| Title |
A Study on Weight Optimization of DNN-Based Voltage Control for PV-Integrated Distribution Networks |
| Authors |
임진우(Jin-Woo Lim) ; (Peng Y. Lak) ; 이현탁(Hyun-Tak Lee) ; 최종길(Jong-Gil Choi) ; 남순열(Soon?Ryul Nam) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.4.741 |
| Keywords |
Renewable Energy; Distribution Networks With Distributed Generation; Voltage Control; Optimal Power Flow (OPF); Chance- Constrained Optimization; Differential Evolution; Deep Neural Network (DNN) |
| Abstract |
This paper proposes a Differential Evolution (DE)?based weight optimization framework built upon a DNN-based optimal power flow (OPF) structure capable of maintaining stable voltage regulation under limited observability. The DNN-based voltage control has previously been introduced as an efficient OPF surrogate achieving voltage regulation performance comparable to scenario-based Convex-CCOPF with significantly reduced computational burden, and this study extends the framework by treating the objective function weights as design variables. By optimizing the weights based on system responses over an entire scenario horizon, the proposed approach enables systematic performance adjustment while preserving voltage stability. Simulation results show that all bus voltages remain within the allowable range, while the voltage deviation from the nominal value slightly increases in terms of RMSE but stays within acceptable limits. Under these conditions, the daily total energy loss is reduced from 5.88421 MWh to 4.54900 MWh, corresponding to a 22.7% reduction, and the cumulative OLTC tap operations decrease from 9 to 8. Using a Total Performance Index that combines stability and efficiency metrics, the proposed DE-based weight optimization achieves an overall performance improvement of 27.81% compared to the baseline weight setting. These results demonstrate that incorporating weight optimization into a DNN-based OPF framework provides a practical and extensible voltage control solution that enables performance tuning according to operational objectives while maintaining system stability. |