Evaluating the potential of Gaussian process regression for data-driven renewable energy management

Lubbe, Foster Joachim (2019-12)

Thesis (MEng)--Stellenbosch University, 2019.


ENGLISH ABSTRACT: The variable nature of renewable energy sources presents a challenge to the stability of the electricity grid. The probabilistic modelling of renewable energydata could assist in overcoming this challenge. In this context, the machine learning technique of Gaussian process regression is explored. It is posited that Gaussian process regression could form part of a solution to the variability problem by allowing for better-informed energy management decisions. To test this hypothesis, Gaussian process regression is applied to wind speed and solar radiation datasets in order to regress and forecast its behaviour. Weather data is acquired from the Southern African Universities Radiometric Network and the forecasts are compared to the measured values. Meter malfunction is simulated within the solar radiation dataset in order to evaluate its effect on the regression. Gaussian process regression is furthermore used in an attempt to improve the quality of interval-deficient wind speed data. Attention is given to constructing a customised kernel for the modelling of renewable energy system data. The best regression and forecasting results for solar radiation data were obtained by employing a kernel consisting of an exponential sine-squared component and a rational quadratic component. No meaningful advantage was obtained by increasing the complexity of the kernel any further. It was also found that meter failure can be bridged by employing Gaussian process regression. The success achieved in modelling solar radiation data using Gaus- sian process regression is encouraging and could open up new avenues in the development of an effective renewable energy management system. No significant improvement in the quality of interval deficient wind speed data was observed after applying Gaussian process regression using the Matérn kernel. Python and Matlab were used for modelling and analysis.

AFRIKAANSE OPSOMMING: Die veranderlike aard van hernubare energiebronne kan die stabiliteit van die elektrisiteitsnetwerk ontwrig. Ten einde 'n oplossing te vind, word die probabilistiese masjienleertegniek van Gaussiese proses-regressie verken. Hierdie tegniek kan moontlik bydra tot 'n oplossing vir die stabiliteitsprobleem deur ingeligte energiebestuursbesluite in die hand te werk. Om hierdie hipotese te toets, word Gaussiese proses-regressie op windspoed- en sonstralingsdata toegepas om soedoende die gedrag daarvan te voorspel. Weerdata is van die South African Universities Radiometric Network verkry en voorspellings is met metings vergelyk. Die wanfunksionering van metingsinstrumentasie word in die sonstralingsdata gesimuleer om die effek daarvan op die regressiemodel vas te stel. Daar is ook gepoog om Gaussiese proses-regressie te gebruik om die kwaliteit van lae-resolusie-winddata te verbeter. Aandag word geskenk aan die samestelling van 'n kovariansiefunksie vir die modellering van hernubare energie data. Die beste regressie- en voorspellingsresultate is verkry deur gebruik te maak van 'n kovariansiefunksie bestaande uit 'n eksponensieel-sinus-kwadraat komponent en 'n rasioneel-kwadratiese komponent. Geen betekinsvolle voordeel is verky deur die kompleksiteit van die kovariansiefunksie verder te verhoog nie. Daar is ook bevind dat die wanfunksionering van metingsinstrumentasie oorbrug kan word deur Gaussiese proses-regressie. Die suksesvolle modellering van sonstralingsdata is belowend en kan nuwe kanale skep vir die ontwikkeling van 'n effektiewe hernubare-energie-bestuurstelsel. Die toepassing van Gaussiese proses-regressie deur gebruik te maak van die Matérn kovariansiefunksie het nie 'n noemenswaardige verbetering in die kwaliteit van windspoeddata teweeg gebring nie. Python en Matlab is gebruik vir modellering en analise.

Please refer to this item in SUNScholar by using the following persistent URL: http://hdl.handle.net/10019.1/107207
This item appears in the following collections: