PhD project title :Data-driven risk-averse control of offshore floating wind turbines with embedded GPUs
Outline description,including interdisciplinary, intersectoral andinternational dimensions:
Floating offshore wind turbines are exposed to wind and wave-induced motions that tend to reduce their productivity, cause fatigue to the overall structure, jeopardise their long-term structural integrity, and may even have a destabilising effect with dire consequences, especially under adverse weather conditions. Stochastic model predictive control methodologies offer unparalleled reliability, safety and performance properties, but come with two important challenges: (i) they require prior knowledge of the system dynamics and probabilistic properties of disturbances which, in this case, are not fully available, (ii) the associated computational cost is a limiting factor that hampers their applicability. The highly nonlinear dynamics of floating offshore wind turbines only exacerbate this situation.
To address the first challenge, in this project the ESR will develop novel data-driven methods within the framework of risk-averse model predictive control: a novel control-theoretic framework that allows for partial knowledge of the underlying dynamics and distributions of disturbances. This will lead to a unique amalgamation of control theory and statistical learning that will lead to notions of stochastic stability with high probability – this is a highly interdisciplinary aspect of this project. This will be in contrast to empirical machine learning approaches that come with no theoretical guarantees and rigorous control methods that come with strong assumptions on prior knowledge of the system. Special focus will be on formulations with probabilistic constraints.
To address the second challenge, the ESR will develop new numerical optimization methods for large-scale nonconvex problems. The objective will be to devise semismooth Newton-type and quasi-Newtonian methods that will exploit the problem structure to achieve a fast and accurate solution of non-convex non-smooth riskaverse optimization problems. Most importantly, the ESR will devise numerical methods that are amenable to parallelization on the lockstep architecture of modern embedded GPUs (such as the NVIDIA Jetson TX1 and Nano).
Key words/descriptors: High-performance computing, Floating offshore wind turbines, Data-driven control, Model predictive control,
Fit to CITI-GENS theme(s)
First Supervisor: Dr Pantelis Sopasakis School: EEECS
Second Supervisor: Dr Madjid Karimirad School: Natural & Built Environment
Third Supervisor: Prof Alberto Bemporad Company: ODYS Srl
Name of non-HEI partner(s) ODYS Srl
Contribution of non-HEI partner(s) to the project:
ODYS Srl is a private SME and one of the few companies worldwide that design and implement advanced embedded optimization solutions for demanding applications (e.g., automotive, aerospace, energy and process control). ODYS will:
• Host and supervise the ESR for 6 months. The student will be under the supervision of the company founders, Prof Alberto Bemporad (5% FTE) and Dr Daniele Bernardini (5% FTE). The ESR will be hosted in ODYS operating office in Milano. The office is located in a start-up hub that offers a suitable
environment for interactions with other innovative companies,
• Provide training on intellectual property.
Prof. Alberto Bemporad is a full professor of Control Systems at IMT School for Advanced Studies Lucca and has published more than 300 papers in the areas of MPC, hybrid systems, automotive control, multiparametric optimization, computational geometry, robotics, and finance. He is the author/coauthor of various MATLAB toolboxes for MPC design, including the Model Predictive Control Toolbox (The Matworks, Inc.). He received the IFAC High-Impact Paper Award for 2011-14. He is IEEE Fellow since 2010.
Subject area: Control/Electrical/Mechanical Engineering
Queen's University Belfast is committed to Equality, Diversity and Inclusion.
For more information please read our Equality and Diversity Policy.
Queen's University Belfast is registered with the Charity Commission for Northern Ireland NIC101788
VAT registration number: GB 254 7995 11