While power systems are interconnected more closely and deregulated in electricity markets

distributed dynamic simulations among different dispatch centers are highly required for online full system analysis and control. In this paper a new algorithm for distributed dynamic simulation of interconnected power systems is presented. This algorithm is based on a Jacobian-Free Newton-GMRES (m) method

which requires only exchanges of states of boundary buses among different sub-areas. Therefore

it has strong scalability in distributed computing environments built on heterogeneous computing resources

such as the power grid system. Moreover

several accelerating methods are developed to enhance its efficiency

including continuously preconditioning with self-adaptive preconditioners

predicting boundary conditions and multi-step scheduling. Tests on an IEEE standard system are performed. The results show that these accelerating methods can enhance the convergence rate of the proposed algorithm greatly and reduce communication costs remarkably

which make the new algorithm more feasible and adaptable to highlatency distributed computing environments.