Open Access

ARTICLE

Data-Driven Voltage Control for Distribution Networks with High Penetration of PVs via Improved DDPG

Xiaolong Xiao^1,*, Linghao Zhang², Ziran Guo¹, Xiaoxing Lu¹, Wenqiang Xie¹, Shukang Lv¹, Peishuai Li³
1 Electric Power Research Institute, State Grid Jiangsu Electric Power Co., Ltd., Nanjing, China
2 State Grid Jiangsu Electric Power Co., Ltd., Nanjing, China
3 School of Automation, Nanjing University of Science & Technology, Nanjing, China
* Corresponding Author: Xiaolong Xiao. Email:
(This article belongs to the Special Issue: AI and Advanced Computational Techniques for Sustainable Renewable Energy Systems)

Energy Engineering https://doi.org/10.32604/ee.2026.076751

Received 26 November 2025; Accepted 12 February 2026; Published online 17 April 2026

Abstract

The highly increasing integration of distributed photovoltaics (PVs) has brought significant challenges for distribution network voltage control, which struggles with fast and stochastic fluctuations. This paper proposes a data-driven voltage control method utilizing an improved deep deterministic policy gradient (DDPG) to minimize power losses while maintaining voltage within a safe operating range. The data-driven voltage control framework is established first, with the controlling center formulated as the agent. Based on the distribution network system and the voltage control original model, the data-driven voltage control model is formulated via the Markov decision process (MDP). A voltage sensitivity-guided DDPG algorithm, which incorporates the voltage-reactive power injection sensitivity matrix into the actor policy gradient, is developed to embed physical grid knowledge and reduce the actor’s reliance on critic action-gradient estimates, thus the practicality and generalization of data-driven voltage control is improved. Numerical simulations on the modified IEEE 33-node test system show the method maintains all nodal voltages within limits and reduces total system losses under varying PV generation, demonstrating its effectiveness and potential for future sustainable power systems.

---

Keywords

Distributed photovoltaics (PVs); distribution network; voltage control; deep reinforcement learning (DRL); improved DDPG

---