Abstract：With ever enlarging scale of the underground cavern group，recognition of relevant parameters and scheme optimization would be highly nonlinear and of multi-extreme values in the solution space. It is much in need of finding an available method of global optimization and parallel computation. So a new parallel evolutionary neural network FEM is put forward. Through the sample construction by FEM calculating，the mapping relationship among the calculating schemes，the maximum displacement and the volume of damage zone is set up by parallel evolutionary neural network，and a group of initial feasible schemes are given by genetic algorithms(GAs). Evaluated by the maximum displacement of key spots and volume of damage zone，a group of new schemes are generated by operation of GAs. The operation is done until the reasonable scheme is found. The parallel computation is carried out on independently developed parallel environment(RsmVPC) based on WINDOWS platform. The methodology makes it possible to solve the large scale optimization problems parallelly on PC machine groups，and to improve the computing speed，scale and precision to a large extent with the proposed method.
