| Title |
A Study on Prediction and Accuracy Improvement of SVR Voltage Control in Distribution System Based on Machine Learning |
| Authors |
유병찬(Byungchan Yoo) ; 최원나(Wonna Choi) ; 정승민(Seungmin Jung) |
| DOI |
https://doi.org/10.5370/KIEE.2025.74.1.7 |
| Keywords |
Distribution system; Reverse power flow; Step voltage regulator; Machine learning |
| Abstract |
The proportion of distributed energy sources in the distribution network is rapidly increasing. Along with this, voltage management is becoming more difficult, and stability issues in the power system are being reported. Until recently, the voltage of distribution lines has generally been regulated using voltage regulators, predominantly deployed in sections where rapid voltage drops occur.
However, as the characteristics of distributed energy sources are reflected, the efficiency of the voltage regulator diminishes.
Therefore, cooperation with substations is necessary for stable operation of the distribution network. This paper aims to predict the operation of a machine learning-based voltage regulator to evaluate voltage fluctuations occurring in distribution lines and the corresponding control operations in advance. Long Short-Term Memory is capable of prediction on time scales. Therefore, the output characteristics of distributed energy sources can be considered. Additionally, to improve prediction accuracy, cases are generated in the OpenDSS-Python environment and additional training is performed on the LSTM model. As a result, it was confirmed that the prediction accuracy improved when the proposed method was applied. The machine learning model was verified using four evaluation indices. |