The invention discloses a neural network machine translation model based on iterative bidirectional migration, and relates to the natural language processing. By establishing the multi-time two-way migration between a source field and a target field and utilizing the effective interaction of the two fields and the beneficial flow of the related knowledge, the model capabilities of different fieldsare continuously improved, so that a better translation effect is achieved. The model is popularized from the one-to-one field migration to the many-to-one field migration, and an effective solutionto the migration sequence problem from different source fields to the target field in the many-to-one field migration is provided, so that the model can more fully utilize the corpus resources of multiple fields. During the training process, a knowledge distillation method is used for more effectively guiding the convergence of the model, the problems of catastrophic forgetting and knowledge sparseness are avoided, and the win-win of the translation models in two fields is achieved.