Ju Feng authored 4 years ago and Mads M. Pedersen committed 4 years ago

Replacing scipy.interpolate.interp1d with a user defined class, as interp1d can be slower in some cases.

Ju Feng authored 4 years ago and Mads M. Pedersen committed 4 years ago

Replace lambda function with scipy interpolate.interp1d, to avoid pickle error in certain cases using multiprocessing.

Implemented wake_radius() function for the different wake models and modified args4turb to be independant from deficit

- add test, which tests aep probability normalizing for two turbines
- remove not-needed autograd switch

Added different superposition models for turbulence intensity, as StenFrandsen uses effective turbulence intensity instead of the usual added turbulence intensity

Move the split line between wake and blockage is set slightly downstream to handle numerical inaccuracy in the trigonometric functions that calculates dw_ijlk

add selfsimilarity blokage model
prepare for gradients
New structure. Old WakeModel mainly in new EngineeringWindFarmModel
add deflectionmodel
update documentation

Prepare for iterative calc_wake backend by adding an the source-wt dimension(i) to all calc_deficit arguments

Ju Feng authored 5 years ago and Mads M. Pedersen committed 5 years ago

Solve two issues in noj: 1. variables passed to np.arccos() could be lareger than 1.0 or less than -1.0 for some combination of R1, R2 and d to cal_overlapping_area_factor. 2. Ct values passed to np.sqrt(1-Ct) could be larger than 1. These two issues both result errors/nan values. Now these variables/values have to pre-treated to avoid such results.