Hybrid Deep Reinforcement Learning with Leaky LMS-ANN for Active Power Filter-Based UPQC in PV-Integrated System
Document Type : Research Article
Authors
- Sritam Parida 1
- Bhabani Patnaik 2
- G Mohan 3
- P Maniraj 4
- S Kaliappan 5
- Maheswar Prasad Behera 6
- Manoj Kumar Sahu 7
- Omkar Tripathy 8
- Supriya Sahu 9
- BIBHU PRASAD GANTHIA 6
1 Electrical Engineering, Indira Gandhi Institute of Technology, Sarang, Dhenkanal, odisha, India
2 Electrical and Electronics engineering, GIFT Autonomous Bhubaneswar
3 Associate Professor Department of Mathematics, K.S.Rangasamy College of Technology, Tiruchengode
4 Department of Electrical and Electronics Engineering, Thalavapalayam, Karur, Tamilnadu - 639 113
5 Electrical and Electronics Engineering, Kumaraguru College of Technology, Saravanampatti, Coimbatore, India
6 Assistant Professor, Electrical Engineering, Indira Gandhi Institute of Technology, Sarang, Dhenkanal, odisha, India
7 Department of Electrical Engineering, Centre for Advanced Post Graduate Studies, Biju Patnaik University of Technology, Rourkela, Odisha, India
8 Electrical Engineering, Indira Gandhi Institute of Technology, Sarang, BPUT, Rourkela, Odisha, India.
9 Mechanical Engineering, Indira Gandhi Institute of Technology, Sarang, Dhenkanal, Odisha, India
Abstract
The growing integration of photovoltaic (PV) systems into conventional distribution networks has intensified major power quality challenges such as 18–25% harmonic distortion, power factor deviation between 0.92 and 0.95, frequent voltage fluctuations, and load imbalance exceeding 8%. This study proposes a Hybrid Deep Reinforcement Learning (DRL) with Leaky LMS-ANN controlled Unified Power Quality Conditioner (UPQC) to achieve enhanced active power filtering in PV-integrated systems. The hybrid controller leverages the decision-making capability of DRL alongside the adaptive parameter tuning of the Leaky LMS-ANN algorithm, enabling real-time optimization under variable irradiance, nonlinear loads, and load-switching conditions. MATLAB/Simulink validation demonstrates substantial performance gains: total harmonic distortion is reduced from 18.4% to 2.97% (84% reduction), reactive power compensation improves by 55.6%, and voltage imbalance declines from 8.5% to 1.2%. The DC-link voltage stability increases by 23%, while the power factor is maintained near unity at 0.998. Compared with conventional ANN, LMS, and MPC controllers, the proposed approach delivers superior harmonic mitigation, voltage regulation, and system reliability, making it well suited for advanced smart grid applications.
Keywords
)