Physics-Informed-Neural-Networks-for-AC-Optimal-Power-Flow

This repository contains the code for Physics-Informed Neural Network for AC Optimal Power Flow applications and the worst case guarantees When publishing results based on this data/code, please cite: R. Nellikkath and S. Chatzivasileiadis "Physics-Informed Neural Networks for AC Optimal Power Flow", 2021. Available online: https://arxiv.org/abs/2107.00465

Author: Rahul Nellikkath E-mail: rnelli@elektro.dtu.dk

This code is distributed WITHOUT ANY WARRANTY, without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

This code requires the following:

conda(python) and tensor flow installations with environment activated Matlab (R2018b) or above YALMIP (https://yalmip.github.io/download/) Gurobi (Gurobi V9.1.2, https://www.gurobi.com/) MATPOWER (matpower7.1, https://matpower.org/) The data for the test cases are reproduced from the IEEE PES Power Grid Library - Optimal Power Flow - v19.05 (https://github.com/power-grid-lib/pglib-opf)

To run the code to re-create the simulation results please follow the steps below: Copy network details from "Test_Network(Copy to Matpower)\Modified Network" to matpower folder 0.Run "A1_Create_Data_Sets" to generate the data sets 1.Run PINN_DC_OPF_Main.py in python: 1.1 It contains the python code used for the PINN algorithms 1.2 Teset cases can be 39 bus system, 118 bus system or 162. They can be changed by changing the "n_buses" 1.3 All the fuctions used by the algorith is given in Folder PINNs 1.4 After the training the weigts and biases will be stored in the respective folder in:"MILP_For_Worst_Case_Guarantees\Trained_Neural_Networks" 2. Run "Evaluate_Average_And_Worst_Performance_Data.m" from "MILP_For_Worst_Case_Guarantees"to get statistical average and worstcase results from the data 3. The Run "E2" and "E3" to get the worst case performance and run "E4" for Input Domarin Reduction results