With the rapid development of blockchain technology, cryptocurrency exchanges, as a critical application, have experienced a substantial increase in transaction volume. This surge provides opportunities for illicit activities such as money laundering and market manipulation, necessitating an effective regulatory framework. In this paper, we propose a regulatory approach for monitoring trading volumes. Our method entails a detailed analysis of trading data from multiple exchanges. We conduct feature engineering on the data and enhance features from a time series perspective. Utilizing deep learning models, we effectively capture the dynamic characteristics of the data and improve prediction accuracy. Furthermore, we establish refined dynamic threshold intervals based on data dynamics, enabling more efficient anomaly detection. We validated using real transaction data from Ethereum and compared it with conventional methods. The results demonstrate that our framework achieves higher prediction accuracy, greater sensitivity in anomaly detection, and more comprehensive detection results, thereby enhancing regulatory.