<p>The variable stiffness exoskeleton for sit-to-stand assistance is a device that enables individuals with lower limb impairments to perform the transition. However, its complex mechanism makes it difficult to accurately obtain dynamic parameters. To resolve this challenge, this paper establishes a dynamic model of the system and proposes a parameter identification method based on a triple-loop iteration. First, the dynamic model of the exoskeleton is established. This model describes the dynamic relationship between the collective motor torque of each joint, the exoskeleton’s motion and friction torques, and it exhibits strong nonlinearity. Some motions in the mechanism exhibit monotonic correlation characteristics. Therefore, their resulting friction torques can be simplified to reduce dynamic parameters. Subsequently, the dynamic parameters are identified through triple-loop iteration. Two inner loops identify the linear and nonlinear parameters, respectively. The outer loop coordinates their interaction until the results converge to the optimal dynamic parameters. Additionally, backpropagation neural network (BPNN) is introduced to improve the accuracy of model-based torque prediction. Finally, the proposed method is evaluated by experiments. The results demonstrate that the triple-loop iterative method exhibits better parameter identification accuracy compared to the existing method, and the BPNN effectively enhances torque prediction accuracy.</p>