Abstract: Soft foundation sluice is vulnerable to floor void due to water erosion and other factors,which greatly threatens sluice safety.To solve the detection of soft foundation sluice floor void,the dynamic detection method was proposed based on “Gate Recurrent Unit (GRU) neural network surrogate model and multi-objective gray wolf optimal algorithm (MOGWO)”.The surrogate model was constructed based on GRU neural network to characterize the nonlinear relationship between floor void parameters and modal parameters of soft foundation sluice; The multi-objective optimization function was established for void parameters inversion based on the intrinsic frequency and normalized vibration mode.The MOGWO was used to solve the multi-objective optimization problem of void parameters inversion.The proposed method was applied to the inversion of two void conditions of an indoor physical soft foundation sluice model.The GRU surrogate model accuracy is better than the Back Propagation (BP) surrogate model and the polynomial response surface model.The relative errors between the inverse void area and actual void area of the model are 6.76% and 5.58% for the two void conditions.The results indicate that the proposed method is superior to the single-objective inversion method.