{"id":{"repo_id":"strathclyde","oai_identifier":"oai:strathclyde:ht24wj46b"},"canonical_url":"https://search.dev.ndltd.org/etd/strathclyde/oai:strathclyde:ht24wj46b","repository":{"repo_id":"strathclyde","name":"University of Strathclyde","base_url":"https://stax.strath.ac.uk/catalog/oai"},"display":{"title":"Wind turbine dynamics identification using gaussian process machine learning","abstract":"Wind turbine controllers require dynamic information about the turbine for design and operation purposes. These dynamics are currently determined from simulation models during the turbine design stages. Hence, the dynamics for a given operational turbine will not be identical to those assumed by the controller due to manufacturing and construction variations. Furthermore, turbine dynamics are known to change over time due to environmental effects such as blade erosion. There are currently no known methods by which such information can be determined for an operational turbine. This thesis presents such a method.;The determination of sought dynamic information is formulated as a regression problem involving data available to a wind turbine controller. The nature of the dynamics identification problem is shown to necessitate a regression method which is able to process data in batches, updating predictions as new data becomes available.;Gaussian process machine learning is chosen as the regression approach best suited for application in this problem. However, a review of existing batched Gaussian process theory results in the identification of gaps in the current knowledge base which render existing methods unsuitable. A new approach to batched Gaussian process regression is therefore developed, Sufficient-Subset Gaussian process iteration, which addresses the questions for which existing theories come up short. In the process of developing this new method fundamental contributions have been made to the areas of Gaussian process polynomial regression and sparse Gaussian process approximation theory.;Sufficient-Subset Gaussian process iteration is applied to both simulated and real turbine data and shown to be able to identify the sought dynamics to within a 3% error threshold. Additionally, a related regression formulation corresponding to maximum efficiency tracking is shown to present a potential method for turbine monitoring and fault detection.","abstract_html":"Wind turbine controllers require dynamic information about the turbine for design and operation purposes. These dynamics are currently determined from simulation models during the turbine design stages. Hence, the dynamics for a given operational turbine will not be identical to those assumed by the controller due to manufacturing and construction variations. Furthermore, turbine dynamics are known to change over time due to environmental effects such as blade erosion. There are currently no known methods by which such information can be determined for an operational turbine. This thesis presents such a method.;The determination of sought dynamic information is formulated as a regression problem involving data available to a wind turbine controller. The nature of the dynamics identification problem is shown to necessitate a regression method which is able to process data in batches, updating predictions as new data becomes available.;Gaussian process machine learning is chosen as the regression approach best suited for application in this problem. However, a review of existing batched Gaussian process theory results in the identification of gaps in the current knowledge base which render existing methods unsuitable. A new approach to batched Gaussian process regression is therefore developed, Sufficient-Subset Gaussian process iteration, which addresses the questions for which existing theories come up short. In the process of developing this new method fundamental contributions have been made to the areas of Gaussian process polynomial regression and sparse Gaussian process approximation theory.;Sufficient-Subset Gaussian process iteration is applied to both simulated and real turbine data and shown to be able to identify the sought dynamics to within a 3% error threshold. Additionally, a related regression formulation corresponding to maximum efficiency tracking is shown to present a potential method for turbine monitoring and fault detection.","abstract_has_math":false,"creators":["Hart, Edward"],"institution":"University of Strathclyde","degree_name":"phd","degree_level":"doctoral-pg","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":["Leithead, Bill","Feuchtwang, Julian"],"committee_chairs":[],"committee_members":[],"year":2018,"date_issued":"2018","date_published":"2018","updated_at":"2026-07-24T04:47:42Z","subjects":[],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier.doi","label":"DOI","values":["10.48730/80p3-j963"],"render_values":[{"text":"10.48730/80p3-j963","href":"https://doi.org/10.48730/80p3-j963","code":true}]},{"key":"dc:identifier","label":"Identifier","values":["T15071"],"render_values":[{"text":"T15071","href":null,"code":true}]},{"key":"dc:creator.authoridentifier","label":"Author Identifier","values":["201458655"],"render_values":[{"text":"201458655","href":null,"code":true}]}]},"links":{"outbound_url":"https://stax.strath.ac.uk/concern/theses/ht24wj46b","outbound_label":"Repository record","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Leithead, Bill","Feuchtwang, Julian"]},{"key":"dc:creator","label":"Author","values":["Hart, Edward"]},{"key":"dc:creator.authoridentifier","label":"Author Identifier","values":["201458655"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2018"]},{"key":"dc:date.issued","label":"Date","values":["2018"]},{"key":"dc:publisher.department","label":"Dc Publisher Department","values":["Department of Electronic and Electrical Engineering","Centre for Doctoral Training in Wind and Marine Energy Systems"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Strathclyde","University of Oxford","University of Edinburgh"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["doctoral-pg"]},{"key":"dc:type.qualificationname","label":"Dc Type Qualificationname","values":["phd"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["T15071"]},{"key":"dc:identifier.doi","label":"DOI","values":["10.48730/80p3-j963"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://stax.strath.ac.uk/concern/theses/ht24wj46b"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Wind turbine controllers require dynamic information about the turbine for design and operation purposes. These dynamics are currently determined from simulation models during the turbine design stages. Hence, the dynamics for a given operational turbine will not be identical to those assumed by the controller due to manufacturing and construction variations. Furthermore, turbine dynamics are known to change over time due to environmental effects such as blade erosion. There are currently no known methods by which such information can be determined for an operational turbine. This thesis presents such a method.;The determination of sought dynamic information is formulated as a regression problem involving data available to a wind turbine controller. The nature of the dynamics identification problem is shown to necessitate a regression method which is able to process data in batches, updating predictions as new data becomes available.;Gaussian process machine learning is chosen as the regression approach best suited for application in this problem. However, a review of existing batched Gaussian process theory results in the identification of gaps in the current knowledge base which render existing methods unsuitable. A new approach to batched Gaussian process regression is therefore developed, Sufficient-Subset Gaussian process iteration, which addresses the questions for which existing theories come up short. In the process of developing this new method fundamental contributions have been made to the areas of Gaussian process polynomial regression and sparse Gaussian process approximation theory.;Sufficient-Subset Gaussian process iteration is applied to both simulated and real turbine data and shown to be able to identify the sought dynamics to within a 3% error threshold. Additionally, a related regression formulation corresponding to maximum efficiency tracking is shown to present a potential method for turbine monitoring and fault detection."]},{"key":"dc:description.abstract","label":"Abstract","values":["Wind turbine controllers require dynamic information about the turbine for design and operation purposes. These dynamics are currently determined from simulation models during the turbine design stages. Hence, the dynamics for a given operational turbine will not be identical to those assumed by the controller due to manufacturing and construction variations. Furthermore, turbine dynamics are known to change over time due to environmental effects such as blade erosion. There are currently no known methods by which such information can be determined for an operational turbine. This thesis presents such a method.;The determination of sought dynamic information is formulated as a regression problem involving data available to a wind turbine controller. The nature of the dynamics identification problem is shown to necessitate a regression method which is able to process data in batches, updating predictions as new data becomes available.;Gaussian process machine learning is chosen as the regression approach best suited for application in this problem. However, a review of existing batched Gaussian process theory results in the identification of gaps in the current knowledge base which render existing methods unsuitable. A new approach to batched Gaussian process regression is therefore developed, Sufficient-Subset Gaussian process iteration, which addresses the questions for which existing theories come up short. In the process of developing this new method fundamental contributions have been made to the areas of Gaussian process polynomial regression and sparse Gaussian process approximation theory.;Sufficient-Subset Gaussian process iteration is applied to both simulated and real turbine data and shown to be able to identify the sought dynamics to within a 3% error threshold. Additionally, a related regression formulation corresponding to maximum efficiency tracking is shown to present a potential method for turbine monitoring and fault detection."]},{"key":"dc:title","label":"Title","values":["Wind turbine dynamics identification using gaussian process machine learning"]}]}],"canonical_facts":{"dc:contributor.advisor":["Leithead, Bill","Feuchtwang, Julian"],"dc:creator":["Hart, Edward"],"dc:creator.authoridentifier":["201458655"],"dc:date":["2018"],"dc:date.issued":["2018"],"dc:description":["Wind turbine controllers require dynamic information about the turbine for design and operation purposes. These dynamics are currently determined from simulation models during the turbine design stages. Hence, the dynamics for a given operational turbine will not be identical to those assumed by the controller due to manufacturing and construction variations. Furthermore, turbine dynamics are known to change over time due to environmental effects such as blade erosion. There are currently no known methods by which such information can be determined for an operational turbine. This thesis presents such a method.;The determination of sought dynamic information is formulated as a regression problem involving data available to a wind turbine controller. The nature of the dynamics identification problem is shown to necessitate a regression method which is able to process data in batches, updating predictions as new data becomes available.;Gaussian process machine learning is chosen as the regression approach best suited for application in this problem. However, a review of existing batched Gaussian process theory results in the identification of gaps in the current knowledge base which render existing methods unsuitable. A new approach to batched Gaussian process regression is therefore developed, Sufficient-Subset Gaussian process iteration, which addresses the questions for which existing theories come up short. In the process of developing this new method fundamental contributions have been made to the areas of Gaussian process polynomial regression and sparse Gaussian process approximation theory.;Sufficient-Subset Gaussian process iteration is applied to both simulated and real turbine data and shown to be able to identify the sought dynamics to within a 3% error threshold. Additionally, a related regression formulation corresponding to maximum efficiency tracking is shown to present a potential method for turbine monitoring and fault detection."],"dc:description.abstract":["Wind turbine controllers require dynamic information about the turbine for design and operation purposes. These dynamics are currently determined from simulation models during the turbine design stages. Hence, the dynamics for a given operational turbine will not be identical to those assumed by the controller due to manufacturing and construction variations. Furthermore, turbine dynamics are known to change over time due to environmental effects such as blade erosion. There are currently no known methods by which such information can be determined for an operational turbine. This thesis presents such a method.;The determination of sought dynamic information is formulated as a regression problem involving data available to a wind turbine controller. The nature of the dynamics identification problem is shown to necessitate a regression method which is able to process data in batches, updating predictions as new data becomes available.;Gaussian process machine learning is chosen as the regression approach best suited for application in this problem. However, a review of existing batched Gaussian process theory results in the identification of gaps in the current knowledge base which render existing methods unsuitable. A new approach to batched Gaussian process regression is therefore developed, Sufficient-Subset Gaussian process iteration, which addresses the questions for which existing theories come up short. In the process of developing this new method fundamental contributions have been made to the areas of Gaussian process polynomial regression and sparse Gaussian process approximation theory.;Sufficient-Subset Gaussian process iteration is applied to both simulated and real turbine data and shown to be able to identify the sought dynamics to within a 3% error threshold. Additionally, a related regression formulation corresponding to maximum efficiency tracking is shown to present a potential method for turbine monitoring and fault detection."],"dc:identifier":["T15071"],"dc:identifier.doi":["10.48730/80p3-j963"],"dc:identifier.uri":["https://stax.strath.ac.uk/concern/theses/ht24wj46b"],"dc:publisher.department":["Department of Electronic and Electrical Engineering","Centre for Doctoral Training in Wind and Marine Energy Systems"],"dc:publisher.institution":["University of Strathclyde","University of Oxford","University of Edinburgh"],"dc:title":["Wind turbine dynamics identification using gaussian process machine learning"],"dc:type.qualificationlevel":["doctoral-pg"],"dc:type.qualificationname":["phd"]},"updated_at":"2026-07-24T04:47:42Z"}