Plantwide surrogate modelling playground

This repo is a small sandbox for building and experimenting with surrogate models on process data.

• Environment is managed with uv and a local .venv (see pyproject.toml).
• Core dependencies: pandas, numpy, scikit‑learn, xgboost, matplotlib, Jupyter.
• All data files live under the repo (no external services required).

For now, the main work lives in the analysis1/ directory.

analysis1/ – dynamic surrogate analysis

analysis1 contains a first warm‑up case where we build surrogate models for a simple dynamic process:

• Inputs: U1, U2
• Outputs: Y1, Y2

Inside analysis1 you’ll find:

• A notebook that:
  • loads synthetic training/testing data from Excel,
  • trains static models (Random Forest, XGBoost) that use only current inputs,
  • trains dynamic models that also use short histories of U1, U2, Y1, Y2 as features,
  • compares train/test metrics and plots actual vs predicted trajectories.
• A short README in analysis1/ that explains the dynamic setup and how to interpret the plots.

Use this as a template for future cases: drop new data/configs into a sibling analysisX/ folder and clone the notebook to try different model structures or feature choices.