# Learning Warm-Start Points for AC Optimal Power Flow

## TL;DR

This paper proposes a machine learning approach using Random Forests to generate warm-start solutions for AC optimal power flow problems, aiming to improve solution speed and efficiency.

## Contribution

It introduces a multi-target learning framework that predicts approximate ACOPF solutions directly from network loads without needing detailed system parameters.

## Key findings

- Learned solutions can serve as effective warm starts for ACOPF solvers.
- The approach shows variable benefits depending on the network and solver used.
- Numerical tests on IEEE networks demonstrate promising results.

## Abstract

A large amount of data has been generated by grid operators solving AC optimal power flow (ACOPF) throughout the years, and we explore how leveraging this data can be used to help solve future ACOPF problems. We use this data to train a Random Forest to predict solutions of future ACOPF problems. To preserve correlations and relationships between predicted variables, we utilize a multi-target approach to learn approximate voltage and generation solutions to ACOPF problems directly by only using network loads, without the knowledge of other network parameters or the system topology. We explore the benefits of using the learned solution as an intelligent warm start point for solving the ACOPF, and the proposed framework is evaluated numerically using multiple IEEE test networks. The benefit of using learned ACOPF solutions is shown to be solver and network dependent, but shows promise for quickly finding approximate solutions to the ACOPF problem.