Show simple item record

dc.creatorChen, Xiaomei
dc.date.accessioned2021-11-11T21:03:21Z
dc.date.available2021-11-11T21:03:21Z
dc.date.created2021-05
dc.date.issued2021-05
dc.date.submittedMay 2021
dc.identifier.urihttps://hdl.handle.net/2346/88269
dc.description.abstractWind power has been increasingly integrated into electricity grid as renewable power supply through worldwide power system year by year. The intermittent and variable characteristics of wind energy will be pronounced in large-area wind power integrated power systems. When a large amount of wind power production changes, it would have significant impact on the stable operation of power systems because of the significant power imbalance introduced by large-area wind generation. Since current power system is more prepared to handle small variations of wind power, the above situation would pose a significant yet new challenge for power system operators to maintain the stability and reliability of system operations, schedule power planning, prepare non-spinning reserves, make unit commitment, and also manage economic market. Large wind power ramps are the especially critical events which are sudden and significant change in wind power generation in a relatively short time period. This makes large-area wind power ramp forecasting extremely important for power system operators. Since wind power ramps occurs very stochastically, it would be much useful and reliable to forecast them in a short time horizon. Normally, the wind speed data from local measurement site can be used for forecasting wind power ramps of individual wind turbine or wind farm. However, when forecasting large-area wind power ramps, there are multiple wind speed data available from different measurement sites covered over the corresponding geographical region which should be considered to predict more accurately. NWP (Numerical Weather Prediction) models can also be used for wind power ramp forecasting, whereas it is not accurate when predicting hourly and intra-hour wind power ramps because of its low refreshing rate of output data which typically updates only every 3 or 6 hours limited by their computational complexity and complicated postprocessing. This issue should be resolved to make it possible for NWPs to provide timely predictions of wind power ramps. This dissertation would be focused on addressing the above-mentioned issues, which would be predicting the large-scale wind power ramps in an extended region with hourly ahead forecast results. Three innovative methodologies have been proposed for the forecasting purpose in a timely manner. First, the ordinal levels of wind power ramp events have been defined as the forecasting information for power system operators. Multinomial logistic regression has been developed based on the discovery of the correlation of real-time meso-scale wind speed measurements from multiple Mesonet sites with regional wind power data. Further, the probabilistic output of individual multinomial logistic regressive models been combined to an aggregate model formed with different weights which are calculated by minimizing the Brier skill score (BSS) on the individual models. Based on the observation of the multivariate wind speed measurements from diverse locations in the extended region, those measurements are highly correlated. Thus, sparse primary component analysis (PCA) has been developed to utilize the above correlation information for further data fusion and feature extraction to improve the performance of forecasting models. Besides, ensemble NWP model has been developed with the weighted linear combination of several individual NWPs based on ensemble learning method. This method calculates the weights by minimizing the difference between forecast values and real-time wind speed measurements through gradient boosting algorithm. All the above developed models can be trained offline and carried out for online wind power ramp forecasting. The developed methods been tested and evaluated with real-world data using several metrics compared with other currently proposed methods. The results have shown the effectiveness and outperformance of these methods for improving wind power ramp event forecasting.
dc.description.abstractEmbargo status: Restricted until 06/2022. To request the author grant access, click on the PDF link to the left.
dc.format.mimetypeapplication/pdf
dc.language.isoeng
dc.subjectRegional Wind Power Ramp Forecasting
dc.subjectReal-Time Measurements
dc.subjectMultinomial Logistic Regression
dc.subjectSparse Primary Component Analysis (PCA)
dc.subjectNumerical Weather Prediction (NWP) Model
dc.subjectEnsemble Learning Method
dc.titleRegional wind power ramp forecasting using real-time Mesonet measurements
dc.typeThesis
dc.date.updated2021-11-11T21:03:22Z
dc.type.materialtext
thesis.degree.nameDoctor of Philosophy
thesis.degree.levelDoctoral
thesis.degree.disciplineElectrical Engineering
thesis.degree.grantorTexas Tech University
thesis.degree.departmentElectrical and Computer Engineering
dc.contributor.committeeMemberGiesselmann, Michael G.
dc.contributor.committeeMemberBayne, Stephen B.
dc.contributor.committeeMemberDu, Pengwei
dc.contributor.committeeChairHe, Miao
dc.rights.availabilityRestricted until June 2022.
dc.creator.orcid0000-0002-6748-3102
local.embargo.terms2022-05-01
local.embargo.lift2022-05-01


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record