The detection of dissolved acetylene in transformer oil is a critical component for early fault detection in power equipment. The optical thermal spectroscopy platform demonstrates significant advantages in high-sensitivity detection of low-concentration C2H2 gas. With the advancement of intelligent power transformer detection, developing rapid and accurate quantitative analysis methods for fault characteristic gases has become crucial. This study proposes a 1DCNN-LSTM model for acetylene concentration prediction, which integrates the local feature extraction capability of convolutional neural networks (CNN) with the sequence-dependent modeling advantages of long short-term memory networks (LSTM). Compared to traditional 1DCNN models, the 1DCNN-LSTM achieves superior predictive performance: the model's test set coefficient of determination (R2) reaches 0.988 with a root mean square error (RMSE) of 31.89, representing a 2.4% increase in R2 and a 53.2% reduction in RMSE compared to 1DCNN. This effectively enhances the accuracy and goodness of fit in acetylene concentration prediction, providing valuable references for quantitative analysis and detection of fault characteristic gases in transformers.