EAGER-DynamicData: Minimizing Wind Farm Operation and Maintenance Cost Using Data-Driven Models

在过去几年中，风电场安装的增长导致了达到其制造保修期的风力涡轮机数量的增加。因此，风力发电行业现在面临着不定期维护成本的上升，这增加了运营和维护支出。该研究项目将开发新的操作和维护策略，以提高风电场系统的可靠性，从而降低风能成本。风力涡轮机在导致风力涡轮机部件故障的恶劣条件下操作，这是难以预测的。因此，它是具有挑战性的，以安排维护行动，使这样的组件故障可以避免或最小化。由于美国风力发电场运行和管理的成本不断上升，设计使用可用风力涡轮机传感器数据的方法对于降低风力发电场运行成本至关重要。为了实现这一目标，本研究考虑新的维护调度模型和算法，考虑到数据的不确定性，涡轮机的状态，天气条件，以及执行维护所需的资源的可用性。如果成功的话，这种探索性的研究将能够更快的初始维护响应和更好地利用风能系统中有限的维护资源。 有效利用昂贵的资源将促进竞争力，并将有助于降低风能成本，实现电价稳定，减少对全球燃料市场的依赖。有必要制定准则，以降低风能系统的运营和维护成本。为了实现这一目标，本研究的目的是建立随机在线数据支持模型和算法如何能够导致风力涡轮机快速损伤检测和故障减少。随机在线优化是这个问题的一个合适的框架，因为它显式地吸收随机数据，随着时间的推移演变到优化模型，使强大的决策之前，观察整个随机数据流顺序。该项目的动机来自于对风能维护规划数据驱动方法的需求。拟议的工作将解决基本的科学和工程挑战，以成功地推导出一个数据驱动的随机在线优化算法，用于风能系统的运行和维护。这项研究将通过贡献计算和数据支持的概念，模型和算法来推进风电场运营和维护的最新技术。这项研究的成果将通过引入以下内容将风能系统中的维护计划扩展到一个新的水平：1）维护和资源调度的数据驱动优化模型和2）随机在线优化的新算法。特别是，新的数据驱动的方法，集成随机规划与随机在线优化，这通常是孤立对待，将产生一个新的和可行的方法随机数据同化到决策下的不确定性复杂的多实体工程系统，如风电场。

The growth in wind farm installations in the past years has led to an increase in the number of wind turbines reaching the end of their manufacturing warranties. Therefore, the wind industry is now faced with a rising cost of unscheduled maintenance which is increasing operation and maintenance expenditures. This research project will develop new operation and maintenance strategies for improving the reliability of wind farm systems so that wind energy cost can be reduced. Wind turbines operate under harsh conditions that lead to wind turbine component failures, which are difficult to predict. Consequently, it is challenging to schedule maintenance actions so that such component failures can be avoided or minimized. Because of the continuously escalating cost of wind farm operation and management in the United States, devising methods for using available wind turbine sensors data is critical to decreasing wind farms operational costs. To accomplish this objective, this research considers new maintenance scheduling models and algorithms that take into account data uncertainties in turbines status, weather conditions, and the availability of the resources needed to perform maintenance. If successful, this exploratory research will enable faster initial maintenance response and better utilization of limited maintenance resources in wind energy systems. The efficient utilization of costly resources will foster competitiveness and will contribute towards reducing the cost of wind energy, achieve electricity price stability, and reduce dependency on global fuel markets. There is a need to establish guidelines to reduce operation and maintenance costs in wind energy systems. In pursuit of this goal, the objective of this research is to establish how stochastic online data-enabled models and algorithms can lead to wind turbine rapid damage detection and failure reduction. Stochastic online optimization is a suitable framework for this problem since it explicitly assimilates stochastic data that evolve over time into the optimization model, enabling robust decisions to be made sequentially prior to observing the entire stochastic data stream. The project motivation comes from the need for a data-driven methodology for maintenance planning in wind energy. The proposed work will address basic scientific and engineering challenges toward the successful derivation of a data-driven stochastic online optimization algorithm for the operation and maintenance of wind energy systems. This research will advance the state-of-the-art in wind farm operation and maintenance by contributing computational and data-enabled concepts, models and algorithms. The outcomes of this research will extend how maintenance is scheduled in wind energy systems to a new level beyond the current state of practice by introducing: 1) data-driven optimization models for maintenance and resource scheduling and 2) new algorithms for stochastic online optimization. In particular, the new data-driven methodology, integrating stochastic programing with stochastic online optimization, which are usually treated in isolation, will give rise to a new and viable approach for stochastic data assimilation into decision-making under uncertainty for complex multi-entity engineered systems such as wind farms.