With the development of the social informatization, the demand for optical fiber communication system capacity continues to grow. Channel impairment has always been the key factor limiting the capacity in optical fiber communication system. As the basic function and key technology in optical fiber communication system, channel impairment compensation is constantly improved. Applying deep learning technology in channel impairment compensation of optical fiber communication has attracted the attention of many researchers. This paper introduces the common deep learning models used and the development of deep learning technology for impairments compensation of optical fiber communication systems in recent years.