Table 5 Numerical results for test Case III.

From: Surrogate modeling of passive microwave circuits using recurrent neural networks and domain confinement

Domain

Modeling method

Average relative RMS error

NB = 50

NB = 100

NB = 200

NB = 400

NB = 800

NB = 1600

Original (X)

Kriging

63.6%

53.8%

45.2%

40.0%

35.1%

32.3%

RBF

68.9%

55.2%

43.9%

40.8%

37.2%

34.7%

GPR

70.2%

69.1%

56.4%

47.9%

42.3%

70.2%

SVR

79.6%

72.6%

68.4%

63.2%

61.0%

59.6%

ANN 1

78.3%

77.4%

65.2%

59.4%

37.2%

33.3%

ANN 2

82.2%

78.3%

66.2%

61.2%

49.8%

42.3%

ANN 3

79.3%

77.8%

64.8%

59.7%

45.5%

44.1%

RNN-LSTM (this work)

75.3%

55.6%

37.8%

26.8%

17.8%

16.3%

Reduced (Xd)

Kriging

38.9%

28.7%

23.5%

16.6%

12.5%

8.4%

RBF

42.5%

31.3%

26.0%

18.1%

14.1%

9.9%

GPR

60.7%

51.8%

41.1%

41.1%

33.7%

25.8%

SVR

55.5%

51.9%

47.7%

44.0%

39.1%

37.2%

ANN 1

52.4%

34.0%

23.2%

16.8%

12.0%

10.0%

ANN 2

47.5%

35.7%

24.6%

19.5%

14.6%

11.1%

ANN 3

48.7%

36.2%

25.1%

18.7%

15.1%

11.4%

RNN-LSTM (this work)

50.7%

29.3%

18.8%

11.9%

7.6%

5.6%