Firstly, this paper improves the standard Fruit Fly Optimization Algorithm (FOA) by adaptively adjusting the number of the fruit fly populations and the search step size and optimizing the initial iteration position, so as to improve the local search ability and search efficiency of the algorithm. Through combining the optimized FOA (AFOA) and Echo State Network (ESN), a two-stage combined prediction model (AFOA-ESN) is proposed. The AFOA optimizes the ESN to obtain its key parameters, which are inputted into the ESN to form the final combined prediction model. Finally, this model is used to predict tourism demand. The experimental results show that the AFOA-ESN model has higher prediction accuracy than the ARIMA, SVM, BPNN, standard ESN and other models.