Projects · 2025–2026
Research projects
Each project has its own page with the research question, the method, a step-by-step pipeline and the current status of the paper.
All projects
- RF-based UAV detection under frequency-hopping shifts Treating a frequency-hopping-related change in RF drone features as a transformation to identify and undo, so that an existing detector keeps working without retraining.
- Attack localization in federated solar forecasting A federated solar forecaster whose shared update is small and physically interpretable, so the server can not only resist a corrupted update but say which part of the forecast was attacked.
- Quantile-calibrated ensemble for daily GHI forecasting A chronology-preserving stacked ensemble for daily solar irradiance over central India, with quantile forecasts calibrated by conformal prediction so that its uncertainty can be trusted.
- Look-ahead leakage in decomposition-based solar forecasting Measuring how much decomposition-based solar forecasters overstate their accuracy when the decomposition is computed over the whole series, and comparing models fairly once it is computed causally.
- Causal multi-scale PV forecasting A multi-horizon irradiance forecaster that is causal by construction: band forecasts from a trailing-window wavelet decomposition, combined with horizon-dependent weights and calibrated prediction intervals.
- Energy-bounded routing and wireless charging for electric shuttle fleets Planning routes and wireless charging for an electric shuttle fleet in two stages that can each be checked: a feasibility-first router, then an exact per-route charging optimiser.
- Rooftop solar PV and India’s energy transition Why the same national subsidy for rooftop solar produces very different results across Indian states, and how solar readiness should be judged across countries: by mechanism, not by installed capacity alone.
- EV charging load forecasting A neural-network forecaster for electric-vehicle charging load built from real charging sessions, and what its patterns mean for demand-side management of the grid.
- Real-time Indian Sign Language translation A camera-only approach to recognising Indian Sign Language gestures and translating them into English, comparing convolutional, recurrent and Transformer configurations.
- RfGANNNet 2.0 for rainfall forecasting A review of machine-learning approaches to rainfall forecasting and a proposed hybrid framework that combines random forests, spatio-temporal graph convolution and a physics-guided GAN.