The Application of Machine Learning for the Voltage and Reactive Power Control in Power Distribution Network

Recently, the integration of renewable energy sources into the distribution system is a significant effect on the operation of the conventional voltage regulation system, this challenge could be required more control actions. This study presents the applied machine learning (ML) for controlled voltage and reactive power in the power distribution system with a presence distributed generator (DG). The study aims to validate the effectiveness of ML algorithms to create the predictive model for pre-defining the coordinated operation of the reactive power compensator for voltage regulation. An hourly dataset in three months was collected from Electricite du Laos (EDL) Savannakhet branch, which applied to perform in the simple radial feeder in the distribution network by using Particle swarm optimization (PSO) for allocation the optimal operated scheduling of switched capacitor banks and the control scheme is assumed based on centralized management to communicate to the reactive power support devices. Also, a new dataset was created from the reactive power control process that provides the knowledge for using to train the ML algorithms. The utilized different algorithms consist of Decision Trees, Support Vector Machine (SVM) and k-Nearest Neighbors and the feature selection is considered. The simulation results demonstrate that the classification algorithm of ML provided the satisfying accuracy and can be proof the voltage regulation, which can deploy the model to assist the monitoring and pre-decision for the operator to control the system.