Convolutional Neural Network for Voltage Sag Source Azimuth Recognition in Electrical Internet of Things

Table 3

Classification indicator.

Fault type	Classification algorithm	Classification feature	Accuracy	Kappa
PPF	CNN	Multidimensional feature	100.00%	100.00%
	KNN	Multidimensional feature	83.24%	82.19%
		Disturbance power	88.53%	87.81%
		Equivalent impedance	76.18%	74.69%
		System slope	82.35%	81.25%
		Real part of current	82.35%	81.25%
	EL	Multidimensional feature	64.12%	61.88%
		Disturbance power	64.41%	62.19%
		Equivalent impedance	64.12%	61.88%
		System slope	64.71%	62.50%
		Real part of current	64.71%	62.50%
DLGF	CNN	Multidimensional feature	100.00%	100.00%
	KNN	Multidimensional feature	90.59%	90.00%
		Disturbance power	93.24%	92.81%
		Equivalent impedance	82.35%	81.25%
		System slope	89.41%	88.75%
		Real part of current	81.18%	80.00%
	EL	Multidimensional feature	63.82%	61.56%%
		Disturbance power	58.82%	56.25%
		Equivalent impedance	64.41%	62.19%
		System slope	64.12%	61.88%
		Real part of current	64.41%	62.19%
TLGF	CNN	Multidimensional feature	100%	100%
	KNN	Multidimensional feature	97.65%	97.50%
		Disturbance power	79.12%	77.81%
		Equivalent impedance	80.00%	78.75%
		System slope	31.18%	90.63%
		Real part of current	76.76%	75.31%
	EL	Multidimensional feature	64.12%	61.88%
		Disturbance power	35.29%	31.25%
		Equivalent impedance	35.29%	31.25%
		System slope	47.06%	43.75%
		Real part of current	64.12%	61.88%