Abstract: In order to balance the accuracy and efficiency of neutron transport simulation, performing self-adaptive mesh refinement and domain decomposition parallelization for unstructured meshes are technologies that worth exploring. For the neutron flux distribution of each energy group generally has significant difference with distributions of other groups, using a common mesh in the simulation for all energy groups cannot best match the flux spatial profiles of every group. In that case, the common mesh has to be excessively globally refined, or the accuracy is expected to lose some extent. An energy-group wise local mesh refinement scheme for multi-group neutron transport simulation was adopted in this paper, which was based on independent posterior error estimations for the neutron flux of all energy groups. The obtained unstructured fine meshes were stored separately in multiple containers, they are different on the active top level, but share a common coarse mesh on the base level where the fine meshes are derived from. These multi-sets of hierarchical meshes were named base-derived meshes hereafter. Obviously, these base-derived meshes can best match the flux distribution of each energy group respectively, but the data transfer and interpolation between groups for calculation of inter-group fission and scattering sources come to be the key issues. A couple algorithm for multi-group transport simulation on such group-wise refined base-derived meshes was introduced. It achieves high accuracy and efficiency by successive adaptive refinement cycles, and finally reaches the mesh insensitive state automatically regardless the resolution of initial coarse base mesh. The transport code ENTER-Ⅱ was developed based on this energy-group wise self-adaptive (GWA) refinement algorithm, and the discontinuous finite element method was applied to enable the flexible modeling and meshing of complex geometry. Domain decomposition based parallel computing capability was realized with the help of the open-source library DEAL.Ⅱ. Preliminary verifications to the group-wise self-adaptive refinement algorithm were presented, which show that the simulation accuracy is acceptable, and the time efficiency is remarkably better. The time costs in solving angular flux on the successively refined meshes are quite low, due to the good estimations of initial values of flux that are transferred from the solutions on the coarser meshes and then interpolated on the fine meshes. It concludes that, with the help of energy-group wise self-adaptive refinement, the initial meshing of geometry domain does not need to rely on expertise, and the high-fidelity simulation efficiency raises effectively.