Probabilistic streamflow forecasts in hydropower systems operation

Authors: Osina Torrez, Michael
Advisor: Tilmant, Amaury; Anctil, François
Abstract: In Canada, like in many OECD (Organization for Economic Co-operation and Development) countries, the more efficient use of existing hydropower assets is becoming increasingly relevant.The optimal operation of a hydroelectric system is a sequential decision making problem. A sequence of release decisions must be determined over a given planning period taking into account a variety of physical and ecological constraints. Since this planning period may extend over a more or less distant future, release decisions are influenced by the availability of reliable hydrologic forecasts, including hydrological ensemble prediction systems (H-EPS).Hydrologists often rely on statistical scores to assess the reliability and accuracy of H-EPS,but those scores do not give any indication of the economic value of the forecasts. This studyseeks to identify the most relevant attributes of ensemble hydrological forecasts in hydropower production. To do this, a large set of forecasts is built from 20 hydrological models and ensemble meteorological forecasts of 50 members over a period of 6 years (2011-2016). From this large set, several H-EPS are then produced (configured) and used by a hydroelectric optimization model. The management of the water system is then simulated on a rolling horizon over a period of 6 years (2011-2016). The simulation results indicate that there is a trend between the overall quality and the value of the forecast in terms of energy production,but that this relationship is not directly proportional (1: 1). The multi-model setup works a bit better than the other setups. In addition, the simulation results show that the ensemble forecast at short-term (ST) has value, but the room for improvement is clearly in the forecastat mid-term (MT, seasonal), as a large reservoir upstream controls the availability of water throughout the system. In addition, probabilistic forecasts give better performance than determinists, because they provide information on the uncertainty of the optimization model.Finally, ST forecasts have value while ST-MT optimization models are coupled.
Document Type: Mémoire de maîtrise
Issue Date: 2021
Open Access Date: 11 October 2021
Permalink: http://hdl.handle.net/20.500.11794/70593
Grantor: Université Laval
Collection:Thèses et mémoires

Files in this item:
Description SizeFormat 
37495.pdf12.02 MBAdobe PDFThumbnail
View/Open
All documents in CorpusUL are protected by Copyright Act of Canada.