ew 18: e33

Research Article

Modelling state spaces and discrete control using MILP: computational cost considerations for demand response

Download30 downloads
  • @ARTICLE{10.4108/eai.23-12-2020.167787,
        author={P. L. Magalh\"{a}es and C. H. Antunes},
        title={Modelling state spaces and discrete control using MILP: computational cost considerations for demand response},
        journal={EAI Endorsed Transactions on Energy Web: Online First},
        volume={},
        number={},
        publisher={EAI},
        journal_a={EW},
        year={2020},
        month={12},
        keywords={computational performance, state space, discrete control, mixed-integer linear programming, multiple-choice programming},
        doi={10.4108/eai.23-12-2020.167787}
    }
    
  • P. L. Magalhães
    C. H. Antunes
    Year: 2020
    Modelling state spaces and discrete control using MILP: computational cost considerations for demand response
    EW
    EAI
    DOI: 10.4108/eai.23-12-2020.167787
P. L. Magalhães1,*, C. H. Antunes2
  • 1: INESC Coimbra, Department of Electrical and Computer Engineering, Rua Sílvio Lima, Pólo II, 3030-290 Coimbra, Portugal
  • 2: University of Coimbra, Department of Electrical and Computer Engineering, Rua Sílvio Lima, Pólo II, 3030-290 Coimbra, Portugal
*Contact email: pmlpm@deec.uc.pt

Abstract

INTRODUCTION: Demand response (DR) has been proposed as a mechanism to induce electricity cost reductions and is typically assumed to require the adoption of time-differentiated electricity prices. Making the most of these requires using automated energy management systems to produce optimised DR plans. Mixed-integer linear programming (MILP) has been used for this purpose, including by modelling dynamic systems (DS).

OBJECTIVES: In this paper, wecompare the computational performance of MILP approaches for modelling state spaces and multi-level discrete control (MLDC) in DR problems involving DSs.

METHODS: A state-of-the-art MILP solver was used to compute solutions and compare approaches.

RESULTS: Modelling state spaces using decision variables proved to be the most efficient option in over 80% of cases. In turn, the new MLDC approaches outperformed the standard one in about 60% of cases despite performing in the same range.

CONCLUSION: We conclude that using state variables is generally the better option and that all MLDC variants perform similarly.