First Advisor

Martin Zwick

Term of Graduation

Winter 2023

Date of Publication

2-15-2023

Document Type

Dissertation

Degree Name

Doctor of Philosophy (Ph.D.) in Systems Science

Department

Systems Science

Language

English

Subjects

Graphical modeling (Statistics), Machine learning

DOI

10.15760/etd.8190

Physical Description

1 online resource (xi, 195 pages)

Abstract

This research focuses on the investigation of two machine learning methodologies, Reconstructability Analysis (RA) and Bayesian Networks (BN). Both methods are probabilistic graphical modeling (PGM) methodologies. RA was developed in the systems community and has applications including time-series analysis, classification, decomposition, compression, pattern recognition, prediction, control, and decision analysis. BNs have origins in path models and have applications similar to those of RA. BNs are another graphical modeling approach for data modeling that is closely related to RA; where BN overlaps RA the two methods are equivalent, but RA and BN each has distinctive features absent in the other methodology.

The primary aim of this research is to make theoretical contributions through the unification of the RA and BN methods by developing and integrating the RA and BN neutral and directed system lattices and developing an algorithm to generate the joint RA-BN neutral system lattice of structures for any number of variables. This analysis is done exhaustively for four variables, which is sufficient to elucidate the formal relationship between these two PGM approaches.

The secondary aim of this research is to apply RA and BN to a real world problem in the electricity industry to identify predictive variables and obtain a new stand-alone model that improves prediction accuracy and reduces the INC and DEC Resource Sufficiency Requirements for Western Energy Imbalance Market participants.

The primary research aim was addressed by developing a lattice of structures for RA and BN that offers an expanded set of models to represent complex systems more accurately or more simply. The conceptualization of this lattice also offers a framework for additional innovations. Specific contributions include integrating RA and BN by developing and visualizing: (1) a BN neutral system lattice of general and specific graphs, (2) a joint RA-BN neutral system lattice of general and specific graphs, (3) an augmented RA directed system lattice of prediction graphs, and (4) a BN directed system lattice of prediction graphs. Additionally, it (5) extends RA notation to encompass BN graphs and (6) offers an algorithm to search the joint RA-BN neutral system lattice to find the best representation of the structure of the system variables. All lattices are for four variables, but the theory and methodology presented are general and apply to any number of variables. These methodological innovations are contributions to machine learning and artificial intelligence and more generally to complex systems analysis. These innovations also suggest extensions of RA and BN modeling that could enhance their power and flexibility.

The secondary aim was addressed by applying RA and BN, as well as Neural Networks and Support Vector Regression, to build predictive models of Net Load Imbalance for the Resource Sufficiency Flexible Ramping Requirement in the Western Energy Imbalance Market. This research identified predictive variables wind forecast, sunrise/sunset and the hour of day as primary predictors of net load imbalance, among other variables, and show that the average size of the INC and DEC capacity requirements can be reduced by over 25% with the margin of error currently used in the industry while also significantly improving closeness and exceedance metrics. The reduction in INC and DEC capacity requirements would yield an approximate cost savings of $4 million annually for one of nineteen Western Energy Imbalance market participants. Reconstructability Analysis performed best among the machine learning methods tested.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

Persistent Identifier

https://archives.pdx.edu/ds/psu/39793

Download

Share

COinS