Open AccessWhen using laser methods of gas analysis, one of the arising problems is instability in results of defining a quantitative composition of gases under control of multicomponent mixes in the conditions of real noise of measurements. It leads to demand for using the special algorithms to process results of laser measurements.
For multicomponent gaseous mixes, when solving a problem of quantitative gas analysis based on the results of multispectral laser measurements, use of methods for solving incorrect mathematical tasks is efficient.
If mix is stationary (i.e. there is a possibility for a series of measurements) it is possible to use a much simpler method to determine concentration of gases, i.e. the least-squares method based on the minimization of residual function.
However, the estimates obtained by the least-squares method are effective if distribution of measurement errors is according to the normal law. In practice, the law of errors distribution is often non-normal, and loss of estimate efficiency achieved by the least-squares method occurs even at a small share of bursts.
With bursts available in the measuring signal, it is necessary to use the stationary estimation methods allowing the significantly reduced impact on the estimate of considerable bursts.
To estimate an efficiency of the robust methods for defining a quantitative composition of the multicomponent stationary gas mixes from multispectral laser measurements a mathematical simulation was performed. A gas mixture was considered to be stationary, and n measurements (at each wavelength) were taken ( n were specified from 2 to 6) to define a quantitative composition of gases in the mixture. Simulation was implemented for gas mixes with the number of components from 4 to 6.
Results of mathematical simulation show that the robust estimate based on the residual function ( x ) arctg x , allows us, in conditions of the bursts of a variable signal, to reduce significantly the error of determining concentrations of the gas mix components as compared to the averaging, by the least-squares method and other options of function ( x ).