Dépôt numérique

Nonparametric versus parametric (both unimodal and mixed) probability distribution in hourly wind speed modelling for some regions of Tamil Nadu state in India.

Natarajan, Narayanan et Latif, Shahid (2024). Nonparametric versus parametric (both unimodal and mixed) probability distribution in hourly wind speed modelling for some regions of Tamil Nadu state in India. Stochastic Environmental Research and Risk Assessment , vol. 38 , nº 2. pp. 535-569. DOI: 10.1007/s00477-023-02587-x.

Ce document n'est pas hébergé sur EspaceINRS.


It is crucial to accurately predict the probability distribution of long-term wind speed patterns to evaluate the potential for wind energy. This could involve testing various probability density models to ensure they correctly match the wind speed (WS) characteristics provided. Parametric models have prior distributional assumptions that limit their flexibility and are unsuited for skewed, uni-modal, or multimodal wind regimes. Nonparametric kernel density estimation (KDE) is a data-driven model free from prior distribution assumptions. This study offers a thorough approach to modelling the probability density of hourly WS observations covering 11 locations in Tamil Nadu state in India. The efficacy of nonparametric Gaussian KDE with six bandwidth selectors was examined in modelling WS probability distribution. Additionally, both 1-component and 2-component mixture models are fitted to WS via the maximum likelihood estimation (MLE), the L-moment method (LMOM), and the expectation–maximization approach. The performance of some standard kernel functions, BOX, Epanecknokov, Triweight and Biweight, fitted with the direct-plug-in (DPI) method, are also compared. The model performance of all candidate models is examined thoroughly by employing different goodness-of-fit test measures. Investigation reveals that nonparametric KDE with an unbiased cross-validation approach outperformed all other nonparametric and parametric, uni- and bi-modal distributions for all the stations, except at Kanchipuram. The Gaussian KDE with Silverman rule-of-thumb best fits station Kanchipuram. Also, the best-fitted parametric model, among the 1-component model, outperformed the 2-component mixture models for all selected stations. When comparing the performance of some other kernel densities with DPI bandwidth selectors, it performed better than all parametric models. The hourly WS observation in this case study does not favour any fitted mixture models compared to nonparametric KDE density and 1-component density. Each station's selected model is employed further in estimating non-exceedance probabilities and return periods (RPs). Finally, the design WS quantiles are estimated at different univariate RPs (1, 2, 3, 5, 10, 15, 20, 30, 40, 50, 70, 80, 100 years) for all selected stations.

Type de document: Article
Mots-clés libres: hourly wind speed; 1-component parametric models; nonparametric models; mixture parametric models; gaussian kernel density estimation; goodness-of-fit; cumulative distribution function; return period
Centre: Centre Eau Terre Environnement
Date de dépôt: 01 mai 2024 14:30
Dernière modification: 01 mai 2024 14:30
URI: https://espace.inrs.ca/id/eprint/15478

Gestion Actions (Identification requise)

Modifier la notice Modifier la notice