Statistical surrogate modelling of a deterministic financial model for tidal energy generation
LE3 .A278 2017
Master of Science
Mathematics and Statistics
The Bay of Fundy is a region of the Atlantic Ocean located between the Canadian provinces of New Brunswick and Nova Scotia. It is known for having the highest tidal range on Earth, creating great potential for harvesting tidal energy. In order to make the harvesting of tidal energy economically successful, there are many variables that the decision makers within an organization will want to consider, such as how many turbines to deploy, which turbine design to use, and who to hire for installation and maintenance. A deterministic computer program that models several economic parameters of interest and uses over 50 input variables was developed by Roc, Funke,and Thyng (2015). The economic outputs are internal rate of return (IRR), levelized cost of energy (LCOE), and overall benefit (OB). Some of the variables that the deterministic computer program uses as inputs have values that are proprietary information that companies are not always willing to release, while others are not proprietary, but difficult to measure. A model that requires only the variables necessary to obtain an accurate prediction can be quite desirable. This research treats the deterministic financial model as within a black-box, and uses it as a basis for developing surrogate statistical models which approximate the deterministic computer program with a known degree of accuracy, and depend on fewer input variables. This can provide insight into whether there are combinations of only a few variables which determine the financial success of a tidal energy project in the Bay of Fundy. The benefit of this can be seen as three-fold: First, some of the variables that may have been difficult to obtain values for will no longer be required. Second, any variables which are shown to dominate the likelihood of the economic success of the project will give a project manager insight as to where they need to spend the most time exploring options. Third, results that contradict intuition can lend insight as to whether the deterministic computer program is accurately representing the effects of the variables. This black-box analysis reveals that by using only 15 of the original 51 input variables, surrogate models can be constructed that capture up to 79%, 98%, and 94% of the variability of IRR, LCOE, and OB, respectively.
The author grants permission to the University Librarian at Acadia University to reproduce, loan or distribute copies of my thesis in microform, paper or electronic formats on a non-profit basis. The author retains the copyright of the thesis.