In this paper, two efficient accelerating schemes for decreasing the simulation time of the substrate coupling modelling and analysis are proposed. The first method is based on partitioning the substrate to two parts, including two different boundary conditions, so that to apply FDM for simulation of two parts independently. This partitioning is performed using a MLP neural network to approximate the potential of the middle surface of the substrate. Using parallel processing reduces the simulation time by 57 percent for 50 simulation points. The second method is based on applying weights to the results of each stage of Multi-Grid processing. Then the results of the two previous stages are used for obtaining the magnitudes of the new stage variables. We show that, this method with the best weight factors, results in 75% reduction in the simulation time for the substrate modeling compared to the simple Multi-Grid method.