Hybrid Learning Aided Inactive Constraints Filtering Algorithm to Enhance AC OPF Solution Time

TL;DR

This paper introduces a hybrid machine learning algorithm that efficiently predicts active and inactive constraints in AC optimal power flow, enabling faster solutions by reducing the problem size.

Contribution

A novel hybrid supervised learning approach that accurately identifies inactive constraints in AC OPF, leading to a truncated problem with reduced computational complexity.

Findings

High accuracy in active/inactive constraint prediction

Significant reduction in solution time for test systems

Open access code and dataset provided

Abstract

The Optimal power flow (OPF) problem contains many constraints. However, equality constraints along with a limited set of active inequality constraints encompass sufficient information to determine the feasible space of the problem. In this paper, a hybrid supervised regression and classification learning based algorithm is proposed to identify active and inactive sets of inequality constraints of AC OPF solely based on nodal power demand information. The proposed algorithm is structured using several classifiers and regression learners. The combination of classifiers with regression learners enhances the accuracy of active / inactive constraints identification procedure. The proposed algorithm modifies the OPF feasible space rather than a direct mapping of OPF results from demand. Inactive constraints are removed from the design space to construct a truncated AC OPF. This truncated optimization problem can be solved faster than the original problem with less computational resources. Numerical results on several test systems show the effectiveness of the proposed algorithm for predicting active and inactive constraints and constructing a truncated AC OPF. We have posted our code for all simulations on arxiv and have uploaded the data used in numerical studies to IEEE DataPort as an open access dataset.