This work deals with scheduling of the continuous casting of steel making, i.e.determination of the number of fixed capacity pots of each grade of steel and the charge sequence in each casting machine. Since a huge number of binary variables make the full-space model mixed-integer linear programming model computationally intractable, we propose a two-level algorithm. At the top level, we solve the planning problem which determines the number of pots of each grade for every planning period by solving the relaxed mixed integer linear model. At the lower level, the scheduling problem is solved by an algorithm which combines ideas from parallel simulated annealing and shuffled frog-leaping algorithm. Real-world steel-plate production data are utilized to demonstrate the effectiveness of the proposed algorithm.