Nikola Tesla Institute
Miroslav Dragićević
Forecast of Voltage-Reactive States Within the Natural Voltage Zone
Abstract:
The goal of this research is to develop appropriate machine learning models for predicting voltage values on the 400kV side of the grid. The data used for training the models include two years of historical voltage data, along with hydrometeorological variables, with temperature being the most significant factor, supported by meteorological yearbooks. Additional input data include air humidity, wind direction, wind speed, precipitation, and the appearance of ice on power lines.
The prediction of voltage values aims to forecast the engagement of static reactive reserves, with a sampling period of 10 minutes (min/max/avg). This data enables annual planning of energy imports and maintenance periods, as well as daily planning of capacity engagement. The next step in the research is the forecast of dynamic reactive reserves, where samples with a frequency of one second and more frequent samples would be used for model training.
Keywords: machine learning, forecasting Voltage-Reactive states, decision trees, gradient boosting, regression
Biography of the presenter:
Miroslav Dragićević, dipl ing E. E acquired a degree in electrical engineering from the School of Electrical Engineering, University of Belgrade, in the study group for power systems.
Miroslav enhanced his professional skills at the Nikola Tesla Electrical Engineering Institute in Center for automation and regulation, in the areas of managing industrial-technological processes using process computers, turbine regulation, and complete management of small hydroelectric power plants. He is the deputy head of the Specialized laboratory for turbine governing systems.
Miroslav is a member of Engineering Chamber of Serbia as well as the Committee N057 – Control and communication in the power systems at the Serbian Institute for standardization.
