@MASTERSTHESIS\{IMM2003-02546,
    author       = "T. I. Halldorsson",
    title        = "Unit commitment for combined heat and power production - Using simulated annealing",
    year         = "2003",
    keywords     = "Unit commitment, economic dispatch, heuristic, combinatorial optimization, Simulated Annealing, combined heat and power, optimization",
    school       = "Informatics and Mathematical Modelling, Technical University of Denmark, {DTU}",
    address      = "Richard Petersens Plads, Building 321, {DK-}2800 Kgs. Lyngby",
    type         = "",
    note         = "Supervisor: Victor Vidal",
    url          = "http://www2.compute.dtu.dk/pubdb/pubs/2546-full.html",
    abstract     = "In this project, Simulated Annealing is used to solve the unit commitment problem (UCP) for combined heat and power production system. The {UCP} involves scheduling the on/off status of units in a power production system in order to meet load demand while satisfying operational constraints. The {UCP} is a combinatorial optimization problem. A sub-problem of the {UCP} is the economic dispatch problem (EDP) which determines the production output for a given unit schedule. In this project the {EDP} is formulated as a linear programming problem. 

A complicating factor in solving the {EDP} is that the power production system is composed of 4 different unit types were heat and power production is dependent and the number of constraints is large. The solution approach used, was to implement a fast heuristic method that solves the {EDP}. The heuristic was tested in comparison to an optimal solver and was found to produce near optimal solutions in all cases. 

Simulated Annealing (SA) is used to solve the unit scheduling. A method that generates random feasible solutions for {SA} was implemented. {SA} was then tested on diŁerent sized problems and the results compared to optimal or lower bound solutions calculated in Gams . Near optimal solutions were obtained with {SA} and the computation time was considerably faster for {SA} than Gams . Finally, the performance of {SA} was compared to the performance of Tabu Search implemented by a fellow student working on the same project."
}