A multi-objective dynamic programming-based metaheuristic to solve a bi-objective unit commitment problem using a multi-objective decoder. (2016)

• Record Type: Journal Article
• Authors: Jacquin, Sophie; Jourdan, Laetitia; Talbi, El-Ghazali
• Is Part Of: International journal of metaheuristics. Volume 5:Number 1(2016)
• Page Start: 3
• Page End: 30
• Publication Date: 2016
• Subjects: hybrid optimisation -- indirect representation -- multi-objective decoder -- multi-objective dynamic programming -- multi-objective evolutionary algorithms -- unit commitment problem -- multi-objective UCP -- constrained scheduling -- metaheuristics -- production costs -- greenhouse gases -- GHG emissions; Heuristic algorithms -- Periodicals; 006.3105
• Languages: English
• ISSNs: 1755-2176

Abstract

The unit commitment problem (UCP) is a heavily constrained scheduling problem, where the on/off scheduling and production amounts of heterogeneous power production units have to be determined for a discrete time horizon. Due to environmental concerns, the traditional UCP to solely minimise the production cost is no longer adequate, and a second objective, to minimise the gas emissions has to be added to properly model reality. In this paper, we propose an efficient metaheuristic to solve this multi-objective version of the UCP. The proposed method, MO-DYNAMOP, is a generalisation of DYNAMOP (DYNAmic programming-based Metaheuristic for Optimisation Problems), a state-of-the-art hybrid optimiser which was successfully applied to the single-objective unit commitment problem. The main difficulty in extending DYNAMOP to the multi-objective UCP is that it uses an indirect representation of solution that gives the on/off scheduling of each unit. The real production amounts are computed by an exact sub-optimiser which minimises the production cost assuming that the on/off scheduling is fixed. Since the sub-optimiser now has to solve a multi-objective problem, each on/off scheduling induces an entire Pareto optimal set of solutions. We handle this complication by assigning an approximation of the corresponding set to each on/off scheduling solution. A comparison study with methods previously proposed in the literature indicates that MO-DYNAMOP performs considerably better on many benchmark instances.