In order to overcome the defects of shuffled frog leaping algorithm (SFLA) such as slow searching speed in the late evolution and easily trapping into local extremum, an Improved Shuffled Frog Leaping Algorithm (ISFLA) based on the basic ideas of Molecular Dynamics (MD) simulations is put forward in this paper in which the population is regarded as molecular system. We propose a new intermolecular force instead of the classic two-body Lennard-Jones force and use Velocity-Verlet algorithm and Gaussian mutation instead of the original SFLA update strategy, which balance the population diversity and search efficiency effectively. The test results on high-dimensional and multi-modal optimization problems indicate that ISFLA improves the capacity of escaping from local maximum and the global searching performance is superior to SFLA.