Detection and statistical modelling of output power ramp events for utility scale wind energy facilities

Date
2021-03
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: Large and rapid variationsin wind power,i.e. ramp events,hasreceived increased attention in recent years. It is generally accepted that accurate forecasting and quantification of ramp events are considered crucial in managing the risks associated with large-scale integration ofwind energy resources.The modelling and characterisation of wind power ramp events are of major importance in this context.The research objectives associated with thisproject focusses strongly on the investigation of existing ramp models and ramp detection algorithms, as well asthe development of improved ramp models and ramp detection algorithms.Some of the most commonly used ramp detection algorithms to date areinvestigated, including the swinging door algorithm, optimised swinging door algorithm and L1-ramp detect with sliding window.Three new ramp detection models were proposed andinvestigated, including a multi-parameter segmentation algorithm, a multi-parameter segmentation algorithm with particle swarm optimisationand regression-based segmentation algorithms. Furthermore,thewind power rampsdetectedby the multi-parameter segmentation algorithm for optimal parameter values were used to perform statistical analysis of the key ramp features in order to gain insightsinto wind power ramp events, including the distribution and severity of the ramp events, the frequency of occurrence and seasonality of the ramp events and the distribution of the interarrival times of ramps.The application of cluster analysis tothe ramp detection results obtained via the multi-parameter segmentation algorithmfor optimal parameter valueswas investigated to characterise awind energy facility site in terms of ramping mode.A diverse range of clustering algorithms, distance measures and linkage criteria wereinvestigated to determine the optimal clustering procedurefor the dataset of interest. The most important contribution of the work is the development of the multi-parameter segmentation algorithm. Focus was, therefore, placed on evaluating the performance of the multi-parameter segmentation algorithm by comparing its detection behaviour to that of the swinging door algorithm, optimised swinging door algorithm andtheL1-ramp detect with sliding windowfor optimal parameter values. It was concluded that the multi-parameter segmentation algorithmperforms significantly better compared to the original swinging door algorithm, as well as similarly or better compared to the L1-ramp detect with sliding windowand optimised swinging door algorithm, while also being more computationally inexpensive.The ramp detection performance of the multi-parameter segmentationalgorithmisespeciallysuperior to the ramp detection performance of the swinging door algorithm, optimised swinging door algorithmand L1-ramp detectwith sliding windowbased on the detection accuracy of the start-and end-points of the ramps, as itcorrectly identifies the start-and end-points of all the detected ramp events.
AFRIKAANSE OPSOMMING: Onlangs was meer aandag geskenk aan groot en vinnige veranderinge in windkrag, naamlik helling-gebeurtenisse. Die akkurate voorspelling en kwantifisering van helling-gebeurtenisse word as noodsaaklik beskou ten einde die risiko's verbonde aan grootskaalse integrasie van wind energie bronne te bestuur. Die modellering en karakterisering van helling-gebeurtenisse is dus van groot belang in hierdie konteks.Die doelwitte van die projek fokus grootliks om die bestaande hellingsmodelle en algoritmes wat helling-gebeurenenisse identifiseer te ondersoek, asook omnuwe hellingsmodelle en algoritmes te ontwikkel wat beter isas die bestaande modelle en algoritmes.Die mees algemeenste algoritmes wat helling-gebeurtenisse identifiseer, naamlik die swaai deur algoritme, die geoptimaliseerde swaai deur algoritme endie“L1-ramp detect with sliding window” was ondersoek. Drie nuwe hellingsmodelle was voorgestel en ondersoek, naamlik die multi-parameter segmenterings algoritme, die multi-parameter segmenterings algoritme met optimering van deeltjieswerm en segmenterings algoritmes wat op regressie gebaseer is. Die windkrag helling-gebeurtnenisse wat deur die multi-parameter segmenterings algoritme geidentifiseer is vir optimale waardes van die parameters, word gebruik om ‘n statistiese analise van die hoof eienskappe van diehelling-gebeurtenisse uit te voer. Die doel daarvan is om insigte oor die windkrag helling-gebeurtense te kry, insluitend die verspreiding en erns van die helling-gebeurtenisse, die seisoengerigtheidvan die helling-gebeurtenisse en hoe gereeld dit voorkom, asook die verspreiding van die tydsduur tussen die helling-gebeurtenisse.Die toepassing van groeperings-analise word ondersoek op die helling-gebeurtenisse wat deur die multi-parameter segmenterings algoritme geidentifiseer is, ten einde ‘n wind energy aanleg te karakteriseer in terme van ‘n hellingsmodus. Verskeie groeperings algoritmes, afstand maatstaweenkoppelings maatstawe was ondersoek om te bepaal wat die optimale groepering prosedure is vir die spesifieke datastel is. Die belangrikste bydrae van die projek is die ontwikkeling van die multi-parameter segmenterings algoritme. Daar worddus grootliks gefokus op die evaluering van die prestasie van die multi-parameter segmenteringsalgoritme deur dit met die swaai deur algoritme, die geoptimaliseerde swaai deur algoritme en die “L1-ramp detect with sliding window”te vergelyk. Die gevolgtrekking is dat die multi-parameter segmenterings algoritme aansienlik beter presteer as die oorspronklike swaai deur algoritme, asook soortgelyk of beter presteer in vergelyking met die “L1-ramp detect with sliding window”en geoptimaliseerde swaai deur algoritme, terwyl die loop-tyd ook aansienlik vinniger is.
Description
Thesis (MEng)--Stellenbosch University, 2021.
Keywords
Statistical modelling, UCTD, Wind power, Ramp detection, Renewable energy sources
Citation