Pipeline-Oriented Multi-Fidelity Computational Workflow for Large-Scale Wind Farm Micro-Siting Optimization

Abstract

Modern wind farms have expanded in size and complexity to the point that even with high-fidelity wake and flow models, micro-siting optimization is a computationally expensive subject that remains open to optimization research. Traditional micro-siting methods generally are either monolithic or single-fidelity optimization pipelines which have the disadvantage of being prohibitively expensive in terms of computational cost, and poorly scale to large numbers of turbines and design variables. In addition, the current multi-fidelity approaches tend to be devoid of systematic workflow orchestration and this construct is restrictive when it is used in large and distributed computing environments. In this paper, a pipeline-based multi-fidelity computational workflow is presented that can optimise the problems of large-scale wind farms using micro-siting optimization, which is aimed at simultaneously addressing the issues of accuracy, efficiency, and scalability. The suggested framework combines the formal micro-siting problem formulation with the hierarchical fidelity modeling, automated pipeline organization and fidelity-sensitive optimization. Low-fidelity models are used in more situations where the higher-fidelity evaluations are activated selectively to be refined and validated within a single flow. The implementation of the workflow is through the distributed execution strategies that take advantage of the task-level parallelism, dependency management and allocation of resources that is scalable. A significant amount of experimental work as to wind farm cases of growing magnitude show that the proposed pipeline is much faster to compute than single-fidelity and non-pipeline multi-fidelity case, and is comparable or better in energy yield optimization performance. Scalability performance indicates that the performance increases almost linearly with increased computational resources. The findings emphasise the usefulness of pipeline-oriented multi-fidelity workflows as a useful and scalable method of addressing computerally intensive optimization problems in renewable energy, and reinstate their more general applicability as large-scale engineering processes that need automated and distributed execution.