During real-time hybrid simulation(RTHS) of nonlinear specimens, the interaction between the specimen and the loading system can lead to variations in the specimen's behavior, consequently affecting the time delay in the servo system. Online estimation of the system's time delay enables the application of an adaptive time-delay compensation method for controlling time-varying systems. Nevertheless, during the initial stages of parameter identification, the estimated values frequently exhibit notable fluctuations, which can have a detrimental impact on the effectiveness of control. To this end, a two-stage adaptive time-delay compensation method for RTHS based on the inverse model is proposed, that is: first, the inverse model controller of the system is used to perform coarse compensation to eliminate the test error caused by the main time delay; and then the adaptive delay compensation method based on recursive least squares is used to compensate the remaining delay to further control the accuracy. Using the two-layer shear frame as the prototype and the self-centering viscous dampers as the specimens, a time-delay compensation RTHS was carried out simultaneously on the two experimental substructures. Numerical simulations and experimental results show that the control accuracy of the proposed method is higher than that of the single-stage time-delay compensation method, and it can be applied to RTHS involving multiple experimental substructures.