Model-based prosthetic control strategies are still incompetent for practical implementations because modeling uncertainties and errors limit their precision. This paper investigates identifying the dynamic parameters of ankle-knee prostheses based on Coulomb-viscous friction for a newly designed powered prosthesis in the laboratory. Firstly, a powered ankle-knee prosthesis dynamic model, considering the fixed transmission ratio for the knee joint and the nonlinear transmission ratio for the ankle joint, is developed based on the Lagrange method. Secondly, a Coulomb-viscous friction model is employed to describe the joint friction characteristics in the prosthesis dynamic model. Finally, the dynamic parameters of the powered prosthesis are identified by the particle swarm optimization algorithm. Compared with CAD estimation, the root-mean-square errors between the actual and the reconstructed torques after parameter identification are reduced by 99.07% for the knee joint and 83.33% for the ankle joint, indicating that the model precision has been significantly enhanced. The effectiveness, as a consequence, provides a solid technical foundation for accurate prosthesis control.