Thermal Unit Commitment Instances for Paper: A Polyhedral--Based Approach Applied to Quadratic Cost Curves in the Unit Commitment Problem

Alejandro Angulo, Daniel Espinoza, Rodrigo Palma

The aim of the thermal unit commitment problem is to decide which of the available generating units in a power system will be used to cover the demand and reserve requirements at a minimum cost. The solution of the UC problem has to satisfy the operational constraints of the generating units, with typical planning horizons ranging from one day to a couple of weeks. The state–of–the–art approach to tackle the UC problem is based on monolithic mixed–integer programming (MLP) formulations, which are solved solved via a particular branch & bound (BB) scheme. For example, it has been reported by the PJM interconnection that the MIP–BB approach has been able to improve the UC problem's solution by about 1% when compared against strategies based on Lagrangian relaxations (LR). Every percent of improvement translates into millions of dollars of savings, thus the UC problem continues to be a topic of active research.

In order to evaluate the performance of different formulations of this problem, we have built a set of 162 daily instances based on code developed by Frangioni, which was modified to write the output files in AMPL format.

All instances include the following parameters: number of thermal generation units (NG), number of hourly periods (NP), unit's id (IDG), initial state (UINI), initial power generation (PINI), ramp-up limit (GRADS), ramp-down limit (GRADB), minimum up time (TMINON), number of hours on at init (TONINI), minimum down time (TMINOFF), number of hours off at init (TOFFINI), cuadratic coefficient of generation cost (A), linear coefficient of generation cost (B), fixed part of generation cost (C), minimum power (PMIN), maximum power (PMAX), shutdown cost (SDCOST), startup power (PSU), suthdown power (PSD), number of startup cost stages (NSUC), startup cost pairs (SUTIME,SUCOST), hourly demand (DEMANDA) and hourly reserve level (RESERVA).

Regarding the number of generators involved, instances were divided into three groups:
  • Small size instances: 57 cases with 10-20 units.
  • Medium size instances: 69 cases with 50-75-100 units.
  • Large size instances: 36 cases with 150-200 units.


Here you can download our instances in AMPL format.

   Type of instances 
Small size
Medium size
Large size