A stage-wise surrogate modelling algorithm for tidal power simulators
LE3 .A278 2015
2015
Chipman, Hugh Ranjan, Pritam
Acadia University
Bachelor of Science
Honours
Mathematics and Statistics
Mathematics & Statistics
The Bay of Fundy has the highest tidal range in the world causing large amounts of water to flow at high speeds through the Minas Passage, a narrow passage in the Bay of Fundy. This feature gives the Minas Passage signifi cant potential to generate electrical power using in-stream tidal turbines. An important problem in generating this power is to determine the optimal locations of the turbines. There exist computer simulators able to predict the average power a turbine would generate at any location in the Minas Passage. These simulators update each time a turbine location is selected to reflect the effects of the turbine on the system. However, accurate simulators are computationally expensive and using them to determine the optimal turbine locations is impractical. Statistical models, such as Gaussian Process (GP) models, fit using a sample of simulator evaluations can estimate a surface of predicted simulator output. In a naive approach to surrogate modelling, a new model can be estimated over the full input space and its predicted surface optimized to select each turbine location. We propose a stage-wise surrogate modelling approach to determine the optimal sequential turbine locations with decreased computational cost and increased accuracy compared to the naive approach. Rather than generating a new full model after each turbine location is selected, the algorithm models the effects of the new turbine and updates the previous model. We also propose a new 2-stage GP model to estimate a more accurate fit of the initial power surface than the standard GP model, particularly in regions of high power. We implement the algorithm to select 50 turbine locations using a simplified computer simulator for a section of the Minas Passage. Our algorithm selects turbine locations which generate 92.15 MW of power, equal to approximately 91.8% of the power that could be extracted with the same number of turbine locations selected using the computer model.
The author retains copyright in this thesis. Any substantial copying or any other actions that exceed fair dealing or other exceptions in the Copyright Act require the permission of the author.
https://scholar.acadiau.ca/islandora/object/theses:1262