Parallel Dynamic Step Size Sphere-Gap Transferring Algorithm for Solving Conditional Nonlinear Optimal Perturbation

Intelligent algorithms have been extensively applied in scientific computing. Recently, some researchers apply intelligent algorithms to solve conditional nonlinear optimal perturbation (CNOP) which is proposed to study the predictability of numerical weather and climate prediction. The difficulty of solving CNOP using the intelligent algorithm is the high dimensionality of complex numerical models. Therefore, previous researches either are just tested in ideal models or have low time efficiency in complex numerical models which limited the application of CNOP. In this paper, we proposed a parallel dynamic step size sphere-gap transferring algorithm (DSGT) to solve CNOP in complex numerical models. A dynamic step size factor is also designed to speed up convergence of sphere-gap transferring algorithm. Through the singular value decomposition, the original problem is reduced into a low-dimensional space to hunt the coordinate of the optimal CNOP with the DSGT algorithm. Moreover, in order to accelerate the computation speed, we parallelize the DSGT method with MPI technology. To demonstrate the validity, the proposed method has been studied in the Zebiak-Cane model to solve the CNOP. Experimental results prove that the proposed method can efficiently and stably obtain a satisfactory CNOP, and the parallel version can reach the speedup of 7.18 times with 10 cores.