Code (and demo data) for the study: “Electric vehicles accelerate pavement damage but deliver net climate and cost benefits”.

This repository is organized to match the paper workflow:

1. Fitting & shifting axle load spectra (ALS) under EV scenarios
2. Pavement survival analysis (RSF) under EV traffic
3. Pavement roughness (IRI) modeling (XGBoost) under EV traffic

Repository structure (three modules):

• als-fitting-shifting/ — fits parametric ALS models and applies Monte Carlo-based shifting under EV scenarios
• pavement-survival-analysis/ — Random Survival Forest (RSF) time-to-threshold modeling (example: asphalt IRI)
• pavement-roughness-modeling/ — XGBoost regression for IRI evolution (asphalt and concrete)

Each module includes core scripts plus small demo Excel datasets so you can run examples without the full LTPP extraction pipeline.

---

Quickstart (demo run)

1) Create a Python environment

Recommended: Python 3.9–3.11.

Option A — conda (recommended for scikit-survival):

conda create -n ev_pavement python=3.10 -y
conda activate ev_pavement
conda install -c conda-forge numpy pandas scipy scikit-learn matplotlib joblib -y
conda install -c conda-forge scikit-survival lifelines optuna -y
pip install xgboost scikit-optimize shap seaborn

Option B — pip only (may be harder for scikit-survival on some OS):

pip install numpy pandas scipy scikit-learn matplotlib joblib xgboost scikit-optimize shap seaborn optuna lifelines scikit-survival

2) Run each module (from its folder)

ALS fitting & shifting

cd als-fitting-shifting
python als_fitting_example.py
python als_shifting_example_1.py
python als_shifting_example_2.py

Survival analysis (RSF)

cd ../pavement-survival-analysis
python RSF_IRI_asphalt.py

IRI modeling (XGBoost)

cd ../pavement-roughness-modeling
python IRI_xgb_asphalt.py
python IRI_xgb_concrete.py

Notes on reproducibility

• Some scripts were exported from notebooks and may contain plotting/output sections intended for demonstration.
• Paths: If you see absolute local paths (e.g., saving a scaler to a user-specific directory), replace them with relative paths (e.g., ./outputs/).
• Runtime: hyperparameter search can be compute-intensive (Optuna trials / Bayesian search iterations). For a quick smoke test, reduce:
  • n_trials in Optuna
  • n_iter in BayesSearchCV

---

Using your own data

Demo Excel files are included in each module. For your own datasets:

• Keep the same column names expected by the scripts (see each module README for schema).
• Put your input Excel next to the script (or update pd.read_excel(...) paths accordingly).

---

License

MIT License (see LICENSE).