EV-AI-CCUS Regional Energy Pressure Intelligence System

Overview

This project combines electric vehicle demand, AI data center electricity demand, and CCUS project capacity into a reproducible regional energy pressure analysis workflow.

The main research question is:

Do EV adoption and AI data center expansion create regional electricity pressure, and can CCUS capacity act as a mitigation-readiness proxy?

Current Deliverables

The current final presentation artifact is:

EV_AI_CCUS_Interactive_Dashboard.html

It is a standalone interactive HTML dashboard at the repository root. It supports year, region, EV scenario, AI scenario, and CCUS scenario filters, plus hover tooltips and synchronized tables.

Final PNG exports are stored in results/:

results/pressure_ranking_2030.png
results/combined_demand_by_region.png
results/ev_ai_mix_2030.png
results/ccus_buffer_ratio_2030.png
results/scenario_pressure_comparison_2030.png
results/feature_importance.png
results/cluster_scatter.png

Older SVG and prototype HTML outputs have been removed from results/.

Data Sources

Canonical source files live in data/raw/. Any same-named CSV/XLSX files in the repository root are duplicate convenience copies and are not required by the current pipeline.

File	Description
IEA-EV-Data-2023.csv	IEA EV data, 2023 release
IEA-EV-Data-2024.csv	IEA EV data, 2024 release
IEA-EV-Data-2025.xlsx	IEA EV data, 2025 release, GEVO_EV_2025 sheet
Data_Energy_and_AI.xlsx	AI data center capacity, electricity consumption, PUE, load factor, and scenarios
IEA-CCUS-2026.xlsx	IEA CCUS project database

Project Structure

config/
  pressure_score_weights.json

data/
  raw/          Raw source files copied from the repository root
  interim/      Reserved for intermediate files
  processed/    Fact tables, feature tables, quality reports, model outputs
  warehouse/    DuckDB analytical database

logs/           Pipeline logs
results/        Final PNG chart exports
scripts/        Environment, pipeline, dashboard, and test helpers
src/            Main pipeline source code
tests/          Pytest smoke and integration tests

EV_AI_CCUS_Interactive_Dashboard.html
starter_ev_ai_project.py
requirements.txt
requirements-optional.txt

starter_ev_ai_project.py is kept as a legacy exploratory reference. The modular pipeline lives under src/.

Installation

On Windows:

py -3 -m venv .venv
.\.venv\Scripts\python.exe -m pip install --upgrade pip setuptools wheel
.\.venv\Scripts\python.exe -m pip install -r requirements.txt

Optional research dependencies:

.\.venv\Scripts\python.exe -m pip install -r requirements-optional.txt

DuckDB is the primary database backend. SQLite fallback is only allowed when ALLOW_SQLITE_FALLBACK=1 is explicitly set.

Pipeline Commands

.\.venv\Scripts\python.exe scripts\check_env.py
.\.venv\Scripts\python.exe src\pipeline.py --step all

Individual steps:

.\.venv\Scripts\python.exe src\pipeline.py --step env
.\.venv\Scripts\python.exe src\pipeline.py --step schema
.\.venv\Scripts\python.exe src\pipeline.py --step ingest
.\.venv\Scripts\python.exe src\pipeline.py --step features
.\.venv\Scripts\python.exe src\pipeline.py --step scenarios
.\.venv\Scripts\python.exe src\pipeline.py --step models
.\.venv\Scripts\python.exe src\pipeline.py --step explain
.\.venv\Scripts\python.exe src\pipeline.py --step dashboard
.\.venv\Scripts\python.exe src\pipeline.py --step quality

Convenience scripts:

.\scripts\run_pipeline.ps1
.\scripts\run_tests.ps1
.\scripts\run_dashboard.ps1

Main Data Outputs

Key files in data/processed/:

File	Purpose
fact_ev.csv	Ingested EV fact table
fact_ai_energy.csv	Ingested AI energy fact table
fact_ccus_project.csv	Standardized CCUS project fact table
feature_ev_region_year.csv	EV region-year features
feature_ai_region_year.csv	AI region-year features
feature_ccus_region_year.csv	CCUS region-year cumulative capacity
ml_region_year_features.csv	Core ML-ready feature table
scenario_region_year_features.csv	Scenario-expanded region-year features
scenario_comparison_2030.csv	2030 scenario comparison table
model_regression_metrics.csv	Regression model metrics
model_feature_importance.csv	Model feature importance
region_clusters.csv	Regional clustering output
data_quality_report.csv	Data quality report
environment_report.csv	Environment validation report
database_schema_report.csv	Database schema report

Database

The main warehouse is:

data/warehouse/energy_ai.duckdb

Main tables:

fact_ev
fact_ai_energy
fact_ccus_project
dim_region
feature_ev_region_year
feature_ai_region_year
feature_ccus_region_year
ml_region_year_features
scenario_region_year_features
scenario_comparison_2030
pipeline_metadata

Pressure Score

The regional pressure score is normalized to 0-100. Higher values indicate higher pressure.

The current formula combines:

demand_pressure_component
ev_growth_pressure_component
ai_growth_pressure_component
ccus_buffer_component

Weights live in:

config/pressure_score_weights.json

Default weights:

demand: 0.35
ev_growth: 0.20
ai_growth: 0.20
ccus_buffer: 0.25

Pressure classes:

Score	Class
< 25	Low Pressure
25-49	Medium Pressure
50-74	High Pressure
>= 75	Critical Pressure

Models

When scikit-learn is available, the project uses:

Task	Models
Regression	Ridge, RandomForestRegressor, GradientBoostingRegressor
Clustering	KMeans with PCA coordinates
Classification	LogisticRegression, RandomForestClassifier, GradientBoostingClassifier

Numpy fallbacks are retained for restricted environments.

Tests

Run tests after the pipeline has produced the DuckDB database:

.\.venv\Scripts\pytest.exe

The tests validate the database backend, non-empty fact and feature tables, CCUS standardization, pressure score ranges, scenario engine output, and dashboard files.

Notes On Cleanup

The old root-level processed/ folder and Anaconda file-browser cache are obsolete and can be removed. The canonical processed outputs now live under data/processed/.

.pytest_cache/ is ignored. If Windows leaves an ACL-locked empty .pytest_cache directory behind, it is safe to leave it in place because it is not part of the project.

Known Limitations

• CCUS capacity is a mitigation-readiness proxy, not a physical offset for electricity demand.
• Region mapping is intentionally conservative and should be expanded before publication-level analysis.
• AI regional scenario coverage is limited outside the workbook's world-level scenario rows.
• The final interactive HTML is static and self-contained; it does not refresh automatically when CSV files change.

Future Work

• Add power mix, renewable share, carbon intensity, GDP, population, and electricity price data.
• Add SHAP or permutation-based explainability.
• Add richer Streamlit drill-down pages.
• Add CI checks for environment, pipeline, tests, and dashboard generation.