithm and the k-nearest neighbor, random forest, extreme gradient boosting and artificial neural network algorithms were trained. As a result, the extreme gradient boosting algorithm showed the greatest accuracy and performance. The conclusions provide recommendations for the practical use of the developed methodology and directions for further research.