This repository contains the codebase for STORM-OPF: Robustness Benchmarking for AC-OPF Methods.

STORM-OPF Workflow

Repository Overview

The core implementation of the Scenario Generator Module and the Evaluator Module is located in src/.

All baseline implementations are provided under src/baselines. Each baseline directory includes its own README.md file with instructions for training and evaluation.

Complete Pipeline

The overall STORM-OPF workflow is as follows:

1. Choose the correct env Using virtual environments is recommended, since different baselines may require different dependencies. In this implementation, all baselines use the same environment specified in env_opf.yml, except for the OPF-DNN baseline. The OPF-DNN baseline requires a separate environment defined in src/baselines/OPF_DNN/env_opf_dnn.yml.

2. Generate stress-test scenarios.
   First, choose the stress scenario(s) on which you want to evaluate an OPF solver. In this work, we define 10 scenarios, including the in_distribution case. If you want to use these predefined scenarios, run src/create_scenario_bank.py to generate the scenario networks. See src/README.md for additional details.

   To add a new stress scenario:

   • add the scenario name to config.json, and
   • register and implement the scenario in src/stress_test_scenarios.py.

3. Train and evaluate baselines.
   The repository currently includes 5 baselines. To train or evaluate any of them, refer to the corresponding README.md file in that baseline's folder under src/baselines.

4. Add a new baseline.
   To integrate a new baseline into STORM-OPF:

   • place the baseline implementation inside src/baselines,
   • use the generated training and test scenarios to prepare the required datasets, and
   • extend the appropriate evaluator from the relevant src/global_evaluation_... file depending on the OPF setting, such as AC-OPF, SC-ACOPF, DC-OPF, or SC-DCOPF.