Combined CNN-BiLSTM-Att tourism flow prediction based on VMD-MWPE decomposition reconstruction

Predicting tourist flow is crucial for making effective management decisions at scenic spots. To further improve the accuracy of scenic passenger flow prediction, the current study proposed a data preprocessing approach that combines variational modal decomposition (VMD) with multiscale weighted permutation entropy (MWPE). Additionally, a tourist flow prediction method was introduced, utilizing a combination of bi-layer convolution, bidirectional long short-term memory, and an attention mechanism (CNN-BiLSTM-Att). First, the historical passenger flow data was decomposed into multiple intrinsic modal functions using VMD. Second, the IMF subsequence was measured using the MWPE metric, leading to the construction of a new feature matrix. Finally, the CNN-BiLSTM prediction model, incorporating an attention mechanism and optimized using a genetic algorithm, was applied to generate the final prediction results. Using Lushan Scenic Area (China) as a case study, our experimental results demonstrate that the proposed model significantly outperforms benchmark approaches including XGBoost, LSTM and CNN-BiLSTM. The comparative analysis reveals consistent improvements across all evaluation metrics: achieving a minimum 7.46% reduction in the MAPE, at least 7.2% decrease in BRMSE, and a minimum 1.74% enhancement in the R². These statistically significant improvements validate the model's superior predictive capability for tourist flow forecasting. The demonstrated performance, combined with its methodological robustness, suggests strong practical applicability and promising potential for implementation across similar tourism management contexts.