1: \begin{abstract}

2: Matrix splitting iteration methods play a vital role in solving large sparse linear systems.

3: Their performance heavily depends on the splitting parameters, however, 

4: the approach of selecting optimal splitting parameters has not been well developed.

5: In this paper, we present a multitask kernel-learning parameter prediction method 

6: to automatically obtain relatively optimal splitting parameters, which contains simultaneous multiple parameters prediction and a data-driven kernel learning.

7: For solving time-dependent linear systems, including 

8: linear differential systems and linear matrix systems, 

9: we give a new matrix splitting Kronecker product method, as well as its convergence analysis and preconditioning strategy. 

10: Numerical results illustrate our methods can save an enormous amount of time in selecting the relatively optimal splitting parameters compared with the exists methods. 

11: Moreover, our iteration method as a preconditioner can effectively accelerate GMRES. 

12: As the dimension of systems increases, all the advantages of our approaches becomes significantly.

13: Especially, for solving the differential Sylvester matrix equation, the speedup ratio can reach tens to hundreds of times when the scale of the system is larger than one hundred thousand. 

14: \end{abstract}

15: