In order to more accurately predict the demand for shared bicycles, and formulate a reasonable scheduling optimization plan. In view of the periodicity, nonlinearity and randomness of the shared bicycle riding data, the seasonal grey Markov model was proposed to predict the demand for shared bicycles. On this basis, the scheduling optimization plan is formulated according to the results of the bilevel programming model. In the seasonal grey Markov model, the original data was firstly brought into the seasonal GM(1,1) model to obtain the prediction results, and then the Markov model was used to correct the predicted residuals to obtain the final predicted value. In the bilevel programming model, the upper layer target is the dispatch cost of the operator, and the lower layer target is the dispatch time of the dispatch center. The bilevel programming model was solved by the GUROBI solver. Finally, the two models were applied to the analysis of 17 Citi Bike shared bike stations in New York City. The numerical results show that the mean absolute percentage error (MAPE) of the demand forecast from Monday to Friday for the seasonal grey Markov model at 17 sites is 10.68%, which means that the forecast accuracy is relatively high. The scheduling optimization plan based on the bilevel programming model can determine the number, location, scheduling range and scheduling path of the scheduling center, which can optimize the scheduling cost and scheduling time while meeting user needs. The demand forecasting model and scheduling optimization program proposed by the research can provide effective reference for shared bicycle operation departments and traffic management departments.