CC_OPF_Constraint_Informed_Randomness

Table of Contents

• Overview
• Requirements
• Repository Structure and Usage
• Citation

Overview

This repository contains all the implementations needed to reproduce the computational studies in the paper, which include simulating different datasets as randomness, optimization models to solve chance-constrained optimal power flow (cc-opf), out-of-sample reliabitilies validation. In addition, this repository contains the IEEE test cases system in MATPOWER format, where users can use their own preferred test cases or modify.

Installation

Clone and install requirements for Python scripts:

git clone https://github.com/Subramanyam-Lab/Constraint_Informed_CCOPF
cd Constraint_Informed_CCOPF
conda create -n constraint_informed_ccopf python=3.11 -y
conda activate constraint_informed_ccopf
pip install -r requirements.txt

In the previously cloned folder, do the following to set up Julia environment:

git clone https://github.com/Subramanyam-Lab/Constraint_Informed_CCOPF
cd Constraint_Informed_CCOPF
julia --project=. -e 'using Pkg; Pkg.instantiate()'

To install an academic license of the Gurobi solver, please visit (https://www.gurobi.com/academia/academic-program-and-licenses/).

Repository Structure and Usage

1. Datasets

The datasets used for statistical fitting and OPF optimization model are both synthetic and real-world data.

• Gaussian synthetic: Synthetic forecast errors that are Gaussian distributed.
• Cauchy synthetic: Synthetic forecast errors that are Cauchy distributed.
• NordPool Real: Sourced from NordPool production & production forecast over 2 weeks (15 consecutive days) in March. The file contains wind power generation forecasts and actual (in MW) data for 10 European regions at a 15-minute time window. In experiments, these errors will be normalized.

2. Code

• Run this code to get a simple demo of the advantage of constraint-informed over classical on a synthetic Cauchy distributed data and to obtain results in Table 2 and Figure 1 from the paper.

python toy_example_methodology.py 

• Run these two scripts to plot the constraint-informed and classical fitting of Synthetic-C and NordPool datasets in Fig. 3

python Cauchy_data_simulation.py

python NordPool_data_simulation.py

• Run the following script to store BIC values for K=1,2,...,10 when fitting Synthetic-C in 10d and 50d

python Cauchy_data_BIC_scalability.py

• Run the following script to load the BIC values and plot Fig. 6

python Cauchy_scalability_BIC_plot.py

• In Julia, run these scripts to output the PTDF matrix, which will be used for 2D fittings of eta_l

include("add_wind.jl")

include("add_wind_118.jl")

• Run this script to implement the algorithm in Fathabad et al (2023) that finds the optimal placement of piecewise linear approximation by specifying a tolerance, default set to 0.005. The slopes and intercepts will be saved to pwl_segments.csv in the current directory

python pwl.py

Synthetic-C Dataset Fitting and Optimization for 118-bus and Polish case

In Python, run these scripts to do 10 different estimations (seeds 42-51) of Omega (aggregate errors) on the 118 and Polish case with zero means, and write all estimated parameters to the ../data directory.

python Omega_118_zeromean.py

python Omega_polish_zeromean.py

In Python, run these scripts to do 10 different estimations (seeds 42-51) of eta_l on the 118 and Polish case with zero means, using the PTDF matrices generated in the previous step, writing all estimated parameters to the ../data directory.

python eta_l_estimation_118_zeromean.py

python eta_l_estimation_polish_zeromean.py

In Julia, run these scripts to do the CC-OPF optimization using constraint-informed parameters. The worst-case violation of 10 runs (seeds 42-51) will be saved to ../data/118_ci_Cauchy_optimization_summary.csv and ../data/polish_ci_Cauchy_optimization_summary.csv. These are the quantities plotted in Fig. 4.

include("GMM_reformulation_118.jl")

include("GMM_reformulation_polish.jl")

NordPool Dataset Fitting and Optimization for 118-bus

The NordPool dataset is limited in data samples. Use the following script in Python to partition the dataset and perform constraint-informed fitting. The estimated parameters will be written to ../data directory.

python NordPool_data_fitting_CI.py

Use the following script to do the CC-OPF optimization using constraint-informed parameters on the NordPool dataset. The worst-case violation of 10 runs (seeds 42-51) will be saved to ../data/118_ci_NordPool_optimization_summary.csv. These are the quantities plotted in Fig. 4.

include("NordPool_GMM_CI.jl")

In Python, the following script will estimate 10-dimensional GMM of joint forecast errors using the classical approach. The estimated parameters will be written to ../data directory.

python NordPool_data_fitting_Classical.py

In Julia, the following script will use these estimated parameters to solve the CC-OPF on 118-bus system. The worst-case violation of 10 runs (seeds 42-51) will be saved to ../data/118_classical_NordPool_optimization_summary.csv. These are the quantities plotted in Fig. 4.

include("NordPool_GMM_Classical.jl")

Citation

If you have ever used the datasets and estimation/optimization codes, please cite the following paper:

@article{yi2025chance,
  title={Chance-Constrained DC Optimal Power Flow Using Constraint-Informed Statistical Estimation},
  author={Yi, Tianyang and Maldonado, D Adrian and Subramanyam, Anirudh},
  journal={arXiv preprint arXiv:2508.21687},
  year={2025}
}