Data-driven methods for modeling of extreme events
Abstract/Contents
- Abstract
- The motivation of this thesis is to model the risk of extreme events. With rapidly increasing amounts of available data, the statistics on extreme events is more accurate. The developed methodology in this thesis work extends existing actuarial science and financial risk approaches. These approaches use a branch of statistics known as extreme value theory (EVT). One of the extensions is analyzing risk trends across years for each calendar month. This is a highly non-trivial problem because extreme events might not happen every year for a given month. A Bayesian maximum a posteriori (MAP) formulation is applied to tackle this problem. One of the main applications of this novel method is characterizing the risk trends of 100 year extreme high temperature events. We focus on these extreme weather events since their occurrences are immensely disruptive and damaging. The non-triviality of the problem is in distinguishing climate change from the immense variations in weather, both spatially and temporally. The data analyzed is from the continental U.S. from 1979 to 2015. We show that aggregation from multiple locations is valid in order to get more accurate statistics on 100 year extremes. There is a factor of 2.7 relative increase for 100-year extreme high temperature event risk that is found in the last 4 decades of the climate data. This is primarily caused by both a factor 2.3 relative increase in the number of and a 41\% increase in magnitude of extreme high temperature events over the last 37 years. Another method is also developed for building a non-parametric stochastic model of the distribution from large data sets. This is done with an modified version of quantile regression (QR). This new QR formulation also tackles two major issues with the original QR method itself. In the second main application example, we consider probabilistic forecasting of the loads in electrical power grid. A probabilistic forecast is given by our estimated QR model. The probabilistic evaluation of risk is important because load volatility is increasing with the on-going proliferation of the renewable generation. The most important risk factor is that the required amount of the electricity cannot be procured at the spot market, because the stand-by generation capacity is insufficient. An application to electric utility data shows significant improvement in load balancing risk compared to existing methods.
Description
| Type of resource | text |
|---|---|
| Form | electronic; electronic resource; remote |
| Extent | 1 online resource. |
| Publication date | 2016 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Associated with | Shenoy, Saahil |
|---|---|
| Associated with | Stanford University, Department of Physics. |
| Primary advisor | Chu, Steven |
| Primary advisor | Gorinevskiĭ, D. M |
| Thesis advisor | Chu, Steven |
| Thesis advisor | Gorinevskiĭ, D. M |
| Thesis advisor | Boyd, Stephen P |
| Advisor | Boyd, Stephen P |
Subjects
| Genre | Theses |
|---|
Bibliographic information
| Statement of responsibility | Saahil Shenoy. |
|---|---|
| Note | Submitted to the Department of Physics. |
| Thesis | Thesis (Ph.D.)--Stanford University, 2016. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2016 by Saahil Subrao Shenoy
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...