Machine-learning based approach of proportional reactive power dispatch under imposing voltage constraint

Abstract

This paper proposes a modification process for a proportional dispatch method for a strict radial wind farm structure based on a machine learning application. The general approaches of proportional dispatch require greater consideration of the constraints, because of grid expansion. This paper presents a new theoretical approach for a modified proportional dispatch method that is more advanced than the general algorithm and takes the voltage restriction for each wind farm unit into account. A linearization method for calculating the ohmic loss caused by reactive power flow under the voltage constraint is proposed as a major modification. To ensure that the proposed method can be applied to the controller in a stable manner, we design a machine learning-based look-up table operation. Along with the general proportional and even-dispatch methods, the proposed algorithm is simulated in order to verify it under reality-based wind farm conditions, by focusing on both electrical loss and voltage violation. The proposed modified method shows delivery improvements in compliance with the required voltage condition when simulated with PSCAD/EMTDC.