A Critical Review on Wind Turbine Power Curve Modelling Techniques and Their Applications in Wind Based Energy Systems

Table 3

Details of some nonparametric models from various studies.

Ref.	Data set			Model
	Interval	Data	Number of values	Model	Parameters	Structure/transfer function (TF)/training method
[4]	10 min	SCADA data 100 WTs	Total 4347; Training 3476; Testing 871	-NN		Euclidian distance metric
[17]	10 min	12 WTs; (wind speed and direction from two meteorological towers)	Training; 1500 patterns for each WT	ANN	Number of hidden layers = 1; Number of hidden layer neurons = 8	(i) Separate MLP network for each WT; (ii) Training-pattern mode; (iii) TF-hyperbolic (all layers)
[8]	—	Measured data; 100 kW WT	—	Fuzzy; clustering	Number of cluster centers = 8	(i) CFL; (ii) Subtractive clustering
[24]	10 min	Data set 1 generated with method in [9]	Total 1008; Training 50%; Testing 50%	ANN	Number of hidden layer neurons = 5	(i) Feed forward back propagation; (ii)Training: Levenberg-Marquardt; (iii) TF: hidden layer transig; (iv) TF: output layer purlin
		Data set 2	Total 4388; Training 50%; Testing 50%
		Data sets 3, 4, and 5	Total 2208; Training 50%; Testing 50%
		Data sets; 1–5 as above	As above	Fuzzy; clustering	Number of cluster centers = 8	Fuzzy -means
[2]	10 min	SCADA (three 2 MW WTs); (model type 1: wind speed; Model type 2: wind speed and direction, temperature)	32796; Training 60%; Validation 40%	Fuzzy; clustering	Number of cluster centers; Type 1 = 3; Type 2 = 6	CCFL
				MLP; NN	Number of hidden layers = 2	(i) Training-gradient descent; (ii) TF: hidden layer-sigmoid; (iii) TF: output layer-linear
				-NN	Type 1; Type 2	—
				ANFIS		(i) FIS structure Sugeno type; (ii) Training-hybrid learning; (iii) Membership functions; Input space-generalized normal; Output space-linear; (iv) Number of MFs = 3