Improving selectivity of infrared gas analyzer with neural network

Present applied research on how to improve selectivity of dispersive infrared gas analyzer (DIGA) is mostly confined to the improvement of hardware techniques with new structure, material and technology, which have insoluble deficiencies when non-aim gases bring forth cross absorption during the characteristic absorption spectrum of gas to be measured. The arithmetic of BP neural network can be used to eliminate the cross-interfere absorption and consequently improve selectivity of DIGA. When detecting methane in petroleum fission gas, measuring gas methane measurement scale is 0 to approximately 7600 x 10-6 and interfere gas ethene concentration varies 7600 X 10-6. After using neural network data fusion, the selectivity index of DIGA can be increased from 3.17 to 422 and the relevant fluctuation error of main sensor output decreased from 57.9% to 0.65%. The experimental result indicates the method has practical application value.