A Framework for Efficient Paratemporal Simulation

Running and testing highly complex models and simulations has historically been a very computationally intensive and extremely inefficient process, since full simulations are run from start to finish for each realization of a set of stochastic variables until a full distribution of simulation outcomes has been established. Recently, the concept of cloning has been used to speed up stochastic simulations by saving out simulation state and branch probability data at certain common decision points, so that the simulation can be restarted using different branch paths and reuse results previously calculated to reduce computation time. Although cloning has shown to provide significant improvements in computation time for low-complexity simulations, the cloning concept by itself is not scalable when dealing with highly complex models or dealing with simulations that require a large number of stochastic realizations to provide full knowledge of a systems’ potential outcomes. In this work we will develop a framework and branch path optimization techniques that are computationally efficient and scalable, provide full knowledge of outcomes, provide quantified uncertainty in decision making, and are versatile to accommodate a wide variety of simulation types while minimizing development and maintenance burden for end-users. Approved for Public Release | 22-MDA-11339 (13 Dec 22)