The ammonothermal method is considered the most promising method of fabricating bulk gallium nitride (GaN) crystals. This paper improves the ammonothermal growth model by replacing the heater-long fixed temperature boundary with two resistance heaters and considering the real thermal boundary outside the shell. The relationship between power values and temperatures of dissolution and crystallization is expressed by the backpropagation (BP) neural network, and the optimal power values for specific systems are found using the non-dominated sorting genetic algorithm (NSGAII). Simulation results show that there are several discrepancies between updated and simplified models. It is necessary to build an ammonothermal system model with resistance heaters as a heat source. Then large-sized GaN crystal growth is analyzed based on the well-developed numerical model. According to the simulation results, both the increasing rate and maximum stable values of the metastable GaN concentration gradient are reduced for a larger-sized system, which is caused by the inhomogeneity of heat transfer in the autoclave.