This study uses LSTM autoencoder networks to detect anomalies in government bond yields and trading volumes. By analyzing yield data for 6-month, 1-year, 5-year, 10-year, and 30-year government bonds, we found that short-term bonds (6 months and 1 year) showed fewer anomalies, indicating a relatively stable market influenced by expectations of monetary policy changes. In contrast, medium-term bonds (5-year and 10-year) displayed more anomalies, reflecting more frequent market fluctuations and sensitivity to interest rate expectations and market sentiment. No anomalies were detected in long-term bonds (30 years), likely due to their lower volatility and fewer market participants, making them less prone to manipulation. The study also analyzed the trading volume of 5-year government bonds, finding some overlap between trading volume anomalies and yield anomalies, highlighting areas that warrant closer attention. These findings can help regulators improve monitoring systems, particularly focusing on medium-term bonds, which are more susceptible to market volatility and manipulation risks. Future research will further optimize algorithms and incorporate more institutional trading data to enhance the timeliness and accuracy of market surveillance, providing more efficient tools for regulation.