All Projects (0) Under Review (0) Published (0) Featured (0)

Projects

Graphical Abstract

Vision-Informed Multi-Modal Multi-Agent Reinforcement Learning for Active Distribution Network Scheduling

  • Proposed a multimodal multi-agent reinforcement learning-based scheduling framework for active distribution networks. The proposed framework integrates electrical grid measurements with soiling loss features extracted from solar panel images using a hybrid CNN–Vision Transformer (ViT) model, enabling agents controlling energy storage systems, flexible loads, solar plants, and static var compensation devices to learn cooperative and optimal operational policies.
  • Demonstrated significant improvements in grid performance and sustainability. The framework achieved superior scheduling outcomes—reducing operational costs, minimizing network losses, enhancing grid flexibility, and lowering carbon emissions—outperforming conventional scheduling strategies. Moreover, the proposed framework is validated on the large-scale 2289-bus Nizwa distribution network in Oman.
DOI
Graphical Abstract

Deep Learning-based Geospatial Mapping Framework for Large-Scale Electric Power Grids

  • Introduced a novel residual graph convolutional network with attention mechanism that enables precise geospatial prediction of energy infrastructure components (poles, service points, substations) and their spatial interconnections within large-scale distribution and transmission networks.
  • Validated the framework on real-world regional grids (Oman and Nigeria), achieving high link-prediction accuracies (95.88% and 92.98%) and near-perfect regression performance (R² ≈ 0.99), demonstrating strong generalizability and robustness across diverse geospatial contexts.
DOI
Graphical Abstract

Robust Deep Learning Model for Spatiotemporal Forecasting of Renewable Energy Sources

  • Developed a robust CNN–BiLSTM model with a spatiotemporal attention mechanism that jointly captures spatial correlations and temporal dependencies in large-scale renewable energy data, enabling accurate multi-horizon forecasting without requiring site-specific model training or additional preprocessing steps.
  • Enhanced resilience and performance through advanced optimization and robust learning by employing a correntropy-based loss function to mitigate the effects of data contamination (incompleteness, Gaussian noise, and outliers) and introducing a Partial Reinforcement Optimization strategy for efficient hyperparameter tuning—demonstrating superior accuracy over state-of-the-art forecasting methods on real-world photovoltaic (Arizona) and wind (Texas) datasets.
DOI
Graphical Abstract

Microgrid Resource Planning Framework Using Multi-Agent Deep Reinforcement Learning Approach

  • Developed a comprehensive framework utilizing Multi-agent Deep Deterministic Policy Gradient (MADDPG) to integrate renewable energy sources (RESs), battery energy storage systems (BESSs), and electric vehicle charging stations (EVCSs). This framework incorporates dynamic simulation environments and responds to stochastic variables like trip durations and sporadic nature of RESs.
  • Aimed to optimize network power losses, total installation and operational costs, greenhouse gas emissions, and system voltage stability. The model also focuses on improving the state of charge (SOC) for EVs, using practical weather data and EV behavior across multiple grid networks. The approach outperformed contemporary techniques across various performance metrics.
DOI
Graphical Abstract

Techno-economic and Environmental Assessment of Electric Vehicle Charging Stations in low-inertia Microgrids

  • Developed a novel Multi-objective Optimization framework to facilitate optimal allocation of photovoltaic (PV) systems, wind turbines (WT), battery energy storage system-based virtual synchronous generators (BESS-VSG), and electric vehicle charging stations (EVCS) in microgrids. This model uniquely addresses frequency instability through BESS-VSG, which simulates the inertia of traditional synchronous generators.
  • This research aimed to optimize total net present cost (TNPC), levelized cost of electricity (LCOE), energy loss, frequency deviation, voltage stability, and carbon emissions. The approach was tested on real-world grid networks in Masirah Island, Oman, and Ankara, Turkey, showing significant reductions in carbon emissions and improvements in economic and technical metrics compared to existing methods.
DOI

Neural Architecture Search via Reinforcement Learning for Image-based Solar Panel Soiling Loss Prediction

  • Proposed a novel RL–based NAS framework that explores a discrete neural architecture search space for image-based regression of solar panel soiling loss.
  • The project aims to effectively address the computational limitations and adaptability challenges of conventional NAS methods.
  • The work is currently under review.
See Under Review Publications See Featured Publications
Graphical Abstract

Stochastic Multi-objective Architecture for Strategic Integration of Distributed Energy Resources in Microgrids

  • Developed a novel Multi-objective Architecture to optimize the allocation and sizing of renewable energy-based distributed generation (RDG) and the operational strategies of battery energy storage systems (BESS) in microgrids. This model evaluates uncertainty in load demands and renewable outputs to ensure robust system performance under varying conditions.
  • The model aims to enhance voltage stability margins and reduce annual operational costs while ensuring substantial energy transfers during peak and off-peak periods. The model's effectiveness was demonstrated on practical distribution grids, surpassing other contemporary metaheuristics in terms of cost, energy loss, and voltage stability.
DOI
Graphical Abstract

Optimal Planning of Renewable Energy-Integrated Distribution System with Uncertainties

  • Developed a novel planning framework for the optimal allocation and sizing of renewable energy-based distributed generation units (RDGs), including photovoltaic (PV) and wind turbines (WT). This model incorporates uncertainty by employing probability distribution functions to simulate the stochastic nature of renewable outputs, enhancing the reliability of the power network under fluctuating conditions.
  • The project focused on minimizing power losses, improving voltage stability margins (VSM), reducing voltage deviations, and achieving yearly economic savings. The effectiveness of the proposed method was validated on various distribution systems, demonstrating superior performance in enhancing the techno-economic benefits of the distribution systems.
DOI
Graphical Abstract

Electric Vehicle Integration in Bangladeshi Residential Distribution Grid

  • Developed a Monte Carlo simulation-based stochastic load flow algorithm to analyze the effects of EV integration in three different urban residential neighborhoods of Bangladesh. Additionally, implemented an optimization approach to refine EV charging strategies within the distribution networks
  • This study aimed to assess the impact of two EV penetration scenarios (20% and 30%) across different seasons, identifying areas susceptible to voltage collapse and optimizing charging strategies to enhance grid capacity. The research provided insights into how increased EV loads could be accommodated effectively, improving overall grid performance, voltage stability, and reducing power losses, thereby guiding future infrastructure upgrades and policy development for sustainable energy integration.
DOI
Graphical Abstract

Multi-Objective Optimal Planning of Virtual Synchronous Generators in Microgrids with Integrated Renewable Energy Sources

  • Developed a novel Multi-objective Framework for the strategic placement and sizing of renewable distributed generation (RDG) and virtual synchronous generators (VSG). This model accounts for the inherent uncertainties in renewable outputs and load variations, ensuring robust frequency and voltage stability in microgrids under varying operational conditions.
  • The project aimed to minimize frequency deviations and optimize operational costs and energy savings of RDG and VSG units. The effectiveness of this approach was validated across multiple practical distribution networks, showing superior performance in maintaining system stability and reducing energy losses compared to existing systems.
DOI
Graphical Abstract

Optimal Load-Shedding in Distribution System

  • Developed an optimal load shedding technique utilizing the Chaotic Slime Mould Algorithm (CSMA) with a sinusoidal map to enhance the efficiency of islanding operations in distribution systems with Distributed Generation (DG). This approach employs a constrained function focused on maintaining static voltage stability margin (VSM) and optimizing total remaining load after shedding.
  • This study aimed to address the inefficiencies in contemporary load shedding schemes by ensuring optimal load reduction, preserving frequency and voltage stability under various islanding scenarios. The efficacy of the CSMA was validated against traditional methods like the Backtrack Search Algorithm (BSA) and original Slime Mould Algorithm (SMA) using IEEE 33 and 69 bus radial distribution systems, demonstrating superior performance in maintaining load and voltage stability
DOI