Abstract: Using the fault gas carbon dioxide observation data from Tianshui seismic station from 2022 to 2025 as the research object, db4 wavelet was adopted for analytical processing, and short-term low-frequency and high-frequency anomalies were investigated. The results show that the wavelet analysis method has a strong capability of identifying information at different frequencies. It facilitates the separation of trend variations and local fluctuations in data, effectively recognizes and eliminates interference factors, and presents seismic precursor anomalies intuitively and prominently, thus serving as an effective technical approach to seismic precursor data processing. Earthquake case verifications were conducted on the M_S6.2 Jishishan earthquake on December 18, 2023 and the M_S5.6 Longxi earthquake on September 27, 2025 in Gansu Province. Jump anomalies of fault gas carbon dioxide emerged 3～6 months prior to earthquakes with magnitude 5.0 and above within a 300-kilometer range around the station, characterized by carbon dioxide concentration variations exceeding twice the mean square deviation.