An online SO₂ and CS₂ detection system combining weighted elliptical dual-spectrum reconstruction (WEDSR) with...

Area of Science:

• Electrical Engineering
• Analytical Chemistry
• Spectroscopy

Background:

• Sulfur dioxide (SO2) and carbon disulfide (CS2) are crucial indicators for assessing gas-insulated switchgear (GIS) insulation health.
• Significant spectral overlap between SO2 and CS2 in the UV band complicates simultaneous quantification using UV differential optical absorption spectroscopy (UV-DOAS).

Purpose of the Study:

• To develop a novel method for accurately and simultaneously quantifying SO2 and CS2 concentrations in GIS.
• To address the limitations of existing UV-DOAS techniques caused by spectral overlap and low signal-to-noise ratios.

Main Methods:

• Proposed a weighted elliptical dual-spectrum reconstruction (WEDSR) method to enhance and decouple SO2 and CS2 spectral signals.
• Implemented a synergistic algorithm with weighted fitting and iterative optimization to improve ellipse fitting accuracy and suppress noise.
• Developed a convolutional neural network (CNN) model for quantitative detection of SO2 and CS2.

Main Results:

• The WEDSR method effectively decoupled SO2 and CS2 spectral signals.
• The CNN model achieved high accuracy in quantitative detection, with low mean absolute percentage errors (MAPEs) of 0.818% for SO2 and 0.963% for CS2.
• Achieved low detection limits of 7.71 ppb·m for SO2 and 0.35 ppb·m for CS2, demonstrating effective noise suppression under ultra-low concentrations.

Conclusions:

• The developed system offers a highly accurate and stable solution for simultaneous SO2 and CS2 quantification in GIS.
• This technique holds significant potential for early warning and monitoring of faults during GIS operation, enhancing grid reliability.