Application of Feature-Extraction-Based Extreme Learning Machine Algorithm in Tunable Diode Laser Absorption Spectroscopy

Table 1. Spectral line parameters
View table
Table 1. Spectral line parameters
Molecule Wavenumber /cm^-1 γ_self γ_air Pressure broadening /GHz Doppler broadening /MHz
H₂S 6369.760 0.158 0.074 4.443 403.834
CO₂ 6369.408 0.088 0.069 4.192 354.970
CH₄ 6370.382 0.079 0.066 4.602 588.743

Table 2. Result by ELM

View table

Table 2. Result by ELM

Algorithm	Excitation function	Number of nodes	Training time /s	y_rmse /10^-6	R²
		200	<0.01	2.16	0.9941
		500	0.05	2.02	0.9948
	Line	1000	0.12	1.96	0.9951
		5000	0.33	1.96	0.9951
ELM		10000	0.70	1.97	0.9951
		5000	0.47	6.82	0.9409
		15000	1.84	6.77	0.9419
	Hardlim	20000	1.80	6.02	0.9864
		25000	3.38	6.17	0.9517

Table 3. Regression result by ELM based on feature extraction
View table
Table 3. Regression result by ELM based on feature extraction
Algorithm Excitation function Number of hidden nodes Training time /s y_rmse /10^-6 R²
10 <0.01 1.55 0.9969
NIPALS-ELM Line 20 <0.01 1.46 0.9973
30 <0.01 1.46 0.9973
200 <0.01 1.46 0.9973

Table 4. Comparison among several kinds of regression algorithms
View table
Table 4. Comparison among several kinds of regression algorithms
Algorithm y_rmse /10^-6 R²
CLS 18.19 0.5799
PLS 1.90 0.9954
ELM (line) 1.96 0.9951
NIPALS-ELM (line) 1.46 0.9973

Tools

Get Citation

Copy Citation Text

Lü Xiaocui, Li Guolin, Li Han, Ji Wenhai. Application of Feature-Extraction-Based Extreme Learning Machine Algorithm in Tunable Diode Laser Absorption Spectroscopy[J]. Chinese Journal of Lasers, 2018, 45(9): 911013

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites