RbT: Real-time Carbon and WeatherBench2 Model Evaluation Platform

Overview

RbT is a platform for analyzing, predicting, and evaluating carbon emissions using both real-time UK carbon intensity data and the WeatherBench2 global weather benchmark. At its core, RbT features a Kolmogorov-Arnold Network (KAN) model with a self-improving feedback loop, enabling adaptive, AI-driven environmental forecasting and continuous learning from user feedback.

Core Essence: KAN & Self-Improving Loop

• Kolmogorov-Arnold Network (KAN):
  • The backend model (EcoKAN) is based on KANs, designed for nonlinear, high-dimensional forecasting of carbon emissions and SDG scores.
  • The model architecture is flexible and can evolve over time.
• Self-Improving Feedback Loop:
  • User and system feedback on predictions is collected and stored.
  • When enough new feedback is available, the model is retrained automatically (scripts/retrain.py), updating its weights and architecture for improved accuracy.
  • Each retraining cycle is versioned and performance is tracked, supporting a true self-improving AI workflow.

Features

• Home page with three options: Real-time UK Carbon, WeatherBench2 Evaluation, and Compare Both
• Real-time data analysis and visualization
• Model evaluation on WeatherBench2 public datasets (cloud Zarr)
• Feedback loop and retraining support for continuous model improvement

Quick Start

Prerequisites

• Python 3.9+ (for backend)
• Node.js 16+ and npm (for frontend)
• (Windows) Microsoft C++ Build Tools (for some Python dependencies)

Backend Setup

1. Install dependencies:

   pip install --upgrade pip setuptools wheel
   pip install -r rbt-web/requirements.txt
   pip install git+https://github.com/google-research/weatherbench2.git

2. Start the backend:

   uvicorn rbt-web.backend.app:app --reload

Frontend Setup

1. In a new terminal:

   cd rbt-web/frontend
   npm install
   npm start

2. Open http://localhost:3000 in your browser.

Usage

• Home Page: Choose between Real-time UK Carbon, WeatherBench2 Evaluation, or Compare Both.
• /realtime: View live UK carbon intensity and grid mix analysis.
• /weatherbench: Evaluate the model on WeatherBench2 public datasets (no local download needed).
• /compare: Compare model performance and data analysis side-by-side.

WeatherBench2 Integration

• Uses official WeatherBench2 Python package and public cloud Zarr datasets.
• No need to download large datasets locally; data is streamed from GCS.
• Requires gcsfs, xarray, zarr, and apache-beam (see requirements.txt).

Notes

• For Windows users: If you see C++ build errors, ensure you have the "Desktop development with C++" workload installed via Visual Studio Installer.
• For Docker: Update your Dockerfile to include all Python dependencies if deploying in containers.

License

Apache-2.0