A novel dynamic wind farm wake model based on deep learning

Abstract

Please refer to published version for the most recent bibliographic citation information. If a published version is known of, the repository item page linked to above, will contain details on accessing it.

Data-driven fluid mechanics of wind farms: A review

“…Then, the flow field around each turbine at the hub-height level was extracted from data and was time-averaged, generating three training samples from each LES simulation. The values for inflow wind speed, TI, and yaw angles considered in this study were similar to the values considered in the previous study of the same authors 69 . Finally, 270 training samples, at the expanse of 1 million CPU hours, were generated and utilized in their study.…”

“…In this section, objective, utilized data, and methodology of the studies focusing on data-driven wind-farm flow modeling are discussed in two groups based on their approach, i.e., DDM and PGDDM approaches. The majority of studies reviewed in this article [67][68][69][70][71][72][73][74][75][76][77][78][79][80][81][82][83][84] follow the DDM approach or the black-box modeling without the need for knowing the governing equations of the problem. But we can see that some recent works from Howland and Dabiri 85 , Yan et al 86 , Park and Park 87 , and Sun et al 88 have introduced physics into their data-driven analysis; therefore, they can be labeled as PGDDMs.…”

Spatiotemporal wind field prediction based on physics-informed deep learning and LIDAR measurements

“…Machine learning, in particular deep learning [14], is developing very fast in the past few years, and its applications in wind industry have also seen great successes e.g. in wind power forecasting [15], wind speed forecasting [16,17], and wind farm wake modeling [18,19]. However, the incorporation of physical laws in the training of deep learning models has not been explored in wind energy studies while such ideas are emerging in other physical systems such as the databased turbulence modeling [20,21], the discovery of governing equations [22,23], solving high-dimension partial differential equations (PDEs) [24] and the surrogate modeling of phys-ical systems [25,26].…”

“…2) Previous machine learning based wind prediction studies e.g. [16,18,40] treat machine learning models as "black-box" and require the corresponding input and target values for training. Then they can predict the wind patterns which are present in the training dataset.…”

Wind Farm Power Optimization and Fault Ride-Through under Inter-Turn Short-Circuit Fault

“…The higher C t , the greater the wind velocity deficit of the downstream WT and the less the power that can be produced. In addition, some reducedorder wake models [22] can obtain a relatively good accuracy at a lower computational cost compared with high-fidelity wake models. The choice of the wake model affects the computational time, the complexity of the optimization and the control methods.…”