An online device monitors dissolved gases in oil, ensuring transformer safety and stability. The chromatographic curve of low-concentration gas exhibits noise and irregular peaks, leading to misjudgment and inaccurate gas concentration calculation, impacting transformer condition assessment. To address this, the CEEMDAN algorithm decomposes the chromatographic signal, followed by wavelet threshold denoising to remove noise. Finally, the denoised signal is combined with low-frequency components to generate the final chromatographic curve. A fitted Gaussian waveform is derived from the measured peaks using the Spearman correlation coefficient method. Sliding window analysis identifies chromatographic peaks by measuring the correlation coefficient between the Gaussian waveform and chromatographic curve data, determining peak presence when the correlation exceeds a threshold. The experimental results demonstrate the method's effectiveness in noise reduction and accurate peak identification, laying the foundation for dissolved gas quantification in oil.