为了进一步提高景区客流量的预测精度，提出一种基于变分模态分解（variational mode decomposition，VMD）结合多尺度加权排列熵（multiscale weighted permutation entropy，MWPE）的数据预处理模型，以及通过双层卷积、双向长短期记忆网络及注意力机制的组合（deep convolutional neural network-bidirectional long short-term memory-attention mechanism，DCNN-BiLSTM-Att）客流量预测方法。为检验方法的有效性，本文以中国庐山景区为例对方法进行实验测试。实验结果表明，与DNN、LSTM、XGBoost、BiLSTM、DCNN-LSTM、DCNN-BiLSTM等传统模型相比，基于VMD-MWPE的DCNN-BiLSTM-Att模型对旅游景区客流量的预测具有更好的准确性、鲁棒性和泛化能力。

To further improve the prediction accuracy of tourist flow in scenic spots， a model based on the data preprocessing of variational mode decomposition （VMD） combined with multi-scale weighted permutation entropy （MWPE） was proposed， through the combination of double-layer convolution， bidirectional long short-term memory network and attention mechanism （DCNN-BiLSTM-Att）. Lushan Scenic spot in China was taken as an example to test the effectiveness of the method. The experimental results show that compared with traditional models such as DNN， LSTM， XGBoost， BiLSTM， DCNN-LSTM， and DCNN-BiLSTM， the DCNN-BiLSTM-Att model based on VMD-MWPE has better accuracy， robustness and generalization ability for tourist flow prediction in scenic spots.