Deep Embedded Clustering for Tourist Flow Analysis and Planning Optimization

The distribution of passenger flow shows strong spatiotemporal characteristics, especially during holidays or emergencies, the pattern of passenger flow will undergo significant changes. Traditional prediction methods often rely on static historical data and cannot adapt well to rapidly changing environments. In view of the problems in existing tourist flow analysis, such as insufficient ability to process high-dimensional complex data and unstructured information and difficulty in dynamically capturing the changing patterns of tourist flow, this paper takes tourist attractions in Hainan as the object and applies deep embedded clustering (DEC) to tourist flow analysis and planning and design optimization, aiming to improve the accuracy of tourist flow prediction in tourist attractions and the scientific characteristic of planning decisions. First, the features are initialized using variational autoencoder (VAE), which generates a better adaptive embedding space by modeling the probability distribution of latent variables. Then, a deep network structure is used to extract features, and soft labels and Kullback–Leibler (KL) divergence are adopted as optimization objectives to gradually optimize the data distribution in the embedding space. Finally, combined with the multitask learning framework, the model can perform prediction and clustering tasks simultaneously. The tourist flow data of tourist attractions are clustered at three levels to identify the feature patterns of high-peak, medium-peak, and low-peak. Experimental results show that compared with advanced methods such as autoencoder-based clustering (AEC), density peaks clustering (DPC), and deep K-means clustering, the average accuracy of the DEC model is improved by 8.89%, 5.18%, and 3.23%, respectively. It can be concluded that DEC can help to accurately analyze the tourist flow pattern of tourist attractions and provide new perspectives and support for the optimization of tourism planning and design.