A machine learning approach to solve the DC optimal power flow problem

Tools

He, Runsheng and Bukhsh, Waqquas and Cao, Shengming and Stephen, Bruce; (2025) A machine learning approach to solve the DC optimal power flow problem. In: 2025 10th IEEE Workshop on the Electronic Grid (eGRID). 2025 10th IEEE Workshop on the Electronic Grid (eGRID) . IEEE, GBR. ISBN 979-8-3315-9364-3

[thumbnail of He-etal-2025-A-machine-learning-approach-to-solve-the-DC-optimal]
Preview
 Text. Filename: He-etal-2025-A-machine-learning-approach-to-solve-the-DC-optimal.pdf
Accepted Author Manuscript
License: Creative Commons Attribution 4.0 logo
Download (1MB)| Preview

Abstract

This paper proposes a two-fold data-driven framework for efficiently solving the DC Optimal Power Flow (DCOPF). The Monte Carlo sampling is applied as a strategy with a small set of random perturbations to generator cost coefficients, which eliminates solution degeneracy and yields a continuous, unique mapping from nodal demands to optimal dispatch. We train a simple multilayer perceptron (MLP) on raw–scale variables with generator–output clamping and a transformer–aware PTDF flow–penalty to respect physical constraints. On the IEEE- 24 benchmark, our model achieves 0.33% cost-gap MAPE, 0.34% power-balance gap, and 0.70% overload rate. And we found the limitations of MLP model prediction in the prediction of multiple models when discovering larger grid predictions.

ORCID iDs

He, Runsheng ORCID logoORCID: https://orcid.org/0009-0008-7974-4788, Bukhsh, Waqquas ORCID logoORCID: https://orcid.org/0000-0002-5765-0747, Cao, Shengming ORCID logoORCID: https://orcid.org/0009-0000-6623-0843 and Stephen, Bruce ORCID logoORCID: https://orcid.org/0000-0001-7502-8129;
CORE (COnnecting REpositories)