HyPowerFlow

HyPowerFlow algorithm accelerates power flow convergence across multiple grid scenarios by leveraging GPU parallelism, data compression, structured hyper-sparsity, and refined initial estimates. Efficient data formatting optimizes GPU utilization and eliminates redundant matrix re-creation. Structured hyper-sparsity exploits redundancy in the storage-intensive matrices to minimize memory overhead. Improved initial estimate of node voltage angles reduces iterations and accelerates convergence.

🏆 Award Winning Solution

HyPowerFlow is the winner of Machine Learning for Physical Simulation Challenge - powergrid use case

🚀 How to Run

1. Clone the GitHub repository:

   git clone https://github.com/IRT-SystemX/ml4physim_startingkit_powergrid.git

2. Install the Python dependencies:

   pip install -r requirements.txt

3. Edit parameters.json:

   • Set data_download parameter (1 to download the data, 0 to skip)
   • Set train_batch_size and eval_batch_size as needed

4. Set Dataset-Specific Parameters in parameters.json:

   These parameters must be provided for each dataset before running:

   • base_volt — base voltages for buses
   • topo_vect_unique — unique topologies in the dataset
   • Ybus_sparse_const — base admittance matrix with no outages
   • PQ_unique — PQ vector for each topo_vect_unique
   • PV_unique — PV vector for each topo_vect_unique
   • bus_enable_flag
   • bus_renumber

5. Run the main script:

   python main.py

📄 License

This work is licensed under a Attribution-NonCommercial-NoDerivatives 4.0 International License .