University of Tehran Journal of Solar Energy Research 2588-3097 11 1 2026 01 01 Convolutional Neural Network–Long Short-Term Memory Based Intelligent Adaptive DSTATCOM Control for Enhanced Power Quality in Photovoltaic-Integrated Distribution Networks 2780 2801 105741 10.22059/jser.2026.407530.1677 EN Mir Manjur Elahi Department of Electrical Engineering, Biju Pattnaik University of Technology, Rourkela, Odisha, India Prakash Kumar Ray School of Electrical Sciences, Odisha University of Technology and Research, Bhubaneswar, India Pratap Sekhar Puhan Department of Electrical and Electronics Engineering, Sreenidhi institute of science and Technology Hyderabad, India Journal Article 2025 12 03 This paper presents a Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) based intelligent adaptive control strategy for a DSTATCOM to enhance power quality in photovoltaic (PV)-integrated distribution networks. The proposed controller exploits CNN-based spatial feature extraction of voltage–current waveforms and LSTM-based temporal learning to address nonlinear and time-varying disturbances. Simulation and real-time hardware-in-the-loop (HIL) validation using an OPAL-RT platform confirm superior performance under dynamic irradiance (200–1000 W/m²) and load variations (up to 50 kW). The DC-link voltage is tightly regulated within 758–764 V, with a maximum deviation below 3 V. Total Harmonic Distortion (THD) is reduced from 16.5–22.5% to 3.2–4.5%, achieving 79–82% harmonic suppression and compliance with IEEE-519 limits. The power factor improves from 0.78–0.84 to 0.97–0.99, approaching unity. The proposed controller exhibits a fast dynamic response of 3.8 ms, outperforming PI, Fuzzy-PID, ANN, CNN, and LSTM controllers. Reactive power tracking errors remain below 2.7%, demonstrating high robustness and real-time adaptability for smart PV-integrated grids. https://jser.ut.ac.ir/article_105741_ba3bbf5d2f9fd6818a7dd332d29e3cde.pdf