CyberSEES: Type 1: Collaborative Research: Large-Scale, Integrated, and Robust Wind Farm Optimization Enabled by Coupled Analytic Gradients

风能是一种可再生能源，也是新能源装置最具成本效益的能源之一。如今，风力涡轮机是为隔离环境而设计的，其功率调节策略也是如此。当涡轮机被组装成风力发电场时，它们的尾迹会严重干扰其他涡轮机，导致与预期相比产生的能量不足10%-20%。这种产量不足是增加风能增长的主要障碍。该项目假设，通过同时设计风力机布局、功率调节策略和涡轮机本身，可以显著提高发电量，所有这些都存在随机输入。同时布局-控制-涡轮机设计是具有挑战性的，特别是在考虑不确定输入的情况下。目前的研究和行业实践使用的模拟模型是不可微的或不提供梯度的。因此，大多数风电场布局优化被限制在10-100个变量左右，依赖于顺序设计流程，并且只以简单的方式包含不确定性(如果有的话)。为了能够解决更大规模和更复杂的设计问题，尾流和涡轮机模型必须在考虑可扩展优化的情况下重新实现，并且必须开发新的不确定性量化方法。研究人员最近的工作表明，通过开发提供精确导数的风力涡轮机尾流模型，可以在比目前行业解决的变量多100到1000倍的情况下有效地进行风电场布局。这种可扩展性将使风电场优化包括大量设计变量，集成多个学科，并在设计过程中纳入不确定性。这些建议的贡献旨在促进能源可持续性、科学计算和教育。尾流和涡轮机模型将是大规模优化的，准备让设计师解决以前遥不可及的问题。新的不确定性量化方法将广泛适用于多个学科，特别是随着越来越多的行业走向集成系统设计。最后，一个专门的网站将作为一个教学工具，通过互动的风电场设计问题向普通观众介绍优化和不确定性量化的概念。同时，研究人员将专注于可扩展不确定性量化的基本方法，这些方法既可用于不确定性的正向传播，也可用于统计反演。需要强调可伸缩性，以解决与随机输入变量的数量、感兴趣的输出量的数量以及在极端规模的计算机上并行实现的效率有关的挑战。例如，研究团队正在开发可扩展不确定性量化的新方法，利用这些涡轮机和尾流模型提供的精确导数。该研究计划集中于三个主要目标：1)开发新的具有精确梯度的尾流模型，2)进行布局-控制-涡轮一体化优化，以及3)开发可扩展的不确定性量化方法，以展示在稳健的风电场布局问题上预期的性能改进。

Wind provides a renewable source of energy and is one of the most cost-effective sources for new energy installations. Today, wind turbines are designed for an isolated environment, as are their power regulation strategies. When turbines are assembled into a wind farm their wakes significantly interfere with other turbines resulting in energy underproduction of 10-20% relative to expectations. This underproduction is a major barrier to increased wind energy growth. This project hypothesizes that a significant increase in power production is possible through simultaneous design of wind turbine layouts, power-regulation strategies, and the turbines themselves, all in the presence of stochastic inputs.Simultaneous layout-control-turbine design is challenging, especially when considering uncertain inputs. Current research and industry practices use simulation models that are non-differentiable or do not provide gradients. As a result, most wind farm layout optimizations are limited to around 10-100 variables, rely on sequential design processes, and only include uncertainty in simple ways if at all. To enable design problems of larger size and complexity, wake and turbine models must be reimplemented with scalable optimization in mind, and new methods for uncertainty quantification must be developed. The investigators' recent work suggests that by developing wind turbine wake models that provide exact derivatives, wind farm layout can be done effectively with 100 to 1,000 times more variables than those solved by the industry today. This scalability will enable wind farm optimization that includes a large number of design variables, integrates multiple disciplines, and incorporates uncertainty in the design process. These proposed contributions seek to advance energy sustainability, scientific computing, and education. The wake and turbine models will be large-scale-optimization ready to allow designers to solve problems that were previously out of reach. The new uncertainty quantification methodologies will be widely applicable to multiple disciplines, particularly as more industries move towards integrated system design. Finally, a dedicated website will serve as a teaching tool to introduce optimization and uncertainty quantification concepts to a general audience through interactive wind farm design problems.Concurrently, the investigators will focus on foundational methods for scalable uncertainty quantification that can be used for both forward uncertainty propagation and statistical inversion. The emphasis on scalability is required to address challenges related to the number of random input variables, the number of output quantities of interest, and the efficiency of parallel implementations on extreme-scale computers. As an example, the research team is developing new methodologies for scalable uncertainty quantification that take advantage of the exact derivatives provided by these turbine and wake models. The research plan focuses on three main goals: 1) develop new wake models with exact gradients, 2) perform integrated layout-control-turbine optimization, and 3) develop scalable uncertainty quantification methods to demonstrate expected performance improvements on robust wind farm layout problems.