Research And Results

Type: Agentic Computes electricity prices endogenously based on the simulation of strategic bidding behavior of prototyped market actors.

Type: Foundation Microsoft’s weather prediction model, available on Azure. It can predict global weather patterns and atmospheric processes such as air pollution.

Type: GAN (Generative Adversarial Networks) Generative model that takes a normal scene and synthesizes realistic visual effects of climate‑related disasters such as floods, wildfires, and smog by predicting where the event should appear and rendering it into the image.

Type: Transformer This model classifies patents into "green patents" or "no green patents" based on their titles.

Type: NN (Neural Network) An advanced neural architecture for multi‑step energy load prediction and time‑series forecasting.

Type: Machine‑learning regression (Random Forest) A custom model that forecasts California electricity prices ($/kWh) using features like EV charging demand, solar and wind output, carbon emissions, and storage levels.

Type: Transformer A model that detects contamination, identifies its source and type, measures its extent, and evaluates how contaminants move through groundwater, surface water, and nearby water bodies.

Type: Neural Sparse Retriveal Converts queries and documents into high‑dimensional sparse vectors, where each non‑zero dimension corresponds to a specific vocabulary token and its value reflects that token’s importance for nuclear specific tasks.

Type: NN (Neural Network) Predicts global atmospheric dynamics of various weather and climate variables.

Type: Transformer A cascade machine learning forecasting system for 15-day global weather forecast.

Type: RAG (Retrieval-Augmented Generation) GreenChat is a domain-specific RAG model designed to support environmental decision-making across multiple domains relevant to UN SDGs.

Type: LSTM (Long Short-Term Memory)+PPO (Proximal Policy Optimization) Hybrid LSTM–PPO system for grid optimization, combining weather‑based demand forecasting with reinforcement‑learning control to reduce blackout risk.

Type: Transformer A novel transformer-based framework for image restoration under adverse weather conditions.

Type: Multi-Agent A testbed for the implementation of Multi-Agent Reinforcement Learning (MARL) in building energy coordination and demand response in cities. 

Type: GNN (Graph Neural Network) GNN incorporating grid topology for power‑flow analysis.

Type: Transformer Replaces standard attention with a power‑law‑weighted causal attention mechanism to better capture local, time‑ordered dependencies in time‑series data.

Type: Multi-Agent Power demand forecasting in an on-policy, cooperative MARL algorithm for voltage control problem in the isolated Microgrid system by incorporating a differentiable, learning-based communication protocol, a spatial discount factor, and an action smoothing scheme.

Type: Gradient Boosted Tree Leverages machine learning, satellite imagery, and weather data to predict solar energy output, making it a plug-and-play solution for generating forecasts.

Type: CNN (Convolutional Neural Network) A sky image-based deep convolutional neural network for intra-hour solar forecasting.

Type: LLM (Large Language Model) A base model designed specifically for nuclear power domain as a research tool, responds to chats in a chat-based environment.

Type: LLM (Large Language Model) An instruction model designed specifically for nuclear power domain as a research tool, responds to chats in a chat-based environment.

Type: Foundation Heliophysics model trained on 14 years of observations from NASA’s Solar Dynamics Observatory, helping protect critical infrastructure from space weather.

Type: Transformer Builds and evaluates Transformer‑based neural networks to forecast future electrical load from historical time‑series data.re of critical importance for Transmission System Operators (TSOs) to match electricity supply and demand.

Type: Transformer A PyTorch implementation of a Transformer-based time series model for forecasting electricity load diagrams (hourly).

Type: LSTM (Long Short-Term Memory) Improving the predictions of power generated using wind energy and LSTM as machine learning model to perform model optimization.

Type: GNN (Graph Neural Network) Predicting wind plant performance. It represents the wind plant a graph with nodes representing individual turbines and wake effects encoded by directed edges.