Ongoing Research Projects

Solar Energy Forecasting

This study proposes a hybrid framework integrating a Transformer-based deep learning model for solar radiation forecasting with a Deep Deterministic Policy Gradient (DDPG) reinforcement learning agent for optimizing battery energy storage system (BESS) management in a photovoltaic (PV)-powered microgrid. Leveraging historical meteorological data from the National Solar Radiation Database (NSRDB) India 2014 dataset, the Transformer model predicts Global Horizontal Irradiance (GHI), which is used to estimate PV power output. This predicted PV power drives the DDPG agent, trained over 1000 episodes using MATLAB, to dynamically manage BESS charge/discharge rates. Compared to a rule-based baseline controller, the hybrid approach achieves superior performance, with an energy efficiency of 98.5% (vs. 85.2% for the baseline) and a 64% reduction in unmet demand over a 1749-hour simulation, alongside greater battery utilization.

 Integrating Vision Transformers and Clinical Data for Early Skin Lesion Classification

Skin lesions are key indicators of potential malignancies, particularly in regions where access to dermatological services is limited. This study proposes a dual-stream learning model for early detection and classification of skin lesions, supporting Sustainable Development Goal 3, which focuses on ensuring healthy lives and improving access to healthcare. The model integrates image-based features extracted using a Modified Vision Transformer with patient clinical data processed through Feed Forward Neural Networks. A Progressive Image Editing module is also introduced to simulate realistic lesion progression using the PAD-UFES-20 dataset. Evaluated on 2,298 images with corresponding clinical data, the model achieved 92% accuracy in classifying six lesion categories. The approach enhances diagnostic accuracy and supports improved disease monitoring and accessibility to early cancer detection.