Computational Intelligence and Neuroscience

Research Article

Inversion of Rayleigh Wave Dispersion Curves via Long Short-Term Memory Combined with Particle Swarm Optimization

Table 3

Mean values of Model A and B inverse results (IMV denotes the inverse mean value, RE denotes the relative error, and SD denotes the standard deviation).


Model A
Parameters	True	PSO				LSTM
		IMV		RE (%)	SD	IMV	RE (%)	SD

Vs₁ (m/s)	160	162.81		20.76	6.27	156.84	1.98	0.81
Vs₂ (m/s)	260	311.80		20.24	57.37	264.86	1.87	1.64
Vs₃ (m/s)	200	194.58		14.76	45.42	195.79	2.11	0.96
Vs₄ (m/s)	400	404.44		1.76	10.36	398.89	0.27	1.97
H₁ (m)	2	2.42		19.92	0.53	1.88	5.85	0.36
H₂ (m)	3	2.39		2.71	1.02	2.95	1.63	0.17
H₃ (m)	5	5.74		1.11	2.18	4.91	1.73	0.41

Model A
PSO-LSTM
Parameters	True		IMV		RE (%)		SD
Vs₁ (m/s)	160		160.80		0.51		0.39
Vs₂ (m/s)	260		264.96		1.91		0.59
Vs₃ (m/s)	200		201.79		0.90		0.58
Vs₄ (m/s)	400		397.10		0.73		1.23
H₁ (m)	2		2.04		2.05		0.01
H₂ (m)	3		2.97		1.10		0.02
H₃ (m)	5		4.97		0.57		0.02

Model B
Vs₁ (m/s)	200		203.74		1.87		0.93
Vs₂ (m/s)	160		163.66		2.09		1.24
Vs₃ (m/s)	300		299.09		0.30		1.66
Vs₄ (m/s)	400		403.64		0.91		3.87
H₁ (m)	2		1.98		1.22		0.02
H₂ (m)	4		3.96		1.05		0.04
H₃ (m)	6		5.94		0.93		0.06