The invention discloses a dynamics-based velocity model assisted underwater intelligent navigation method. The method includes the following steps that: (1) on-board sensor information is periodicallycollected; (2) the input variables and output variables of a velocity model to be established are determined; (3) the velocity model is constructed and trained; and (4) when it is detected that a DVL(Doppler Velocity Log) malfunctions or the data of the DVL fail, navigation analysis is performed with the output velocity of the trained model used as a bottom velocity, and therefore, model-assisted navigation can be performed. According to the method of the invention, a new AUV (Autonomous Underwater Vehicle) velocity model is put forward based on the idea of dynamic models; the influence of factors such as rudder angles of rudder blades, and heading angles is considered; as long as the fault data of the DVL are not detected during an operation process, training sets can be increased at any time to train the model; after the fault data of the DVL are detected, the output velocity of the model can be adopted instead, and therefore, the problem of navigation error increase caused by thefailure or fault of the DVL can be avoided; the redundancy method of velocity sensors is provided with no hardware costs increased; the method has good system robustness and can guarantee high-precision navigation.