The work aims to take prediction of screw conveying volume as the target, and propose an improved gray wolf algorithm to optimize the conveying volume prediction model of BP neural network with process factors and structural factors as input, to solve quality issue of concrete conveyed by the distributor required for prefabricated components. The conveying process of the distributor was simulated by EDEM, and the conveying rate and the average mass flow rate were used as two evaluation standards of orthogonal experiment. The range method and matrix analysis method were used to obtain the effect order of various distributor factors on the two standards. The methods of chaotic population initialization, step size update formula and adaptive convergence factor were used to improve the slow convergence speed of gray wolf algorithm and the limitation to local optimal solution. The weights and thresholds of BP neural network were optimized through the improved algorithm. For the optimized conveying volume prediction, the average absolute percentage error was 8%, and the coefficient of determination is 0.95, which was smaller than the predicted value error of the conveying volume model of other experimental comparison groups. The study can provide concrete setting target values for the quantitative distribution of prefabricated components.