A Learning Method for Optimization with Hard Constraints

By Priya L. Donti et al

Published on April 25, 2021

Read the original document by opening this link in a new tab.

Table of Contents

1. Introduction
2. Related Work
3. DC3: Deep Constraint Completion and Correction
4. Experiments

Summary

The document discusses the DC3 algorithm, which addresses optimization problems with hard constraints using deep learning. It introduces methods for enforcing equality and inequality constraints during training and testing, ensuring feasibility of solutions. The experiments evaluate DC3 on convex quadratic programs and the real-world task of AC optimal power flow, demonstrating its effectiveness in achieving near-optimal objective values while maintaining feasibility. Comparison with traditional optimization solvers shows promising results in terms of optimality, feasibility, and speed.