To address the problems of severe cross-sensitivity

strong environmental noise interference

and signal nonlinearity in monitoring SF

6

decomposition products in gas-insulated switchgear(GIS)

a gas monitoring system based on whispering gallery mode(WGM)microcavity sensor arrays and deep learning is proposed.At the sensing layer

a differentially functionalized WGM sensor array is constructed using functionalized materials

such as metal oxides and gold nanoparticles.A reference channel is also introduced to suppress environmental common-mode noise.At the signal processing layer

an intelligent decoupling model combining con

volutional neural network(CNN) and long short-term memory(LSTM)network is developed to extract spatial fingerprints and temporal dynamic features of the array

achieving intelligent decoupling and noise filtering.Experimental results demonstrate that the system can effectively mitigate mixedgas interference

achieving significantly higher accuracy than traditional algorithms.Furthermore

it exhibits excellent robustness against temperature drift and light intensity disturbance

providing a high-sensitivity solution for early warning of GIS.