Adaptive Deep Reinforcement Learning-Augmented Hybrid SMC-ANFIS Controller for Dynamic Voltage Restorer: Enhanced Power Quality Mitigation in Active Distribution Networks with Renewable Energy Integration
Year: 2026
Type: preprint
Abstract: The increasing penetration of renewable energy resources and power electronic converters has intensified power quality disturbances in active distribution networks, necessitating intelligent Dynamic Voltage Restorers (DVRs) capable of rapid and adaptive voltage compensation. This paper proposes an Adaptive Deep Reinforcement Learning-assisted Hybrid Sliding Mode Controller integrated with an Adaptive Neuro-Fuzzy Inference System (ADRL–HSMC–ANFIS) for DVR-based power quality enhancement. The proposed framework combines synchronous reference frame theory for voltage extraction, an ANFIS-based adaptive reference voltage generator, and a chattering-suppressed hybrid sliding mode controller whose control gains are continuously optimized using a Twin-Delayed Deep Deterministic Policy Gradient (TD3) agent. In addition, a novel Integral Square Error with Rate-weighting (ISER) objective function is formulated to simultaneously minimize voltage regulation error and transient oscillations, thereby improving both transient response and steady-state performance. The proposed controller is validated on a modified IEEE 33-bus active distribution network integrated with a 1 MW photovoltaic generation system and a 500 kWh battery energy storage system under balanced and unbalanced voltage sag, voltage swell, dynamic load variation, and renewable intermittency. Simulation results demonstrate a voltage restoration time of 3.2 ms, load voltage regulation of 99.95 ± 0.05%, and a minimum voltage total harmonic distortion (THD) of 0.41%, while providing superior DC-link voltage regulation, reduced chattering, and enhanced robustness compared with conventional PI, PSO-PI, SMC, PSO-SMC, and DQN-SMC controllers. These results demonstrate that the proposed ADRL–HSMC–ANFIS framework provides an effective, adaptive, and computationally efficient control strategy for resilient power quality enhancement in renewable-integrated active distribution networks.
Source: Research Square
Institution: Chettinad Academy of Research and Education
Language: English
Topic: Power Quality and Harmonics
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Keywords: dynamic voltage restorer, power quality mitigation, active distribution networks, renewable energy integration, deep reinforcement learning, sliding mode control, adaptive neuro-fuzzy inference system, TD3, chattering reduction, voltage sag and swell, photovoltaic generation, battery energy storage, DC bus voltage regulation, total harmonic distortion
Open Access status: green
OpenAlex ID: W7235434495
