Abstract:
There have been increasing demands for solving various optimization problems arising from complex physical systems. Conventional optimization approaches are widely used to provide a deterministic solution for decision support based on computer simulators, which is however unable to account for various sources of uncertainties. For complex high-dimensional systems, simplifications are often used to conduct optimization, which leads to the ‘optimal’ solution being suboptimal or nonoptimal. Even where the optimization problem is well resolved, it would still be valuable for both operational and long-term planning purposes to introduce some diversity to the solution. Our work seeks to introduce a general framework for solving optimization for complex systems, such as energy systems. The proposed framework treats an optimization problem into a history match problem, which is addressed using statistical emulation along with proper uncertainty quantification. Emulation constructs fast statistical approximations to the complex computer simulator of a given objective function(s) in order to identify appropriate choices of candidate solutions. Uncertainty quantification is adopted to capture multiple sources of uncertainty attached to each candidate solution and is conducted via Bayes linear analysis. It is demonstrated in a practical wind farm planning case study that applying the proposed methodology can overcome various challenges in conventional optimization approaches.
Speaker Bio:
Hailiang Du is an Associate Professor in the Department of Mathematical Sciences at Durham University, where he has been since 2017. He received his PhD in statistics from the London School of Economics and Political Science (LSE) in 2009. After his PhD, he worked in the Centre for the Analysis of Time Series at LSE for 4 years and in the Center for Robust Decision making on Climate and Energy Policy at the University of Chicago for 3 years as a research scientist. His research interests and work cover a variety of research topics including uncertainty quantification, theory of nonlinear dynamics, machine learning, data assimilation, optimization, multi-model forecasting, Bayesian linear analysis, weather and climate modelling, forecast interpretation and evaluation.
The talk will be in English and is open to all members of DKU community.
Register in: https://duke.qualtrics.com/jfe/form/SV_1ZdEhyZhSlBDBd4
Last registration by Sep. 1, Thursday.
