Skip to content

Latest commit

 

History

4 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

CarbonCast — CBAM Transition Risk Analysis Platform

"AI estimates, Blockchain tracks, Finance connects."

CarbonCast is a CBAM (Carbon Border Adjustment Mechanism) transition risk analysis platform developed for the Hana Financial Group Youth Talent Development Project. It helps Korean export companies and financial institutions quantify, verify, and simulate the cost impact of the EU's carbon border tax.

What is CBAM?

The EU Carbon Border Adjustment Mechanism (EU Regulation 2023/956) imposes carbon costs on imports of steel, cement, aluminum, fertilizers, hydrogen, and electricity starting January 2026. The phase-in schedule gradually increases from 2.5% (2026) to 100% (2034).

The core problem: 87% of Korean SME exporters don't know how to calculate their emissions (KIEP survey). Without verified data, the EU applies default values with a +30% markup — meaning companies pay 2x or more in carbon costs compared to their actual emissions.

Architecture

CarbonCast consists of three layers:

┌─────────────────────────────────────────────────────┐
│                   CarbonCast                         │
│                                                     │
│  ┌───────────────────────────────────────────────┐  │
│  │            AI Engine                           │  │
│  │  Emission estimation (Conformal Prediction)    │  │
│  │  Industry impact analysis (I-O + GNN)          │  │
│  │  Compliance Agent (LLM + RAG)                  │  │
│  └───────────────────────────────────────────────┘  │
│                        ↕                            │
│  ┌───────────────────────────────────────────────┐  │
│  │          Trust Layer                           │  │
│  │  Blockchain audit trail (SHA-256)              │  │
│  │  Digital Product Passport (DPP)                │  │
│  │  Federated Learning                            │  │
│  └───────────────────────────────────────────────┘  │
│                        ↕                            │
│  ┌───────────────────────────────────────────────┐  │
│  │        Financial Layer                         │  │
│  │  READIT OCR integration                        │  │
│  │  PCAF portfolio carbon analysis                │  │
│  │  Credit assessment + ESG finance               │  │
│  └───────────────────────────────────────────────┘  │
└─────────────────────────────────────────────────────┘

Module 1: AI Quality Scoring

Pre-screens emission data reliability before EU accredited verifier review. Uses Isolation Forest + rule-based checks on emission factors, energy intensity, and production volumes. Outputs a 0-100 quality score.

  • Training data: K-ETS registry (~800 facilities, data.go.kr)
  • Scoring: z-score based weighted sum across 4 criteria (emission factor range, energy range, energy-emission consistency, production scale)

Module 2: Blockchain Verification

Records quality-verified emission data on blockchain for tamper-proof audit trails. Stores SHA-256 hashes on-chain with metadata; full data remains off-chain.

  • Prototype: Ethereum Sepolia testnet
  • Production target: Hyperledger Fabric (permissioned)

Module 3: CBAM Cost Simulation

Calculates year-by-year CBAM cost paths (2026-2034) under 4 NGFS climate scenarios.

CBAM Cost = Export Volume × [Emission Factor − Benchmark × (1 − Phase-in Rate)]
            × EUA Price × Exchange Rate
  • Phase-in rates: 2.5% (2026) → 100% (2034) per EU Regulation 2023/956, Annex IV
  • Scenarios: Net Zero 2050, Below 2°C, Delayed Transition, Current Policies (NGFS Phase V)

Project Structure

├── README.md
├── index.html              # Interactive demo (standalone HTML)
├── CarbonCast_app.html     # Full demo application
├── CarbonCast_v2.html      # Demo v2
├── src/
│   ├── 01_fetch_data.py    # EU ETS + market data collection
│   ├── 02_model.py         # LightGBM quantile price model
│   ├── 03_cbam_engine.py   # CBAM cost calculation engine
│   ├── 04_synthetic_data.py# Synthetic K-ETS data generation
│   ├── 05_emission_model.py# Emission factor prediction model
│   ├── app.py              # Streamlit dashboard v1
│   └── app_v2.py           # Streamlit dashboard v2
├── data/
│   ├── cbam_scenarios.json  # NGFS scenario parameters
│   ├── cbam_analysis.csv    # CBAM cost analysis results
│   ├── predictions.csv      # Model predictions
│   ├── model_results.json   # Market model metrics
│   └── emission_model_results.json
└── docs/
    ├── CarbonCast_Architecture.md
    └── [최종] CarbonCast 통합 제안서.md

Key Results (Prototype)

Metric Value
Emission model (with energy data) R² = 0.778, MAPE 8.0%
EU ETS price model (LightGBM) MAE 3.15 EUR, MAPE 4.4%
Steel: EU default vs actual savings ~€154/ton (~22억 won per 10,000t)
Conformal prediction coverage ≥90% (distribution-free guarantee)

CBAM Phase-in Schedule

Year 2026 2027 2028 2029 2030 2031 2032 2033 2034
Rate 2.5% 5% 10% 22.5% 48.5% 61% 73.5% 86% 100%

Source: EU Regulation 2023/956, Annex IV

Tech Stack

  • ML/AI: LightGBM, XGBoost, MAPIE (Conformal Prediction), Isolation Forest
  • Visualization: Streamlit, Plotly, HTML/JS Canvas
  • Blockchain: Web Crypto API (SHA-256), Ethereum Sepolia (prototype)
  • Data Sources: K-ETS registry, NGFS Phase V, EU ETS (ICE), DART Open API

Data Sources

Data Source Access
K-ETS facility data data.go.kr/15053947 Public API
NGFS scenarios ngfs.net Public
EU ETS prices ICE Endex / Yahoo Finance Free (delayed)
Industry I-O table Bank of Korea (ecos.bok.or.kr) Public
Company financials DART Open API Public
CBAM HS code mapping Korea Customs Service Public (2026.02)

Context

This project was developed for the Hana Youth Financial Talent Development Project (하나 청년 금융인재 양성 프로젝트), a competition organized by Hana Financial Group. The platform is part of the broader ESG TradeGuard system, which provides end-to-end trade finance compliance checking including OCR document parsing, regulation matching, carbon verification, and satellite environmental monitoring.

References

  • EU Regulation 2023/956 (CBAM)
  • EU IR 2025/2621 (Default values and benchmarks)
  • NGFS Phase V Scenarios (2024.11)
  • ESPR 2024/1781 (Digital Product Passport)
  • Angelopoulos & Bates (2021) "Conformal Prediction" arXiv:2107.07511
  • Stanford AAAI 2025: "Learning Production Functions for Supply Chains with GNNs"

License

This project is for educational and competition purposes.

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages