Reinforcement Learning-assisted Voltage Stability Analysis of the Nigerian Power Grid Using Dynamic Voltage Restorer (DVR) and Battery Energy Storage System (BESS)
Ndifreke U. Ekanem *
Electrical and Electronics Engineering Department, University of Uyo, Nigeria.
Mfonobong A. Umoren
Electrical and Electronics Engineering Department, University of Uyo, Nigeria.
Kufre M. Udofia
Electrical and Electronics Engineering Department, University of Uyo, Nigeria.
*Author to whom correspondence should be addressed.
Abstract
This study proposed and validated a reinforcement learning (RL)-assisted framework to enhance voltage stability in Nigeria’s power grid through dynamic coordination of a Dynamic Voltage Restorer (DVR) and Battery Energy Storage System (BESS). Addressing the Nigerian grid’s vulnerabilities—including ageing infrastructure, frequent voltage sags/swells, and inadequate reactive power support—the framework integrated a Deep Deterministic Policy Gradient (DDPG)-based RL controller with classical voltage stability methodologies rooted in Newton-Raphson power flow, Jacobian matrix eigenvalue analysis, and voltage stability indices (L-Index, VCPI). A realistic 3-bus MATLAB/Simulink model of the Alaoji-Onitsha 330kV transmission corridor was developed, simulating fault-induced instability scenarios with a 500 MVA generator, 800 MW + 300 MVAR industrial load, and transmission line parameters reflective of Nigeria’s grid. The RL agent was trained to minimise voltage deviations and harmonic distortion, and dynamically optimised DVR voltage injection and BESS charge/discharge cycles achieved a 3.25 ms response time of 0.92 p.u. voltage compensation, and a 96% reduction in total harmonic distortion (THD from 9.06% to 0.36%). Comparative analyses demonstrated the RL controller’s superiority over conventional PI, ANN, and PSO methods, with 75% faster transient recovery and 71% lower THD. Empirical validation under IEC 61000-4-30 and IEEE 519-2022 standards confirmed stable voltage regulation within ±0.9% of nominal during asymmetrical faults, while Jacobian eigenvalue analysis revealed a 40% improvement in stability margins (smallest singular value, σ_min, increased by a factor of 2.8). Through the combination of model-free RL adaptability with physics-based grid modelling, the study provided an adaptable solution for weak grids, reducing dependency on pre-trained datasets and offering a cost-effective strategy for mitigating voltage instability in Nigeria’s power system amid growing renewable integration and load volatility.
Keywords: Reinforcement learning (RL), dynamic voltage restorer (DVR), battery energy storage system (BESS), voltage stability, jacobian matrix analysis, nigerian power grid