Topology-Induced Limits to Voltage Support: The Reactive Locality Barrier in Deep Reinforcement Learning

OVERVIEW: This repository contains the source code, environment models, and datasets required to reproduce the findings presented in the paper. The framework demonstrates how a Deep Reinforcement Learning (DRL) agent identifies topological barriers ("Dark Zones") in voltage control tasks. CONTENTS: 1. createₙetwork. py: Generates the 'stressed' IEEE 118-bus network model (pandapower json). 2. gridₑnv. py: The Gym-compatible environment representing the power system dynamics. 3. ppoₐgent. py: Implementation of the PPO (Proximal Policy Optimization) algorithm. 4. trainₐgent. py: Main training script featuring the curriculum learning and limit discovery logic. 5. trainingₕistory. csv: The actual training logs used to generate the figures in the manuscript. 6. sensitivityₜable. csv: The open-loop sensitivity data confirming the "Dark Zone". HOW TO RUN: 1. Install dependencies: pip install numpy pandas pandapower gym torch 2. Generate the network model: python createₙetwork. py (This will create 'case118ₛtressed. json') 3. Run the training agent: python trainₐgent. py NOTE: The pre-trained model weights are provided in 'preₜrainedₘodels. zip' for direct evaluation.