Abstract: Against the early and rapid monitoring and warning of thermal runaway of lithium-ion batteries , this article proposes a fast Raman detection and recognition algorithm for thermal runaway characteristic gases. The method innovatively develops a least squares convolutional neural network (PLSR-CNN) algorithm, which accurately extracts intrinsic signals from low signal-to-noise ratio Raman gas signals and combines partial least squares regression (PLSR) technology for quantitative analysis, greatly improving the detection and warning effect of Raman gas detection technology. The results show that the accuracy of this method in qualitative analysis is as high as 99.8%, and in quantitative analysis it is as high as 96.4%. Therefore, this method has good application prospects in the field of detection, providing theoretical and technical support for further improving the accuracy of thermal runaway gas identification and improving gas identification models.

Key words: thermal runaway gas, Raman, CNN, PLSR