An advanced hyperbolic divergence cleaning scheme based on "generalized Lagrange multiplier" (GLM) for the equations of "shallow water magnetohydrodynamics" (SMHD) is presented. This scheme is based on the two-step method which is comprised of the standard finite-volume updating step for the nonlinear genuine SMHD system and the divergence cleaning step for the linear GLM-based Maxwell subsystem. The divergence cleaning step can be applied several times per each computational time step, in order to accelerate the transports of the divergence error out of the computational domain. The presented two-step method is compared with the standard GLM method based on operator splitting. It is shown that the standard operator-splitting based method has the shock dissipation problem, particularly when the multiple subcycles of the divergence cleaning step is performed per each time step. On the contrary, the introduction of the multiple subcycles for the new GLM–Maxwell subsystem does not suffer from the dissipation of the shocks, and produces better shock resolution. The presented method can be further applied to the full magnetohydrodynamics equations.