Network traffic measurement plays a crucial role in modern network management. However, existing network measurement approaches struggle to adapt dynamically to current network traffic characteristics while maintaining good bandwidth efficiency and query accuracy, especially in distributed scenarios. In this paper, we propose a novel network traffic measurement framework, DDC sketch, which leverages three key modules to tackle these issues: Division, Dynamic Machine Learning, and Compression. The Division module separates elephant flows from mice flows, enabling targeted handling for each category. The Dynamic Machine Learning module dynamically optimizes query results to improve query accuracy. Meanwhile, the Compression module reduces the size of the query structure transmitted to the remote control center, enhancing overall bandwidth efficiency. We conduct extensive experiments using three distinct datasets to evaluate the performance. Results show that our DDC sketch can outperform comparison techniques by 2.16 times in terms of accuracy.