Tourist flow forecasting is an important research problem in the field of tourism management, which is related to the formulation of tourism policy and the management of tourist attractions. In this paper, a method for predicting the tourist flow of tourist attrac-tions based on the improved marine predator algorithm (MPA) optimization regularized extreme learning machine is proposed. First, in order to adaptively balance the exploration and exploitation status, this paper proposes a MPA based on population diversity and population aggregation, which gives full play to the exploration and exploitation performance of MPA algorithm. The IMPA optimizes the weight and bias of the regularized extreme learning machine (IMPA-RELM), and uses the normalized root mean square error as the fitness function to determine the optimal weight and bias parameters. Finally, the built IMPA-RELM model is applied to the prediction of daily tourist flow in Jiuzhaigou and Chagan Lake scenic spots. The experimental results show that the proposed IMPA-RELM model not only significantly improves the performance of the RELM model, but also has more superior prediction performance and generalization ability compared with the baseline models such as LS-SVM, BPNN and LSTM. Thus, the novel method can provide important reference for the operation and management of scenic spots and the formulation of tourism policies.