In Cloud environment, a number of compute resources are distributed in a resource pool, and a variety of applications can get computing resources, storage space and all kinds of software services if they need through information sharing and real-time communication on the internet. SWF focuses on the data analysis, so the correctness, integrity, consistency and independence of data are the premises to guarantee the normal execution of the workflow. Scientific workflow is based on a large-scale data calculation and dynamic execution of scientific computing is the foundation of efficient execution of the scientific workflow. Here, we construct a platform based on Hadoop and propose an XML-DMM that could execute effectively and flexibly in Cloud environment. Finally, we also design a SST Fusion experiment to prove the rationality and availability of the modeling method, and the superiority of dynamic execution mechanism in scientific workflow.