Research On Distributed-parallel Simulation Algorithms Of Continuous System Model

With the application of Real-Time Simulation becoming more and more popular, more and more complex system model involving in, so it needs much more CPU resources to deel with the High-Speed Computation. The Real-Time simulation based on multi-node that utilizes the local area network technology grow up, in which the node could be either homogeny or isomer. To fully exert computation ability of every node in the system,mean while achieving perfect speed-up and efficiency, before starting simulation,we first divide model into several parts and then distributing them into different nodes in the system, confirm the communication mode and parallel character between nodes. There exists so many High-order mega systems in engineering. Transformming High-order systems into differential equations and then solve them with numerical solution is a common method.Base on implicit Runge-Kutta scheme, this paper lucubrate inner parallel character of Runge-Kutta scheme by two ways and point out two conclusions:(1) Frequent communication between state variables is a key factor to restrict speed-up of parallel distributing simulation.(2) Partition granularity can not be too small while using parallel distributing simulation mode. Create three methods about constructing parallel integral arithmetic for continuous system after analyzing inner parallel characters base on multi-step method and single-step method.Having designed and realized general Runge-Kutta parallel integral engine, parallel integral engine optimized the communication efficiency under the management by itself. Having designed and realized the parallel distributing simulation method based on Simulink and object-oriented simulation.Improving the parallel distributing simulation efficiency by using Memory block, analyzing system error produced by using Memory block and establish math model of error. Analyzing the applicability of using different compensation methods based on state variables with different characters. Concluding compensation methods base on simulation experiments.Concluding Coordinate projection method based on a kind of systems which obey conservation law and time-invariant law.Coordinate projection method find out error speading rules between 1()nX t?,n1X??,n1X? using linearization stage by stage. Analyzing distributed simulation applicability of these systems.Finally, based on nosie usually appeared in sampling stage of Real-time semi-physical simulation, Realized gauss noise function after analyzing their characters. Testing the antinoise ability of lsqe method and lagra method by simulation...