Smart grids are complex systems for which simulation offers a practical way to evaluate and compare multiple solutions before deployment. However, the simulation of a Smart Grid requires the development of heterogeneous models corresponding to electrical, information processing, and telecommunication behaviors. These heterogeneous models must be linked and analyzed together in order to detect the influences on one another and identify emerging behaviors. We apply model-driven engineering to such cyber-physical systems combining physical and digital components and propose SGridSF, the Smart Grid Simulation Framework, which automates tasks in order to ensure consistency between different simulation models. This framework consists mainly of a domain specific language for modeling a cosimulation unit, called CosiML for Cosimulation Modeling Language, a domain specific language for modeling the functional architecture of a Smart Grid, called SGridML for Smart Grid Modeling Language, and a tool implementing different transformation rules to generate the files and scripts for executing a cosimulation. Finally, we illustrate the use of SGridSF on the real use case of an islanded grid implementing diesel and renewable sources, battery storage and intelligent control of the production. We show the sequencing of automatic generation tasks that minimizes the effort and the risk of error at each iteration of the process.