To solve the engineering problem of power loss in wet brakes under non-braking conditions, taking into consideration the influence of the lubricating oil in the clearance of friction pairs on the drag torque of the friction pair, making use of the strong nonlinear fitting ability of the sparrow-search-algorithm-back propagation (SSA-BP) neural network, and taking the no-load operating condition of wet brakes as the input variable and the drag torque as the output variable, an approximate model of wet brakes is established. Compared with the traditional BP model, the prediction accuracy is obviously improved, which can meet the needs of practical engineering. In order to obtain the minimum drag torque, the working parameters are searched and optimized through chicken swarm optimization （CSO） intelligent algorithm, and the best working condition of the wet brake is obtained. The experimental results show that the drag torque and power loss between the friction pairs after optimization are significantly lower than those before optimization. This study provides theoretical research basis and engineering practice experience for the further optimization of wet brake structure.