Piecewise linear approximation for MILP leveraging piecewise convexity to improve performance

TL;DR

This paper introduces a piecewise-convex approximation (PwCA) method for MILP that simplifies non-linear functions into convex constraints, significantly reducing solving time in large-scale engineering optimization problems.

Contribution

The paper presents a novel algorithm for piecewise-linear approximation of multivariate non-linear functions using convex sets, requiring only one auxiliary binary variable, enhancing MILP efficiency.

Findings

PwCA outperforms simplex approximation in solving time.

The method reduces auxiliary variables needed in MILP formulations.

Significantly faster solutions in large MILP problems like unit commitment.

Abstract

Mixed integer linear programming (MILP) has seen a sharp rise in use for engineering optimization applications in recent years. Even for initially non-linear problems, it is often the method of choice. Then, the non-linear functions have to be approximated in a way, that allows for an efficient implementation in MILP. To realize adaptive operation planning with MILP unit commitment, piecewise-linear approximations of the functions that describe the operating behavior of devices in the energy system have to be computed. We present an algorithm to compute a piecewise-linear approximation of a multi-variate non-linear function. The algorithm splits the domain into two regions and approximates each region with a set of hyperplanes that can be translated to a convex set of constraints in MILP. The main advantage of this "piecewise-convex approximation" (PwCA) compared to more general piecewise-linear approximation with simplices is that the MILP representation of PwCA requires only one auxiliary binary variable. For this reason, PwCA yields significantly faster solving times in large MILP problems where the MILP representation of certain functions has to be replicated many times, such as in unit commitment. To quantify the impact on solving time, we compare the performance using PwCA with the performance of simplex approximation with logarithmic formulation and show that PCA outperforms the latter by a big margin. For this reason, we conclude that PCA will be a useful tool to set up and solve large MILP problems such as arise in unit commitment and similar engineering optimization problems.