In order to realize the accurate prediction of the inventory quantity of prefabricated components, a prediction model of the inventory quantity of prefabricated components based on improved Harris Hawk algorithm (HHO) optimized gated recurrent network (GRU) is proposed. The two-stage population initialization strategy of good point set and quantum computing, the nonlinear prey escape energy control strategy, the exploration stage strategy of quasi-reverse learning and slime mold algorithm, and the average differential mutation strategy are introduced to improve HHO. The test function verifies that the improved SDMQHHO optimization ability and local optimal escape ability are improved. By using SDMQHHO to find the optimal hyperparameter combination of GRU model to form SDMQHHO-GRU prediction model, SDMQHHO-GRU, HHO-GRU, GRU and LSTM are compared on the same data set. The coefficient of determination ( \(R^{2}\) ) of this model is at least 7.69% higher than that of other models, and the root mean square error (RMSE), mean absolute percentage error (MAPE) and mean absolute error (MAE) are reduced by at least 46.55%, 45.16% and 40.70% respectively.