Lu Wenxing, Rui Haidong, Liang Changyong, Jiang Li, Zhao Shuping, Li Keqing
School of Management, Hefei University of Technology, Hefei 230009, China.
Ministry of Education Key Laboratory of Process Optimization and Intelligent Decision-Making, Hefei University of Technology, Hefei 230009, China.
Entropy (Basel). 2020 Feb 25;22(3):261. doi: 10.3390/e22030261.
Accurate tourist flow prediction is key to ensuring the normal operation of popular scenic spots. However, one single model cannot effectively grasp the characteristics of the data and make accurate predictions because of the strong nonlinear characteristics of daily tourist flow data. Accordingly, this study predicts daily tourist flow in Huangshan Scenic Spot in China. A prediction method (GA-CNN-LSTM) which combines convolutional neural network (CNN) and long-short-term memory network (LSTM) and optimized by genetic algorithm (GA) is established. First, network search data, meteorological data, and other data are constructed into continuous feature maps. Then, feature vectors are extracted by convolutional neural network (CNN). Finally, the feature vectors are input into long-short-term memory network (LSTM) in time series for prediction. Moreover, GA is used to scientifically select the number of neurons in the CNN-LSTM model. Data is preprocessed and normalized before prediction. The accuracy of GA-CNN-LSTM is evaluated using mean absolute percentage error (MAPE), mean absolute error (MAE), Pearson correlation coefficient and index of agreement (IA). For a fair comparison, GA-CNN-LSTM model is compared with CNN-LSTM, LSTM, CNN and the back propagation neural network (BP). The experimental results show that GA-CNN-LSTM model is approximately 8.22% higher than CNN-LSTM on the performance of MAPE.
准确的游客流量预测是确保热门景区正常运营的关键。然而，由于日游客流量数据具有很强的非线性特征，单一模型无法有效地把握数据特征并做出准确预测。因此，本研究对中国黄山风景区的日游客流量进行预测。建立了一种结合卷积神经网络（CNN）和长短期记忆网络（LSTM）并通过遗传算法（GA）优化的预测方法（GA-CNN-LSTM）。首先，将网络搜索数据、气象数据等数据构建成连续特征图。然后，通过卷积神经网络（CNN）提取特征向量。最后，将特征向量按时间序列输入长短期记忆网络（LSTM）进行预测。此外，使用遗传算法科学地选择CNN-LSTM模型中的神经元数量。在预测前对数据进行预处理和归一化。使用平均绝对百分比误差（MAPE）、平均绝对误差（MAE）、皮尔逊相关系数和一致性指数（IA）评估GA-CNN-LSTM的准确性。为了进行公平比较，将GA-CNN-LSTM模型与CNN-LSTM、LSTM、CNN和反向传播神经网络（BP）进行比较。实验结果表明，GA-CNN-LSTM模型在MAPE性能上比CNN-LSTM高出约8.22%。
