Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.785726
Title: Probabilistic modelling of oil rig drilling operations for business decision support : a real world application of Bayesian networks and computational intelligence
Author: Fournier, François A.
Awarding Body: Robert Gordon University
Current Institution: Robert Gordon University
Date of Award: 2013
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
This work investigates the use of evolved Bayesian networks learning algorithms based on computational intelligence meta-heuristic algorithms. These algorithms are applied to a new domain provided by the exclusive data, available to this project from an industry partnership with ODS-Petrodata, a business intelligence company in Aberdeen, Scotland. This research proposes statistical models that serve as a foundation for building a novel operational tool for forecasting the performance of rig drilling operations. A prototype for a tool able to forecast the future performance of a drilling operation is created using the obtained data, the statistical model and the experts' domain knowledge. This work makes the following contributions: applying K2GA and Bayesian networks to a real-world industry problem; developing a well-performing and adaptive solution to forecast oil drilling rig performance; using the knowledge of industry experts to guide the creation of competitive models; creating models able to forecast oil drilling rig performance consistently with nearly 80% forecast accuracy, using either logistic regression or Bayesian network learning using genetic algorithms; introducing the node juxtaposition analysis graph, which allows the visualisation of the frequency of nodes links appearing in a set of orderings, thereby providing new insights when analysing node ordering landscapes; exploring the correlation factors between model score and model predictive accuracy, and showing that the model score does not correlate with the predictive accuracy of the model; exploring a method for feature selection using multiple algorithms and drastically reducing the modelling time by multiple factors; proposing new fixed structure Bayesian network learning algorithms for node ordering search-space exploration. Finally, this work proposes real-world applications for the models based on current industry needs, such as recommender systems, an oil drilling rig selection tool, a user-ready rig performance forecasting software and rig scheduling tools.
Supervisor: McCall, John ; Petrovski, Andrei ; Barclay, Peter Sponsor: Innovate UK
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.785726  DOI: Not available
Keywords: Business intelligence ; Oil and gas industry ; Bayesian networks ; Metaheuristics
Share: