Wind farm flow models can be ranked according to their “fidelity” level, referring to the degree of sophistication of the underlying mathematical description. When applied within their validation region, these physics-driven models are accurate but will never reach full “real-life” complexity. So, instead of developing models of ever higher fidelity, Schreiber et al. (2020) suggest improving basic models by adding correction terms learned from “real-life” operational data. A recent paper presents how to augment a wake model with SCADA data, and the method is applied to a wind tunnel experiment and an Italian wind farm.

The experiment consists of three wind turbines operating in a boundary layer wind tunnel. The wind turbines experience non-uniform inflow and secondary wake steering - physical effects not included in the baseline wake model. However, the augmented wake model can reproduce these effects. Similar results are found for the Italian wind farm located in complex terrain. Again, the baseline wake model does not include orographic effects, but marked improvements are achieved when augmenting with SCADA data.

Augmented flow models may be the goldilocks between physics-driven and purely data-driven models: they are neither too computationally expensive nor require too extensive datasets.

Schreiber, Johannes, Carlo L. Bottasso, Bastian Salbert, and Filippo Campagnolo. 2020. “Improving Wind Farm Flow Models by Learning from Operational Data.” Wind Energy Science 5 (2): 647–73. https://doi.org/10.5194/wes-5-647-2020.