A tourism demand prediction method based on two-stage feature selection and Time2Vec

Abstract

The present application relates to a kind of tourism demand prediction method based on two-stage feature selection and Time2Vec, belong to tourism data prediction field.The method, two-stage feature selection method based on MCC and ReliefF algorithm is designed, to select the most effective feature subset from data set;Proposed a kind of MCNN-BiLSTM model for predicting tourism demand, make full use of the information in the middle layer, so as to more effectively extract important features from complex passenger flow related time series data;Introduce Time2Vec, integrate the processing process of time series into MCNN-BiLSTM, i.e.T2V-MCNN-BiLSTM, so as to improve the signal processing algorithm commonly used in tourism demand prediction, so that the model can efficiently extract advanced time features, predict tourism demand, so as to improve the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of tourism data prediction, and particularly relates to a tourism demand prediction method based on two-stage feature selection and Time2Vec. BACKGROUND

[0002] Tourism demand prediction is of great significance for tourism management departments and related industry development [1] . Accurate demand prediction can provide scientific basis and policy guidance for strategic development, investment decision-making, revenue management and pricing strategies of tourism and related industries, thereby promoting the contribution of tourism to economic growth [2,3] . However, making accurate tourism demand prediction is a difficult task [4,5] , because visitor flow is usually accompanied by seasonal and unpredictable changes [6] . In addition, the impact of unexpected events on people's travel will also have a serious impact on the tourism industry [7,8] . Therefore, the complexity of visitor flow data and the constant changes in the external environment make it a major challenge to achieve reliable prediction.

[0003] Existing research can be mainly divided into three categories:

[0004] (1) Time series model, which predicts tourism demand based on the historical pattern of tourism demand [9,10] ;

[0005] (2) Econometric model, which focuses on establishing the causal relationship between tourism demand and various explanatory variables [11,12] ;

[0006] (3) Machine learning and deep learning model, which achieves more accurate prediction by capturing potential patterns in high-dimensional feature space [13,14] .

[0007] The main shortcomings of existing technologies are:

[0008] (1) Existing tourism demand prediction models fail to effectively handle the uncertainty and complexity of external variables. Specifically, the ReliefF method commonly used for feature selection only removes redundant features by deleting features with small weights, which may result in insufficient filtering of irrelevant features.

[0009] (2) Although the signal processing method for processing time series data improves prediction accuracy by reducing complexity, the decomposition result is unstable, increasing the complexity and uncertainty of data processing. In addition, separating the time series data decomposition and prediction processes may lose important potential information during the decomposition process, affecting the ability of deep learning models to extract features subsequently.

[0010] (3) Existing tourism demand prediction models do not make full use of sample information during the training phase. That is, the CNN-LSTM hybrid model combines the data in a series manner, ignoring the large amount of useful information contained in the intermediate layers of the neural network.

[0011] References:

[0012] [1]Law, R., Li, G., Fong, DKC, & Han,

[0013] [2]Kaya,K., Y.,Yaslan,Y., G.,& F.(2022).Demandforecasting model using hotel clustering findings for hospitalityindustry.Information Processing&Management,59(1),102816.https: / / doi.org / 10.1016 / j.ipm.2021.102816.

[0014] [3]Wu,B.,Wang,L.,&Zeng,Y.-R.(2023).Interpretable tourism demandforecasting with temporal fusion transformers amid COVID-19.AppliedIntelligence,53(11),14493–14514.https: / / doi.org / 10.1007 / s10489-022-04254-0.

[0015] [4] Silva, E. S., Hassani, H., Heravi, S., & Huang, X. (2019). Forecasting tourism demand with denoised neural networks. Annals of Tourism Research, 74, 134-154. https: / / doi.org / 10.1016 / j.annals.2018.11.006.

[0016] [5] Song, H., Qiu, R. T., & Park, J. (2019). A review of research on tourism demand forecasting: Launching the Annals of Tourism Research Curated Collection on tourism demand forecasting. Annals of Tourism Research, 75, 338-362. https: / / doi.org / 10.1016 / j.annals.2018.12.001.

[0017] [6] Zhang, C., & Tian, Y. X. (2022). Forecast daily tourist volumes during the epidemic period using COVID-19 data, search engine data and weather data. Expert Systems with Applications, 210, Article 118505. https: / / doi.org / 10.1016 / j.eswa.2022.118505.

[0018] [7] M., Soriano, D. R., & Porada-Rochoń, M. (2021). Impact of COVID-19 on the travel and tourism industry. Technological forecasting and social change, 163, 120469. https: / / doi.org / 10.1016 / j.techfore.2020.120469.

[0019] [8] Xu, S., Liu, Y., & Jin, C. (2023). Forecasting daily tourism demand with multiple factors. Annals of Tourism Research. https: / / doi.org / 10.1016 / j.annals.2023.103675.

[0020] [9] Song, H., Qiu, R. T., & Park, J. (2019). A review of research on tourism demand forecasting: Launching the Annals of Tourism Research Curated Collection on tourism demand forecasting. Annals of Tourism Research, 75, 338-362. https: / / doi.org / 10.101.

[0021]

[10] Park, S., Lee, J., & Song, W. (2017). Short-term forecasting of Japanese tourist inflow to South Korea using google trends data. Journal of Travel & Tourism Marketing, 34(3), 357-368. https: / / doi.org / 10.1080 / 10548408.2016.11706516 / j.annals.2018.12.001.

[0022]

[11] Cao, Z., Li, G., & Song, H. (2017). Modelling the interdependence of tourism demand: The global vector autoregressive approach. Annals of Tourism Research, 1-13. https: / / doi.org / 10.1016 / j.annals.2017.07.019.

[0023]

[12] Li, G., Wong, K. K. F., Song, H., & Witt, S. F. (2006). Tourism Demand Forecasting: A Time Varying Parameter Error Correction Model. Journal of Travel Research, 45(2), 175-185. https: / / doi.org / 10.1177 / 0047287506291596.

[0024]

[13] Polyzos, S., Samitas, A., & Spyridou, A. Ef. (2021). Tourism demand and the COVID-19 pandemic: an LSTM approach. Tourism Recreation Research, 175-187. https: / / doi.org / 10.1080 / 02508281.2020.1777053.

[0025]

[14] Zhang, C., & Tian, Y. X. (2022). Forecast daily tourist volumes during the epidemic period using COVID-19 data, search engine data and weather data. Expert Systems with Applications, 210, Article 118505. https: / / doi.org / 10.1016 / j.eswa.2022.118505. SUMMARY

[0026] The present application aims to overcome the shortcomings of the prior art and provide a tourism demand prediction method based on two-stage feature selection and Time2Vec. A two-stage feature selection method based on MCC and ReliefF algorithm is designed to select the most effective feature subset from the data set. An MCNN-BiLSTM model for predicting tourism demand is proposed, which fully utilizes the intermediate layer information, thereby more effectively extracting important features from complex tourist flow related time series data. Time2Vec is introduced, which integrates the time series processing process into MCNN-BiLSTM, i.e. T2V-MCNN-BiLSTM, thereby improving the commonly used signal processing algorithm in tourism demand prediction, enabling the model to efficiently extract high-level time features and predict tourism demand, thereby improving prediction accuracy.

[0027] To achieve the above object, the technical scheme of the present application is: a tourism demand prediction method based on two-stage feature selection and Time2Vec, comprising:

[0028] A two-stage feature selection method based on MCC and ReliefF algorithm is designed to select a key feature subset from the data set;

[0029] An MCNN-BiLSTM model for predicting tourism demand is proposed, and Time2Vec is introduced to integrate the processing process of time series into the MCNN-BiLSTM model to form a T2V-MCNN-BiLSTM model, and based on the key feature subset, the tourism demand is predicted.

[0030] In an embodiment of the present application, the data set is obtained as follows:

[0031] (1) Collect historical passenger flow data: the historical passenger flow of the scenic spot is collected from the official website of the scenic spot;

[0032] (2) Collect Baidu index data: use Python programming to obtain the Baidu index data of the scenic spot related search keywords, the strategy of selecting keywords is to use the central word related to the tourist attraction, and other related keywords or phrases are introduced through the "related search" strategy provided by the Baidu index data platform;

[0033] (3) Collect emergency data: collect the daily increase of emergency related indicators;

[0034] (4) Collect other data: including weather data, holiday data, among which the weather data includes weather condition and temperature, and is collected from weather network.

[0035] In an embodiment of the present application, before use, the data set needs to eliminate repeated invalid data in the data set, and the interpolation method is used to complete the data, and then the normalization processing is performed.

[0036] In an embodiment of the present application, the two-stage feature selection method based on MCC and ReliefF algorithm selects the key feature subset from the data set in the following manner: the MIC of each variable 0-n order is calculated to determine the optimal lag structure, the maximum lag order is set to 30, the time series is arranged according to the lag order, and the key features are determined by using the two-stage feature selection method to obtain the key feature subset.

[0037] In an embodiment of the present application, the implementation of predicting tourism demand based on the key feature subset by the T2V-MCNN-BiLSTM model is as follows:

[0038] (1) Extracting time embedding features: Time2Vec is used to automatically capture periodic and non-periodic time features in the time series of the key feature subset related to tourism demand; after obtaining the time embedding, it is connected with the original feature data in the key feature subset as the input of the MCNN-BiLSTM model;

[0039] (2) Extracting time and variable interaction features: in the MCNN-BiLSTM model, CNN is used to extract features with high information content and high correlation in all input vectors, BiLSTM is used to learn the features extracted by CNN to obtain more context information, and a connection function is introduced to fuse the features extracted by CNN and BiLSTM;

[0040] (3) Output prediction results: the fused features are processed through a fully connected layer to obtain the final prediction results.

[0041] In an embodiment of the present application, the two-stage feature selection method is as follows:

[0042] In the first stage, MCC(θ) is used to calculate the amount of redundant information between each feature in the key feature subset and other features to filter out some redundant features; a self-defined threshold β is used to screen the features, and when θ is greater than β, the feature is retained; the mth feature value of the nth sample is represented by f nm , and the mean value of the ith feature is defined as The covariance v ij and the correlation coefficient r ij are defined as:

[0043]

[0044] The covariance matrix and the correlation coefficient matrix are defined as V={v ij} and R={r ij}, wherein i,j∈(1,2,…,m);

[0045] MCC is calculated based on the correlation coefficient matrix R:

[0046]

[0047] Where θ i is the MCC between the ith feature and other features, r i , r i ' and R i-1 represent the correlation coefficient matrix of the ith feature, the transpose of r i , and the correlation coefficient matrix of other features except the ith feature, respectively;

[0048] The second stage is to determine the final features based on the features screened in the first stage by using a ReliefF algorithm;

[0049] First, a sample S is randomly selected, and q same-type neighbors H nearest to the sample S are selected q ={h1, h2, h3,..., h q}; in addition, q nearest different-class neighbors G q ={g1, g2, g3,..., g q} of the sample S are found, the type of Gq is C, and the weight of each feature F is calculated as follows:

[0050]

[0051] wherein diff(F, S, s) represents the difference between the sample S and s in the feature F, S(F) and s(F) are the numerical values of the sample S and s in the feature F, P(C) and P(class(P)) are the prior probabilities of the type C and the type of the randomly selected sample, i.e., the proportion of the sample of the type in the total number of samples, max(F) and min(F) are the maximum value and the minimum value of the feature F, and m is the iteration number.

[0052] In an embodiment of the present application, the T2V-MCNN-BiLSTM model is implemented as follows:

[0053] (1) Time2Vec

[0054] For a given scalar concept of a time τ, the Time2Vec formula of the i-th element is as follows:

[0055]

[0056] wherein ω i and are learnable parameters, and F is a periodic activation function.

[0057] In general, given a time series, Time2Vec encodes it into a k-dimensional vector.

[0058] (2) MCNN-BiLSTM

[0059] The CNN-BiLSTM model is combined by combining CNN and BiLSTM; first, CNN can extract more abstract and information-rich feature representations from the original time series data; second, BiLSTM as a lower component in the CNN-BiLSTM model receives the time information of the important features of the tourism demand prediction obtained by the CNN; BiLSTM combines the information flow in two directions, i.e., forward and backward, allows the model to consider past and future information while predicting each time step, thereby more comprehensively understanding the time relationship in the sequence, reducing information loss and overcoming the gradient disappearance problem in long sequences; the outputs of the two layers of the forward and backward BiLSTM are connected together to form the final output, and the related calculation process is as follows:

[0060]

[0061]

[0062] Wherein, LSTM represents a long short-term memory network, x i is an input vector of an input sequence at a time step i, are the hidden states of the forward LSTM and the backward LSTM at the time step i respectively, [:, :] represents a connection operation of a vector;

[0063] An additional flattening layer is added after each pooling layer of the CNN to fully utilize the intermediate layer information, and the pooling layer selects a maximum pooling layer.

[0064] In an embodiment of the present application, the periodic activation function uses a sine function sin.

[0065] The present application also provides a tourism demand prediction system based on two-stage feature selection and Time2Vec, comprising a memory, a processor and computer program instructions stored on the memory and capable of being executed by the processor, when the processor executes the computer program instructions, the method steps as described above can be realized.

[0066] The present application also provides a computer readable storage medium having computer program instructions capable of being executed by a processor stored thereon, when the processor executes the computer program instructions, the method steps as described above can be realized.

[0067] Compared with the prior art, the present application has the following beneficial effects:

[0068] (1) The present application can better extract useful information from multi-source data by fusing the features extracted by the intermediate layers of CNN and BiLSTM, thereby realizing more accurate prediction.

[0069] (2) The application utilizes a two-stage feature selection strategy to screen a limited number of features, so that the model has stronger stability, and effectively solves the dimension disaster caused by too many external variables, and the experimental results prove the effectiveness of the method.

[0070] (3) The application uses Time2Vec to replace the cumbersome time series data decomposition process, and integrates it into a hybrid deep learning model composed of different deep neural networks. BRIEF DESCRIPTION OF DRAWINGS

[0071] Figure 1 The method model framework of the application.

[0072] Figure 2 The scenic spot tourism demand prediction result of an example of the application. DETAILED DESCRIPTION

[0073] The technical solutions of the application will be specifically described below with reference to the drawings.

[0074] It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the present application. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0075] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form, and in addition, it should be understood that when the terms "comprise" and / or "include" are used in the specification, there is a feature, step, operation, device, component and / or combination thereof.

[0076] The application provides a tourism demand prediction method based on two-stage feature selection and Time2Vec, comprising:

[0077] A two-stage feature selection method based on MCC and ReliefF algorithm is designed to select a key feature subset from the data set;

[0078] An MCNN-BiLSTM model for predicting tourism demand is proposed, and Time2Vec is introduced to integrate the time series processing process into the MCNN-BiLSTM model to form a T2V-MCNN-BiLSTM model, and based on the key feature subset, the tourism demand is predicted.

[0079] The following is the specific implementation process of the application.

[0080] The embodiment provides a tourism demand prediction method based on two-stage feature selection and Time2Vec. The basic framework of the model is as shown in the accompanying Figure 1 , comprising the following steps:

[0081] First, data collection and preprocessing:

[0082] The prediction model utilizes four kinds of data, namely historical passenger flow data, Baidu index data, emergency event data and other data. The data collection method is as follows: (1) collect historical passenger flow data: the historical passenger flow of the scenic spot is collected from the official website of the scenic spot; (2) collect Baidu index data: use Python programming to obtain the Baidu index data of the scenic spot related search keywords. The strategy of selecting keywords is to use the central words related to the tourist attraction, and to introduce other related keywords or phrases through the "related search" strategy provided by the Baidu index platform; (3) collect emergency event data: collect the daily increase of emergency event related indicators through relevant official channels; (4) collect other data: other data includes weather data, holiday data, etc., among which the weather data includes weather condition and temperature, which is collected from weather network (https: / / lishi.tianqi.com / ).

[0083] The data preprocessing method is as follows: eliminate repeated invalid data in historical passenger flow, emergency event and other data, and use interpolation method to complete the data.

[0084] Second, feature processing

[0085] (1) Data normalization: process weather, holiday and other data into data that the model can learn and understand, use the maximum and minimum normalization method to normalize continuous data to the range of [0, 1], and use one-hot encoding to process discrete data;

[0086] (2) Use two-stage feature selection method to select features: calculate the MIC of 0-n order of each variable to determine the optimal lag structure, set the maximum lag order to 30, arrange the time series according to the lag order, and use the two-stage feature selection method to determine the key features.

[0087] Third, predict tourism demand based on T2V-MCNN-BiLSTM model

[0088] The processed data is input into the T2V-MCNN-BiLSTM hybrid model to obtain the corresponding output value, and the specific steps are as follows:

[0089] (1) Extracting time embedding features: The Time2Vec layer is used to automatically capture important periodic and non-periodic time features in the time series related to tourism demand. After obtaining the time embedding, it is connected with the original feature data as the input of MCNN-BiLSTM;

[0090] (2) Extracting time and variable interaction features: In the MCNN-BiLSTM layer, the superior feature extraction ability of CNN is used to extract high information and highly correlated features in all input vectors, BiLSTM is used to learn the features extracted by CNN to obtain more context information, and a connection function is introduced to effectively fuse the features extracted by CNN and BiLSTM.

[0091] (3) Output prediction results: The fusion data is processed through the full connection layer to obtain the final prediction results. The model architecture of T2V-MCNN-BiLSTM is shown in FIG. 2 (right). Figure 1

[0092] In the above steps, the two-stage feature selection method and the T2V-MCNN-BiLSTM model are the focus of the present application, which will be discussed in detail below.

[0093] 1. Two-stage feature selection method

[0094] The two-stage feature selection method includes two parts: multivariate correlation coefficient (MCC) and ReliefF algorithm.

[0095] In the first stage, the MCC (θ) is used to calculate the amount of redundant information between each feature in the key feature subset and other features to filter out some redundant features; a self-defined threshold β is used to screen the features, and when θ is greater than β, the feature is retained; the value of the mth feature of the nth sample is represented by f nm , and the mean value of the ith feature is defined as The covariance v ij and the correlation coefficient r ij are defined as:

[0096]

[0097] The covariance matrix and the correlation coefficient matrix are defined as V = {v ij} and R = {r ij}, where i, j ∈ (1, 2, …, m). The MCC is calculated based on the correlation coefficient matrix R:

[0098]

[0099] where θ i is the MCC between the ith feature and other features, r i , r i ​and R i-1 respectively represent the correlation coefficient matrix of the ith feature, r i the transpose of and the correlation coefficient matrix of other features except the ith feature;

[0100] The second stage is to determine the final features based on the features screened in the first stage using the ReliefF algorithm.

[0101] First, a sample S is randomly selected, and then q nearest neighbors of the same type H q = {h1, h2, h3,..., h q} of sample S are selected. In addition, q nearest neighbors of different types G q = {g1, g2, g3,..., g q} are found, and Gq is of type C. The weight of each feature F is calculated as follows:

[0102]

[0103]

[0104] where diff(F, S, s) represents the difference between sample S and s in feature F, S(F) and s(F) are the numerical values of sample S and s in feature F, respectively. P(C) and P(class(P)) are the prior probabilities of type C and the type of the randomly selected sample, respectively, i.e., the proportion of samples of this type in the total number of samples, max(F) and min(F) are the maximum and minimum values of feature F, and m is the number of iterations.

[0105] 2. T2V-MCNN-BiLSTM model

[0106] The T2V-MCNN-BiLSTM model includes two components:

[0107] (1) Time2Vec

[0108] Time2Vec is a time vector representation method that can effectively map discrete time information to continuous space, helping the model learn the dependencies in time, such as seasonal changes, trends, and periodicity (Kazemi, S. et. al, 2019). For a given scalar concept at time τ, the Time2Vec formula for the ith element is as follows:

[0109]

[0110] where ω i and are learnable parameters, and F is a periodic activation function. We use the sine function sin as the activation function.

[0111] In general, given a time series, Time2Vec encodes it as a k-dimensional vector, and this adaptive time embedding representation can be learned by other modules of deep learning models, so we apply Time2Vec to our model. By adding Time2Vec, the proposed deep learning model can dynamically extract time features from time series without cumbersome time series processing steps such as EMD, which can further improve the accuracy and efficiency of the prediction model.

[0112] (2) MCNN-BiLSTM

[0113] The CNN-BiLSTM model takes full advantage of the advantages of CNN in feature extraction and the powerful ability of BiLSTM in time series modeling by combining CNN and BiLSTM. First, CNN can extract more abstract and information-rich feature representations from the original time series data. Due to its local connectivity and global sharing characteristics, the network can capture the temporal dependencies and the correlation between the features related to tourism demand prediction, improving the accuracy and generalization ability of the model. Second, BiLSTM, as a lower component in the CNN-BiLSTM architecture, receives the time information of the tourism demand prediction important features obtained through CNN. BiLSTM combines information flow in two directions (forward and backward) to allow the model to consider past and future information while predicting each time step, thus more comprehensively understanding the temporal relationship in the sequence, reducing information loss while overcoming the gradient vanishing problem in long sequences. The outputs of the forward and backward layers of BiLSTM are connected together to form the final output, and the relevant calculation process is as follows:

[0114]

[0115] where LSTM represents a Long Short-Term Memory, which is a variant of Recurrent Neural Network (RNN) that can effectively handle and remember long-distance dependencies in sequence data, x i is the input vector of the input sequence at time step i, are the hidden states of the forward LSTM and the backward LSTM at time step i, respectively, [:, :] represents the concatenation operation of vectors.

[0116] Unlike the CNN-BiLSTM hybrid model, T2V-MCNN-BiLSTM additionally increases a flatten layer after each pooling layer of CNN to fully utilize the intermediate layer information. In the time series prediction task, the performance of the max-pooling operation is usually better than that of the average-pooling operation, so the max-pooling layer is selected.

[0117] Firstly, the Time2Vec layer is used to automatically capture important periodic and aperiodic time features in the time series of tourism demand. After obtaining the time embedding, it is connected with the original feature data as the input of MCNN-BiLSTM. Then, in the MCNN-BiLSTM layer, the CNN feature extraction is used to extract the high information and highly relevant features in all input vectors, and the BiLSTM is used to learn the CNN extracted features to obtain more context information, and the connection function is introduced to effectively fuse the CNN intermediate layer and the BiLSTM extracted features. Finally, the fusion data is processed through the full connection layer to obtain the final prediction result.

[0118] Embodiments

[0119] The present application is based on the daily tourist volume data of two famous domestic scenic spots, and analyzes the effectiveness of the mixed deep learning model based on Time2Vec in tourism demand prediction.

[0120] The present application first uses a two-stage feature selection method to calculate the top 3 features as the optimal subset. Before model training, the important hyperparameters of the model need to be determined, including the kernel size of CNN, the number of units of BiLSTM, etc. Taking the three-step prediction of one of the scenic spots as an example, the best parameter combination is that the number of CNN layers is 2, the filter number and convolution kernel size of the first layer CNN are 32 and 2, the filter number and convolution kernel size of the second layer CNN are 64 and 2, the activation function is ReLU, the number of BiLSTM units is 64, and the number of neurons in the full connection layer is 32.

[0121] In order to prove the effectiveness and superiority of the proposed model in tourism demand prediction, a large number of experiments are carried out. Experiment one is an ablation experiment, which aims to prove the necessity of each component of the model. Experiment two compares the performance with some baseline models to prove the superiority of the proposed model, and the baseline models are as follows:

[0122] SVR: Support vector regression,

[0123] ETS: exponential smoothing

[0124] ARIMAX: autoregressive integrated moving average with explanatory variable,

[0125] SARIMAX: seasonal autoregressive integrated moving average with explanatory variable,

[0126] GRU: Gated recurrent unit networks,

[0127] LSTM-AM: LSTM with the attention mechanism.

[0128] Some results are shown in Appendix 1, Appendix 2, and Appendix 3. Figure 2 As shown.

[0129] Appendix Table 1: Comparison of Predictive Performance of Each Model in Experiment 1

[0130]

[0131] Appendix Table 2 Predictive performance of different scenarios in Scenic Area 1

[0132]

[0133] Experiment 1: To verify the effectiveness of each module, this invention designed several models for ablation experiments, analyzing the impact and contribution of each module in the model network by removing specific modules. First, to verify the effectiveness of each component of MCNN-BiLSTM, four models were designed: CNN, BiLSTM, CNN-BiLSTM, and MCNN-BiLSTM. Next, to verify the impact of Time2Vec on model performance, three models were designed: T2V-CNN, T2V-BiLSTM, and T2V-MCNN-BiLSTM. Finally, we compared the performance of three models—the original data (T2V-MCNN-BiLSTM-O), PCA feature selection (T2V-MCNN-BiLSTM-P), and ReliefF feature selection (T2V-MCNN-BiLSTM-R)—with our proposed two-stage feature selection model. Appendix Table 1 shows the MAE, RMSE, and MAPE results obtained from the 10 model experiments; the best model evaluation metric is indicated in bold. It should be noted that for other models, we used the dataset after the two-stage feature selection. The results show that the MCNN-BiLSTM, Time2Vec layer, and two-stage feature selection method in the hybrid model all help to improve the model performance, and our proposed hybrid model performs the best.

[0134] Experiment two: In order to comprehensively evaluate the prediction effectiveness and superiority of the proposed model, we carried out single-step and multi-step prediction experiments on the scenic spot 1 dataset. Table 2 shows the prediction performance of scenic spot 1 in different scenarios, and the results show that the model T2V-MCNN-BiLSTM proposed in the application performs best. The traditional models ETS and SVR cannot capture the complex nonlinear patterns and dynamic changes in time series data, so the model prediction results are the worst. Compared with other models, the T2V-MCNN-BiLSTM model improves the average MAPE by 44.75%, 50.56%, and 49.79% in single-step, 3-step, and 7-step prediction. Taking 3-step prediction as an example, Figure 2 is a visualization of the prediction results of different models for scenic spot 1 tourism demand, which provides a more intuitive performance comparison. The prediction value of T2V-MCNN-BiLSTM is closest to the actual value, which proves the effectiveness of our model. Figure 2

[0135] The application further provides a tourism demand prediction system based on two-stage feature selection and Time2Vec, which comprises a memory, a processor, and computer program instructions stored in the memory and capable of being executed by the processor, and when the processor executes the computer program instructions, the method steps as described above can be realized.

[0136] The application further provides a computer readable storage medium having computer program instructions capable of being executed by a processor stored thereon, and when the processor executes the computer program instructions, the method steps as described above can be realized.

[0137] Those skilled in the art should understand that the embodiments of the application can be provided as methods, systems, or computer program products. Therefore, the application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0138] The application is described with reference to flowcharts and / or block diagrams according to the methods, devices (systems), and computer program products of the embodiments of the application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions described in the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1 ​means for performing the function specified by the block or blocks.

[0139] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 flow or flows and / or blocks Figure 1 means for performing the function specified by the block or blocks.

[0140] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 flow or flows and / or blocks Figure 1 means for performing the function specified by the block or blocks.

[0141] The above descriptions are only preferred embodiments of the present application, and are not intended to limit the present application to other forms. Any person skilled in the art may make modifications or improvements on the basis of the above disclosed technical content without departing from the technical scope of the present application. Any simple modification, equivalent change and improvement made on the basis of the technical essence of the present application without departing from the technical scope of the present application shall fall within the scope of protection of the present application.

Claims

1. A tourism demand forecasting method based on two-stage feature selection and Time2Vec, characterized in that, include: Design a two-stage feature selection method based on MCC and ReliefF algorithms to select a subset of key features from a dataset; We propose an MCNN-BiLSTM model for predicting tourism demand and introduce Time2Vec to integrate the time series processing into the MCNN-BiLSTM model, forming a T2V-MCNN-BiLSTM model, which predicts tourism demand based on a subset of key features. The T2V-MCNN-BiLSTM model is implemented as follows: (1)Time2Vec For a scalar concept of a given time τ, the Time2Vec formula for the i-th element is as follows: Where, ω i and is a learnable parameter, and F is a periodic activation function; In summary, given a time series, Time2Vec encodes it as a k-dimensional vector; (2) MCNN-BiLSTM The CNN-BiLSTM model combines CNN and BiLSTM; firstly, CNN is able to extract more abstract and information-rich feature representations from the original time series data; Secondly, BiLSTM, as a lower-level component in the CNN-BiLSTM model, receives temporal information from the important features for tourism demand prediction obtained through CNN. By combining information flows in two directions, namely forward and backward, BiLSTM allows the model to consider past and future information while predicting each time step, thus gaining a more comprehensive understanding of the temporal relationships in the sequence, reducing information loss, and overcoming the gradient vanishing problem in long sequences. The outputs of the forward and backward layers of BiLSTM are concatenated to form the final output, and the relevant calculation process is as follows: Where LSTM represents Long Short-Term Memory Network, x t Let be the input vector of the input sequence at time step t. These are the hidden states of the forward LSTM and the backward LSTM at time step t, respectively, and [:,:] represents the vector concatenation operation; An additional flattened layer is added after each pooling layer in the CNN to make full use of the information in the intermediate layers, and the pooling layer is selected as the max pooling layer.

2. The tourism demand forecasting method based on two-stage feature selection and Time2Vec as described in claim 1, characterized in that, The dataset can be obtained in the following ways: (1) Collect historical visitor flow data: The historical visitor flow data of the scenic area is collected from the official website of the scenic area; (2) Collect Baidu Index data: Use Python programming to obtain Baidu Index data of search keywords related to scenic spots. The strategy for selecting keywords is to use central words related to tourist attractions and use the "related search" strategy provided by the Baidu Index data platform to bring out other related keywords or phrases. (3) Collect emergency data: Collect daily updates on emergency-related indicators; (4) Collect other data: including weather data and holiday data, of which weather data includes weather conditions and temperature, collected from the Weather Network.

3. The tourism demand forecasting method based on two-stage feature selection and Time2Vec as described in claim 1, characterized in that, Before using the dataset, duplicate and invalid data should be removed, data should be completed using interpolation, and then normalization should be performed.

4. The tourism demand forecasting method based on two-stage feature selection and Time2Vec as described in claim 1, characterized in that, The two-stage feature selection method based on MCC and ReliefF algorithms selects a subset of key features from the dataset as follows: calculate the MIC of each variable from 0 to n to determine the optimal lag structure, set the maximum lag order to 30, arrange the time series according to the lag order, and use the two-stage feature selection method to determine the key features, thus obtaining the subset of key features.

5. The tourism demand forecasting method based on two-stage feature selection and Time2Vec as described in claim 1, characterized in that, The method for predicting tourism demand using the T2V-MCNN-BiLSTM model, based on a subset of key features, is as follows: (1) Extracting temporal embedding features: Time2Vec is used to automatically capture periodic and aperiodic time features in tourism demand-related time series within a key feature subset; After obtaining the temporal embedding, it is concatenated with the original feature data in the key feature subset and used as the input to the MCNN-BiLSTM model; (2) Extracting time and variable interaction features: In the MCNN-BiLSTM model, CNN is used to extract features with high information content and high correlation from all input vectors, BiLSTM is used to learn the features extracted by CNN to obtain more contextual information, and a connection function is introduced to fuse the features extracted by CNN and BiLSTM. (3) Output prediction results: The fused features are processed through a fully connected layer to obtain the final prediction results.

6. A tourism demand forecasting method based on two-stage feature selection and Time2Vec as described in claim 1 or 4, characterized in that, The two-stage feature selection method is as follows: The first stage uses MCC(θ) to calculate the amount of redundant information between each feature and other features in the key feature subset, thus filtering out some redundant features; features are then selected based on a custom threshold β, and features are retained when θ is greater than β; the m-th feature value of the n-th sample is calculated using f. nm The mean of the i-th feature is defined as follows: covariance v ij and correlation coefficient r ij Defined as: The covariance matrix and correlation coefficient matrix are defined as V = {v ij } and R = {r ij }, where i,j∈(1,2,…,m); MCC is calculated based on the correlation coefficient matrix R: Where θ i It is the MCC between the i-th feature and other features, r i r i ' and R i-1 Let r be the correlation coefficient matrix of the i-th feature, r, and r be the correlation coefficient matrix of the i-th feature, respectively. i The transpose of the matrix and the correlation coefficient matrix of all features except the i-th feature; The second stage is based on the features obtained in the first stage, and uses the ReliefF algorithm to determine the final features; First, randomly select a sample S, then select q nearest neighbors H of the same type as sample S. q ={h1,h2,h3,…,h q Additionally, find q nearest neighbors of different classes, G. q ={g1,g2,g3,…,g q }, Gq is of type C, and the weights of each feature F are calculated as follows: Where diff(F,S,s) represents the difference between samples S and s on feature F, S(F) and s(F) are the numerical values ​​of samples S and s on feature F, respectively, P(C) and P(class(P)) are the prior probabilities of type C and the type of randomly selected samples, respectively, that is, the proportion of samples of type C and type S in the total number of samples in the dataset, max(F) and min(F) are the maximum and minimum values ​​of feature F, and m is the number of iterations.

7. The tourism demand forecasting method based on two-stage feature selection and Time2Vec as described in claim 1, characterized in that, The periodic activation function uses the sine function sin.

8. A tourism demand forecasting system based on two-stage feature selection and Time2Vec, characterized in that, It includes a memory, a processor, and computer program instructions stored in the memory and executable by the processor, which, when executed by the processor, enable the implementation of the steps of the method as described in any one of claims 1-7.

9. A computer-readable storage medium having stored thereon computer program instructions executable by a processor, wherein when the processor executes the computer program instructions, it is able to implement the steps of the method as described in any one of claims 1-7.

Images