@MASTERSTHESIS\{IMM2007-05346,
    author       = "C. V. Hansen",
    title        = "Efficient Numerical Methods for Adaptive Quantile Regression",
    year         = "2007",
    school       = "Informatics and Mathematical Modelling, Technical University of Denmark, {DTU}",
    address      = "Richard Petersens Plads, Building 321, {DK-}2800 Kgs. Lyngby",
    type         = "",
    note         = "Supervised by Prof. Henrik Madsen, {IMM,} {DTU}.",
    url          = "http://www2.compute.dtu.dk/pubdb/pubs/5346-full.html",
    abstract     = "The efficiency and profitability of wind power relies heavily on having precise productivity forecasts in order to predict the need for other power sources.

A significant addition to power production forecasts are uncertainty predictions, which is the focus of this Master thesis. A Matlab implementation of an adaptive quantile regression programhas been converted to C for increased computational performance efficiency, and to break the grounds needed for it to become an integrated part of a commercial power prediction tool (WPPT).

The adaptive algorithm has been improved significantly by the introduction of
a penalty based selection and apart from the simplex implementation for quantile regression, the program has been extended with the interior point method and with a flexible framework for additional algorithms. With these additions to what the program is capable of, the computational performance has still improved significantly.

Particular focus has been placed on qualitatively validating the calculated quantiles in terms of reliability. With the aid of both reliability and skill score tests it has been shown that the quality of quantiles predicted benefits from frequent updates, but this can be only weekly, and not hourly, as previously expected."
}