Platform-Based Design with Limit of Performance Analysis

This repository contains the models used for the paper about Platform-Based Design with limit of performance analysis.

The directory structure is as follows:

- modelica: Contains Modelica models
  - Buildings: Git submodule pointing to the Buildings library
  - IndustrialPilot: Models for the industrial pilot simulation
  - Modelica_Requirements: Git submodule pointing to the Modelica_Requirements library
- controllers: Contains the black-box optimal control models
  - HighResolution: The high-resolution controller
  - MultiResolution: The multi-resolution controller
  - MultiResolutionML-RF1: The multi-resolution with Random Forests (κ=1) controller
  - MultiResolutionML-RF2: The multi-resolution with Random Forests (κ=2) controller
  - MultiResolutionML-GB: The multi-resolution with Gradient Boosting controller
- IDAES: Contains the models for the architectural optimization problem
  -idaes_models/
    - data: Data on annual electricity cost and solar profiles used for architectural optimization.
    - models: Python scripts containing the unit models, cost models and full IDAES flowsheet.

Modelica

To simulate a Modelica model with Dymola on Linux, use

$ cd modelica/
$ export MODELICAPATH=`pwd`/Buildings:`pwd`/Modelica_Requirements
$ cd IndustrialPilot/
$ dymola package.mo &

The top-level models are in the package IndustrialPilot.Examples.

Controllers

The controllers folder contains all of the controllers implemented in the paper. Each controller is packaged with the Dymola FMU evaluator (data/fmu_edit.fmu) and the JADE optimizer implemented in Python (_JADE.py).

The full code implementation (with all hyperparameters as outlined in the paper) for all multi-resolution controllers is located in multires.py, while the high-resolution controller is in highres.py. The requirements.txt file contains all Python dependencies.

Running Locally

To run the controller locally, navigate into the specific controller folder (e.g., MultiResolutionML-GB) and execute the Python script:

cd MultiResolutionML-GB
python multires.py

Running with Apptainer (Reproducibility)

To exactly replicate the environment used in the paper, run the controllers using Apptainer (tested with version 1.4.2):

cd MultiResolutionML-GB
apptainer exec --bind .:/app docker://python:3.11-bookworm /app/run_test.sh

The run_test.sh script contains the specifications regarding the controller execution (including dependency installation).

---

IDAES

The IDAES folder contains the unit models and flowsheets for the multi-objective architectural optimization problem presented in the in the paper. To run the idaes model for the pilot, from inside the IDAES directory, run

python -m idaes_models.models.stepload_case.flowsheets.heating_flowsheet_stepload_additional_constraints 

Due to the size of the optimization problem, running the IDAES model for the full year takes a few days. However, the model can be tested/run for a shorter time interval by passing in the --days flag. For example, to optimize over the first 7 days of the year, run

python -m idaes_models.models.stepload_case.flowsheets.heating_flowsheet_stepload_additional_constraints --days 7

Instructions for installing the IDAES model are in the IDAES folder README.

---