Prediction interval modeling using Gaussian process quantile regression

In this thesis a methodology to construct prediction intervals for a generic black-box point forecast model is presented. The prediction intervals are learned from the forecasts of the black-box model and the actual realizations of the forecasted variable by using quantile regression on the observed prediction error distribution, the distribution of which is not assumed. An independent meta-model that runs in parallel to the original point forecast model is responsible for learning and generating the prediction intervals, thus requiring no modification to the original setup. This meta-model uses both the inputs and output of the black-box model and calculates a lower and an upper bound for each of its forecasts with the goal that a predefined percentage of future realizations are included in the interval formed by both bounds. Metrics for the performance of the meta-model are established, paying special attention to the conditional interval coverage with respect to both time and the inputs. A series of cases studies are performed to determine the capabilities of this approach and to compare it to standard practices.

Department dc:contributor.department

Massachusetts Institute of Technology. Engineering Systems Division.

Grantor dc:publisher

Massachusetts Institute of Technology

Year dc:date.issued

2014

Author dc:creator

• Aguilar Fargas, Joan

Advisor dc:contributor.advisor

• Moshe E. Ben-Akiva and Francisco C. Pereira.

• Engineering Systems Division.
• System Design and Management Program.

Statement dc:rights

• M.I.T. theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission. See provided URL for inquiries about permission.

Licence dc:rights.uri

• http://dspace.mit.edu/handle/1721.1/7582

Language dc:language.iso

eng

Handle dc:identifier.uri

http://hdl.handle.net/1721.1/100361

OAI identifier oai:identifier

oai:dspace.mit.edu:1721.1/100361