Multiple-Periods Locally-Facet-Based MIP Formulations for the Unit Commitment Problem

TL;DR

This paper introduces a systematic approach to formulate tighter multi-period mixed integer programming models for the unit commitment problem, significantly improving computational efficiency for large-scale systems.

Contribution

It develops a multi-period formulation based on sliding windows with explicit derivation of constraints, enhancing solution tightness and computational performance.

Findings

Proposed models are tighter than existing models for window sizes greater than 2.

The approach effectively reduces the search space and computational time.

Validated on systems with up to 1080 generators, showing improved performance.

Abstract

The thermal unit commitment (UC) problem has historically been formulated as a mixed integer quadratic programming (MIQP), which is difficult to solve efficiently, especially for large-scale systems. The tighter characteristic reduces the search space, therefore, as a natural consequence, significantly reduces the computational burden. In literatures, many tightened formulations for a single unit with parts of constraints were reported without presenting explicitly how they were derived. In this paper, a systematic approach is developed to formulate tight formulations. The idea is to use more binary variables to represent the state of the unit so as to obtain the tightest upper bound of power generation limits and ramping constraints for a single unit. In this way, we propose a multi-period formulation based on sliding windows which may have different sizes for each unit in the system. Furthermore, a multi-period model taking historical status into consideration is obtained. Besides, sufficient and necessary conditions for the facets of single-unit constraints polytope are provided and redundant inequalities are eliminated. The proposed models and three other state-of-the-art models are tested on 73 instances with a scheduling time of 24 hours. The number of generators in the test systems ranges from 10 to 1080. The simulation results show that our proposed multi-period formulations are tighter than the other three state-of-the-art models when the window size of the multi-period formulation is greater than 2.