@MASTERSTHESIS\{IMM2004-03067,
    author       = "R. K. Jensen",
    title        = "Robust alignment of wind prognosis",
    year         = "2004",
    keywords     = "Alignment, Optimisation, Expectation Maximization {EM} 
Alignment, Optimisation, Expectation Maximization {EM} Levenberg Maquard, Image analysis",
    school       = "Informatics and Mathematical Modelling, Technical University of Denmark, {DTU}",
    address      = "Richard Petersens Plads, Building 321, {DK-}2800 Kgs. Lyngby",
    type         = "",
    note         = "Supervised by Professor Henrik Madsen",
    url          = "http://www2.compute.dtu.dk/pubdb/pubs/3067-full.html",
    abstract     = "I denne afhandling foresl{\aa}s en robust fejljustering af prognoser for vind styrke, og denne justering unders{\o}ges.

I afhandlingen pr{\ae}senteres et studie af justering/alignment i to og tre dimensioner under anvendelse af robuste match metoder. De grundl{\ae}ggend teknikker og metoder gennemg{\aa}es, vurderes og diskuteres.

Justeringen studeres i to dimensioner via robuste match metoder. Herved drages fordel af en gradvis afgr{\ae}nsning af s{\o}geomr{\aa}det for det individuell punkt match i en sammenh{\ae}ng med {EM} - opdatering. Dette mindsker indflydelsen fra uregelm{\ae}ssigt placerede punkter.

Brugen af {EM} i to dimensioner studeres. Det unders{\o}ges hvordan man kan bruge en ny algoritme til justering af fejlen mellem vejrudsigtens forudsigels af vindstyrken og den faktiske produktion af vindenergi. Nyheden ligger at justere direkte p{\aa} vindstyrke data, betragtet som et billede, og at g{\o}r det robust ved at indlejre den i en {EM} opdatering.

Justeringens problem kan im{\o}deg{\aa}s ved hj{\ae}lp af velkendte line{\ae}re metode til at analysere problemets ikke line{\ae}re opf{\o}rsel. Justering af punktm{\ae}ngde og direkte p{\aa} pixel bliver udf{\o}rt med forskellige minimeringsmetoder. De vises, hvordan de foresl{\aa}ede metoder er anvendelige og robuste p{\aa} simuleringer der er t{\ae}t p{\aa} de virkelige data.

In English:

In this thesis a robust alignment of wind-speed prognosis is proposed and investigated. 

The thesis presents a study of alignment in two and three dimension through robust matching methods. The basic methods and techniques use in alignment are reviewed and discussed.

For alignment in two dimensions, the usage of robust alignment is examined. It takes advantage of gradually constraining the search range of the individual point match in the context of the {EM} - update. This lessens the influence of outliers.

It is investigated how to use a novel algorithm for aligning prognosis of wind-speed to observed energy produce at wind turbines. The novelty lie in the explicit minimization of wind-speed forecasts regarded as images and making it robust by embedding it in an {EM} context.

The alignment problem can be addressed using well known linear methods to analyse the non linear behaviour of the problem. Aligning sets of coordinates or directly on pixels is done using different minimization methods. It is shown how the proposed methods are applicable and robust of simulations very close to that of real data."
}