A tourism management optimization method and system based on tourist behavior analysis

Abstract

The present invention belongs to the field of data modeling and analysis, and provides a tourism management optimization method and system based on tourist behavior analysis, including: periodically obtaining feature data of each node in a scenic spot; obtaining edge feature data of tourists moving between different nodes in the scenic spot; in each data collection cycle, performing convolution operation on the feature data of the node, and using the edge feature data to perform weighted aggregation on the features of neighboring nodes, and fusing the aggregation result with the features of the current node to obtain a convolution output; inputting the convolution output of each collection cycle as time series data into a long short-term memory network; using the long short-term memory network to process these time series data; predicting the tourist flow of each node in the future time step; and dynamically allocating service personnel and resources of each node in the future time step according to the tourist flow prediction results of each node in the future time step. The present invention can optimize scenic spot management and enhance tourist experience.

Description

Technical Field

[0001] The present invention belongs to the field of data modeling and analysis, and in particular relates to a tourism management optimization method and system based on tourist behavior analysis. Background Art

[0002] With the rapid development of tourism, the number of tourists in scenic spots continues to increase, which brings great challenges to the management of scenic spots. Factors such as the peak and trough periods of tourist flow, the popularity of different attractions, and the movement patterns of tourists in scenic spots directly affect the tourists' travel experience and the operational efficiency of scenic spots. Traditional management methods usually rely on manual experience and rough statistical data, which are difficult to accurately predict and adjust in real time, and easily lead to overcrowding of some nodes in the scenic area and idle resources of other nodes, thereby reducing the overall service quality.

[0003] In recent years, with the advancement of data collection technology and big data analysis technology, scenic area managers can obtain a large amount of tourist behavior data. This data includes tourists' stay time, movement paths, number of visitors at different nodes, etc. However, how to efficiently use this data for analysis and prediction to optimize scenic area management and enhance tourist experience is still a problem that needs to be solved. Summary of the invention

[0004] In order to solve the problems in the above background technology, a tourism management optimization method and system based on tourist behavior analysis are proposed.

[0005] In one embodiment, the present invention proposes a tourism management optimization method based on tourist behavior analysis, comprising the following steps:

[0006] Periodically obtain characteristic data of each node in the scenic area, the characteristic data including the number of tourists and node classification, the node classification including ticket points, entrances, scenic area exits, scenic spots, service points, transportation stations and toilets;

[0007] Acquire edge feature data of tourists moving between different nodes in the scenic area, wherein the edge feature data includes the number of tourists moving, average stay time and movement direction;

[0008] Clean, fill missing values ​​and normalize the collected node feature data and edge feature data;

[0009] In each data collection cycle, a convolution operation is performed on the feature data of the node, and the features of the neighboring nodes are weighted aggregated using the edge feature data, and the aggregation result is fused with the features of the current node; an activation function is used to perform a nonlinear transformation on the fused features to obtain a convolution output;

[0010] The convolution output of each acquisition cycle is input into the LSTM network as time series data; the LSTM network is used to process these time series data to capture the temporal dynamic changes of tourist flow; the output of the LSTM network layer is mapped through the fully connected layer to predict the tourist flow of each node in the future time step;

[0011] According to the tourist flow prediction results of each node in the future time step, the service personnel and resources of each node in the future time step are dynamically allocated to optimize the scenic area management and enhance the tourist experience.

[0012] In another embodiment, the present invention further proposes a tourism management optimization system based on tourist behavior analysis, and includes the following modules:

[0013] The acquisition module is used to periodically acquire the characteristic data of each node in the scenic area, the characteristic data including the number of tourists and node classification, the node classification including ticket points, entrances, scenic area exits, scenic spots, service points, transportation stations and toilets; acquire the edge characteristic data of tourists moving between different nodes in the scenic area, the edge characteristic data including the number of tourists moving, average stay time and moving direction;

[0014] The data processing module is used to clean, fill in missing values ​​and normalize the collected node feature data and edge feature data;

[0015] A convolution module is used to perform a convolution operation on the feature data of the node in each data collection cycle, and use the edge feature data to perform weighted aggregation on the features of neighboring nodes, and fuse the aggregation result with the features of the current node; use an activation function to perform a nonlinear transformation on the fused features to obtain a convolution output;

[0016] The prediction module is used to input the convolution output of each acquisition cycle as time series data into the LSTM network; use the LSTM network to process these time series data to capture the temporal dynamic changes of tourist flow; map the output of the LSTM network layer through the fully connected layer to predict the tourist flow of each node in the future time step;

[0017] The adjustment module is used to dynamically allocate service personnel and resources for each node in the future time step according to the tourist flow prediction results of each node in the future time step, so as to optimize the scenic area management and enhance the tourist experience.

[0018] The present invention can produce the following beneficial effects through the above technical solution:

[0019] The solution of this embodiment combines the idea of ​​directed graph and long short-term memory network (LSTM) to predict the tourist flow of each node in the scenic area, and optimize resource allocation and service management accordingly.

[0020] By processing node and edge features through directed graphs, the complex spatial relationships between nodes and their neighbors in the scenic area are captured, and preliminary spatial features are generated. LSTM is used to process these time series data to capture the temporal dynamic changes in tourist flow and improve prediction accuracy. Combining spatial and temporal features enables the prediction model to more accurately reflect the actual situation.

[0021] Identify high-traffic nodes in advance and take preventive measures to reduce congestion and improve tourists' travel experience. Deploy more service personnel and facilities at high-traffic nodes to ensure that tourists can receive good services even during peak hours. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0023] Figure 1 is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0024] Below, the invention is preferably described in conjunction with the accompanying drawings and specific implementation methods.

[0025] This embodiment solves the above problem through the following steps:

[0026] In one embodiment, reference Figure 1 The present invention provides a tourism management optimization method based on tourist behavior analysis, comprising the following steps:

[0027] The characteristic data of each node in the scenic area are periodically obtained, wherein the characteristic data includes the number of tourists and the node classification, wherein the node classification includes ticket points, entrances, scenic area exits, scenic spots, service points, transportation stations and toilets.

[0028] The characteristic data of each node in the scenic area is obtained periodically, where the period is determined according to the time scale required for prediction, such as 30 minutes or 1 hour. The characteristic data includes but is not limited to the number of tourists at each node and the node classification. Node classification refers to dividing the nodes into several categories according to the areas with different functions and locations in the scenic area, so as to conduct more accurate traffic analysis and management. These node classifications specifically include the following:

[0029] Ticket sales point: One of the entrances for tourists to enter the scenic area, which usually gathers a large number of tourists and has high traffic volatility and concentration.

[0030] Entrance: The main entrance of the scenic area, responsible for the initial diversion and management of tourists. There are a large number of tourists and obvious traffic peaks.

[0031] Scenic area exit: The passage through which tourists leave the scenic area. The flow rate has a certain temporal regularity, and usually there is a flow peak before the scenic area closes.

[0032] Scenic spots: The main tourist attractions in the scenic area are where tourists mainly stay and gather. The traffic changes are complex and there are frequent interactions with other nodes.

[0033] Service point: An area that provides tourists with dining, shopping, rest and other services. Tourists stay here for a relatively long time and the traffic has high service demand characteristics.

[0034] Transportation station: A transportation hub within a scenic area, including stops and transfer points for internal transportation. Tourist flow is closely related to traffic management within the scenic area.

[0035] Toilets: Basic service facilities within scenic areas. Although the traffic volume is relatively small compared to other nodes, their distribution and usage have an important impact on the overall service quality of the scenic area.

[0036] By obtaining the characteristic data of the above nodes, we can fully understand the distribution and flow of tourists in each functional area of ​​the scenic area. The periodic collection of these characteristic data can provide dynamic and real-time tourist flow information, providing basic data support for subsequent flow prediction and management optimization.

[0037] To ensure the accuracy and real-time nature of data collection, a variety of technical means can be used, including but not limited to the following methods:

[0038] Surveillance cameras: Install high-resolution surveillance cameras at each node to monitor and count the number of visitors in real time through video analysis technology. Using computer vision technology, the flow of visitors at each node can be automatically identified and counted, and behavioral analysis can be performed.

[0039] Radio Frequency Identification (RFID): Tickets with RFID tags are issued to visitors when they enter the park. RFID readers are deployed in the scenic area to collect real-time data on visitors’ stay and movement at different nodes.

[0040] Wi-Fi probe: Utilize the Wi-Fi signals of tourists’ mobile phones and deploy Wi-Fi probe devices in the scenic area to collect tourists’ real-time location and flow information.

[0041] Bluetooth beacon: Bluetooth beacon devices are deployed in the scenic area to collect tourists’ location information and movement tracks in real time by detecting the Bluetooth signals of tourists’ mobile phones.

[0042] GPS positioning: Use the GPS function on tourists’ mobile phones to obtain tourists’ geographic location information in real time and accurately record their movement paths and stay time in the scenic area.

[0043] Infrared sensing: Infrared sensing equipment is placed at entrances, exits and other important nodes to detect the entry and exit behavior of tourists and collect real-time traffic data.

[0044] Smart bracelets: Equip tourists with smart bracelets with built-in sensors that can monitor their location, activity trajectory and health status in real time, providing more comprehensive behavioral data.

[0045] Through the comprehensive application of the above technical means, it is possible to accurately monitor the number and behavior of tourists in the scenic area and obtain high-quality node feature data. This comprehensive node feature data collection method helps to more accurately analyze and predict tourists' behavior patterns, thereby improving the intelligent level of scenic area management and tourist experience.

[0046] In order to ensure the accuracy and reliability of the data, the frequency of collecting node feature data should be adjusted according to the specific situation of the scenic spot. For example, the collection frequency can be increased during the peak tourist season to capture traffic changes more timely; during the trough period, the collection frequency can be appropriately reduced to save resources.

[0047] Periodically acquiring characteristic data of each node in the scenic area and using a variety of advanced technical means to collect data is an important basis for realizing intelligent tourism management optimization. By refining node classification and accurately collecting dynamic information such as the number of tourists, solid data support can be provided for subsequent traffic forecasting and management decisions.

[0048] Acquire edge feature data of tourists moving between different nodes in the scenic area, wherein the edge feature data includes the number of tourists moving, average stay time and movement direction;

[0049] The edge feature data of tourists moving between different nodes in the scenic area are obtained, and the edge feature data are not limited to the number of tourists moving, the average stay time and the moving direction.

[0050] Tourist movement number: refers to the total number of tourists moving from one node to another in a specific time period. This data can reflect the connection strength between nodes and the main paths of tourist flow. Edges with a high number of movements indicate that tourists move frequently on this path and are key traffic paths.

[0051] Average stay time: refers to the average length of time that tourists stay at a node. This data can help managers understand the attractiveness of different nodes and the activity patterns of tourists. A longer stay time usually means that the node is more attractive to tourists or requires more time to visit and experience.

[0052] Movement direction: refers to the direction in which tourists move from one node to another. This feature can reflect tourists' movement preferences and main flow trends. By analyzing the movement direction, we can understand the activity paths of tourists and identify the main movement lines and potential bottleneck areas in the scenic area.

[0053] In order to obtain these edge feature data, a similar method as the previous step can be used, such as installing surveillance cameras at each node and main channel, and using computer vision technology to monitor and analyze the movement of tourists in real time. After the video data is processed, information such as the number of tourists' movements, stay time, and movement direction can be extracted. At the same time, technologies such as RFID, Wi-Fi probes, and Bluetooth beacons can also be used.

[0054] By comprehensively applying the above technical means, we can comprehensively and accurately obtain the edge feature data of tourists moving between different nodes in the scenic area. These data are of great significance for analyzing tourist behavior patterns, optimizing scenic area management and improving tourist experience.

[0055] The collected node feature data and edge feature data are cleaned, missing values ​​are filled and normalized.

[0056] The collected node feature data and edge feature data are cleaned, missing values ​​are filled and normalized to ensure the accuracy and consistency of the data, thus providing a reliable foundation for subsequent data analysis and model training.

[0057] Identify and handle outliers in the data, such as extremely large or small values, which may be caused by sensor failure, data entry errors, or other reasons. Detect outliers through statistical methods (such as mean and standard deviation) or machine learning algorithms (such as isolation forests) and take appropriate measures to deal with them, such as deleting or replacing them with reasonable values.

[0058] During the data collection process, duplicate records may be generated, which may affect the accuracy of data analysis. Duplicate data can be identified and removed by checking the timestamp and location information of the data records.

[0059] Tourist data is time series data, and missing values ​​can be filled by linear interpolation or spline interpolation. This method assumes that the change of missing values ​​in time is smooth and continuous.

[0060] Scale all feature data to the same range (such as 0 to 1 or -1 to 1) to eliminate differences in the magnitude of different features. This can be done through min-max scaling or standardization (i.e. scaling the data to have a mean of 0 and a standard deviation of 1).

[0061] For time series data with large fluctuations, smoothing techniques (such as moving average or exponential smoothing) are applied to reduce noise and make the data more stable.

[0062] Furthermore, you can also use data visualization tools (such as line charts, histograms, and scatter plots) to check the effect of data processing and ensure that the data processing steps do not introduce new anomalies or errors.

[0063] By cleaning, filling missing values ​​and normalizing node feature data and edge feature data, the quality and consistency of the data can be significantly improved.

[0064] In each data collection cycle, a convolution operation is performed on the feature data of the node, and the edge feature data is used to perform weighted aggregation on the features of the neighboring nodes, and the aggregation result is fused with the features of the current node; an activation function is used to perform a nonlinear transformation on the fused features to obtain a convolution output.

[0065] At the beginning of each data collection cycle, the feature data of each node includes the current number of tourists, node classification information (such as ticket points, entrances, scenic area exits, attractions, service points, transportation stations and toilets), and other possible auxiliary features (such as node area, opening hours, etc.). These initial features are represented as a vector:

[0066] ,

[0067] Based on the data of nodes and edges, a graph structure is constructed, where nodes represent various locations in the scenic area and edges represent the movement paths of tourists between these locations. Each node and edge is accompanied by corresponding feature data.

[0068] For each node, the feature of the node is updated by aggregating the features of its neighboring nodes.

[0069] Use edge features to perform weighted aggregation of neighbor node features. The weighted aggregation process takes into account the strength (number of moves) and direction (movement direction) of the edge, ensuring that the aggregation result can accurately reflect the dynamic relationship between nodes.

[0070] ,

[0071] in:

[0072] is the feature representation of node i at the k+1th layer;

[0073] is the feature representation of node i at layer k;

[0074] is the feature representation of node j at layer k;

[0075] is the set of neighbor nodes of node i;

[0076] is the normalization coefficient between node i and node j, taking the number of neighbor nodes;

[0077] is the weight matrix of the kth layer, which is learned by the back-propagation algorithm based on the training data;

[0078] is the activation function (preferably ReLU).

[0079] The aggregated neighbor node features are fused with the current node features, and a new node feature representation is generated through weighting and superposition. This fusion process is implemented through the neural network layer, so that the new node features contain more spatial information. The details are as follows:

[0080] ,

[0081] in:

[0082] is the fused feature representation of node i at the k+1th layer;

[0083] is the weight of edge (i,j), reflecting the influence of edge features on aggregation. Calculated based on the characteristics of the edge (such as the number of tourists moving, average stay time and movement direction):

[0084] ,

[0085] Where f is a mapping function;

[0086] is the bias vector of the kth layer.

[0087] The fused node features are transformed through a nonlinear activation function (such as ReLU) to increase the expressive power of the model and capture nonlinear features.

[0088] Furthermore, a simple two-layer neural network is used to implement the nonlinear mapping function to obtain :

[0089] ,

[0090] in, is the first-layer nonlinear mapping function, which is used to represent the trend of tourist flow;

[0091] ,

[0092] and are the weight matrices of the first and second layers respectively;

[0093] and is the bias vector;

[0094] is the hyperbolic tangent activation function;

[0095] is the Sigmoid activation function;

[0096] During the training process, these weights and bias parameters are adjusted through the back-propagation algorithm so that the model can more accurately capture the relationship between edge features and weights.

[0097] After the above convolution operation, the features of each node are updated and preliminary spatial features are extracted. These features not only contain the information of the node itself, but also the aggregated information of its neighboring nodes, reflecting the spatial position and relationship of the node in the entire graph structure.

[0098] By performing a convolution operation on the node feature data in each data collection cycle, the complex spatial relationships and dynamic changes between the nodes and their neighbors in the scenic area can be dynamically captured. The extracted preliminary spatial features provide a rich information basis for subsequent prediction and analysis, ensuring that the model can be updated in a timely manner and adapt to changes in tourist flow.

[0099] The convolution output of each acquisition cycle is input into the LSTM network as time series data; the LSTM network is used to process these time series data to capture the temporal dynamic changes of tourist flow; the output of the LSTM network layer is mapped through a fully connected layer to predict the tourist flow of each node in the future time step.

[0100] After each data collection cycle, the convolution output generated by the convolution operation is used as the node feature representation of the current cycle. These convolution outputs contain the preliminary spatial features of the node and the aggregated information of the neighboring nodes.

[0101] The convolution output of each node is arranged in chronological order to form the time series data of the node features. For node 𝑖, its convolution output in different acquisition cycles is arranged in sequence as follows:

[0102] ,

[0103] in Indicates that node i is in The convolution output of acquisition cycles.

[0104] The above time series data is formatted into a form suitable for LSTM input. Specifically, the convolution output time series data of each node is divided into several time windows of fixed length so that LSTM can process and learn the dynamic change patterns of these sequence data.

[0105] Assume that the time window length is , then for node i, its time series data can be divided into several subsequences

[0106] ,

[0107] Each subsequence contains a fixed number of time steps of data, ensuring that LSTM can effectively learn temporal dynamic features.

[0108] The formatted time series data is input into the LSTM network. LSTM is a recurrent neural network (RNN) suitable for processing time series data and can capture long-term and short-term dependencies in the data.

[0109] In LSTM, the input of each time step is combined with the hidden state (hidden layer output) of the previous time step. Through the memory and forgetting mechanism of the LSTM unit, the hidden state is gradually updated to capture the temporal dynamic changes in the sequence data.

[0110] LSTM captures the dynamic changes of tourist flow at different time steps by gradually processing the input time series data. It is updated at each time step and contains information from all previous time steps in the sequence. The continuity and recursiveness of the hidden state enables LSTM to effectively model long-term and short-term dependencies in time series.

[0111] The output of LSTM is represented as , which contains the comprehensive features of node i at different time steps. This output can be regarded as a high-dimensional representation of time series data, capturing important patterns and dynamic changes in the data.

[0112] The output of the LSTM network is mapped through the fully connected layer to obtain the predicted value of tourist flow for each node in the future time step:

[0113] ,

[0114] in:

[0115] is the predicted value of tourist flow at node i in the future time step;

[0116] is the weight matrix of the fully connected layer, which is learned through training data;

[0117] is the bias vector of the fully connected layer, which is learned through training data.

[0118] According to the tourist flow prediction results of each node in the future time step, the service personnel and resources of each node in the future time step are dynamically allocated to optimize the scenic area management and enhance the tourist experience.

[0119] Based on the calculated service personnel demand, service personnel are dynamically allocated to each node. The allocation process can adopt a variety of strategies to ensure maximum resource utilization and optimal service quality. For example:

[0120] Priority allocation strategy: Prioritize the allocation of service personnel to nodes with larger predicted tourist flow to ensure that high-flow nodes have sufficient service support.

[0121] Balanced allocation strategy: Under the premise of meeting the needs of high-traffic nodes, try to balance the number of service personnel at each node to avoid too many or too few service personnel at some nodes.

[0122] Flexible allocation strategy: Dynamically adjust the allocation of service personnel based on real-time tourist flow and emergencies to increase flexibility and response capabilities.

[0123] In addition to service personnel, other resources can also be dynamically allocated based on the tourist flow forecast results, such as adding sanitary facilities at high-flow nodes, supplementing ticket sales points, and preparing more transportation tools to cope with peak tourist demand.

[0124] Resource allocations may include:

[0125] Sanitary facilities: Temporary toilets and trash cans are added at predicted high traffic nodes to meet the needs of tourists.

[0126] Ticketing facilities: Add more ticket windows or self-service ticket machines at ticket sales points to reduce waiting time in queues.

[0127] Transportation facilities: Increase the frequency and capacity of shuttle buses and sightseeing buses within the scenic area to ensure that tourists can move smoothly to various attractions.

[0128] Security and guide personnel: Deploy more security personnel and guide personnel at high-traffic nodes to ensure the safety of tourists and provide guidance services.

[0129] The solution of this embodiment combines the idea of ​​directed graph and long short-term memory network (LSTM) to predict the tourist flow of each node in the scenic area, and optimize resource allocation and service management accordingly.

[0130] By processing node and edge features through directed graphs, the complex spatial relationships between nodes and their neighbors in the scenic area are captured, and preliminary spatial features are generated. LSTM is used to process these time series data to capture the temporal dynamic changes in tourist flow and improve prediction accuracy. Combining spatial and temporal features enables the prediction model to more accurately reflect the actual situation.

[0131] Identify high-traffic nodes in advance and take preventive measures to reduce congestion and improve tourists' travel experience. Deploy more service personnel and facilities at high-traffic nodes to ensure that tourists can receive good services even during peak hours.

[0132] On the other hand, the present invention also provides a tourism management optimization system based on tourist behavior analysis, comprising the following modules:

[0133] The acquisition module is used to periodically acquire the characteristic data of each node in the scenic area, the characteristic data including the number of tourists and node classification, the node classification including ticket points, entrances, scenic area exits, scenic spots, service points, transportation stations and toilets; acquire the edge characteristic data of tourists moving between different nodes in the scenic area, the edge characteristic data including the number of tourists moving, average stay time and moving direction;

[0134] The data processing module is used to clean, fill in missing values ​​and normalize the collected node feature data and edge feature data;

[0135] A convolution module is used to perform a convolution operation on the feature data of the node in each data collection cycle, and use the edge feature data to perform weighted aggregation on the features of neighboring nodes, and fuse the aggregation result with the features of the current node; use an activation function to perform a nonlinear transformation on the fused features to obtain a convolution output;

[0136] The prediction module is used to input the convolution output of each acquisition cycle as time series data into the LSTM network; use the LSTM network to process these time series data to capture the temporal dynamic changes of tourist flow; map the output of the LSTM network layer through the fully connected layer to predict the tourist flow of each node in the future time step;

[0137] The adjustment module is used to dynamically allocate service personnel and resources for each node in the future time step according to the tourist flow prediction results of each node in the future time step, so as to optimize the scenic area management and enhance the tourist experience.

[0138] Furthermore, the specific implementation methods of the above-mentioned tourism management optimization system based on tourist behavior analysis are the same as a tourism management optimization method based on tourist behavior analysis, and all further technical solutions in a tourism management optimization method based on tourist behavior analysis are fully introduced into a tourism management optimization system based on tourist behavior analysis.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

[0140] The present invention does not particularly specify the structure of some modules, which shall be subject to the contents recorded in the prior art. The prior art mentioned in the above background technology section and the specific embodiment section of the present invention can be used as part of the present invention to understand the meaning of some technical features or parameters. The scope of protection of the present invention shall be subject to the contents actually recorded in the claims.

Claims

1. A tourism management optimization method based on tourist behavior analysis, characterized in that: The method comprises the following steps: Periodically obtain characteristic data of each node in the scenic area, the characteristic data including the number of tourists and node classification, the node classification including ticket points, entrances, scenic area exits, scenic spots, service points, transportation stations and toilets; Acquire edge feature data of tourists moving between different nodes in the scenic area, wherein the edge feature data includes the number of tourists moving, average stay time and movement direction; Clean, fill missing values ​​and normalize the collected node feature data and edge feature data; In each data collection cycle, the feature data of the node is convolved, and the features of the neighboring nodes are weightedly aggregated using the edge feature data, and the aggregation result is fused with the features of the current node; the fused features are nonlinearly transformed using an activation function to obtain a convolution output; the convolution output of each collection cycle is input into the long short-term memory network as time series data; the long short-term memory network is used to process these time series data to capture the temporal dynamic changes of tourist flow; the output of the long short-term memory network layer is mapped through a fully connected layer to predict the tourist flow of each node in the future time step; According to the tourist flow prediction results of each node in the future time step, the service personnel and resources of each node in the future time step are dynamically allocated to optimize the scenic area management and improve the tourist experience; At the beginning of each data collection cycle, the feature data of each node includes the current number of tourists, node classification information, and auxiliary features. These initial features are represented as a vector: h i (0) =[number of tourists, node classification, node area, opening hours] Based on the data of nodes and edges, a graph structure is constructed, where nodes represent various locations in the scenic area and edges represent the movement paths of tourists between these locations; Use edge features to perform weighted aggregation on neighbor node features: in: is the feature representation of node i at the k+1th layer; is the feature representation of node i at layer k; is the feature representation of node j at layer k; is the set of neighbor nodes of node i; c ij is the normalization coefficient between node i and node j, and its value is the number of neighboring nodes; W k is the weight matrix of the kth layer, which is learned from the training data through the back propagation algorithm σ is the activation function; Merge the aggregated neighbor node features with the features of the current node: in: It is the fused feature representation of node i at the k+1th layer; α ij is the weight of edge (i, j), reflecting the influence of edge features on aggregation; α ij = f(number of tourists moving ij , average residence time ij , moving direction ij ) Where f is the mapping function; b k is the bias vector of the kth layer; Transform the fused node features through a nonlinear activation function; Use a two-layer neural network to implement the nonlinear mapping function to get α ij : z ij =tanh(W1[number of tourists moving ij , average residence time ij , moving direction ij ]+b1) Among them, z ij is the first-layer nonlinear mapping function, which is used to represent the trend of tourist flow; a ij =σ(W2z ij +b2) W1 and W2 are the weight matrices of the first and second layers respectively; b1 and b2 are bias vectors; tanh is the hyperbolic tangent activation function; σ is the Sigmoid activation function; The convolution output of each node is arranged in chronological order to form the time series data of the node features. For node i, its convolution output in different acquisition cycles is arranged in sequence as follows: in Indicates that node i is in the tth k Convolution output of acquisition cycles; The convolution output time series data of each node is divided into several time windows of fixed length. Assuming the time window length is W, for node i, its time series data is divided into several subsequences Each subsequence contains a fixed number of time steps of data; the formatted time series data is input into the LSTM network; in the LSTM, the input of each time step is combined with the hidden state of the previous time step, and the hidden state is gradually updated through the memory and forgetting mechanism of the LSTM unit, thereby capturing the temporal dynamic changes in the sequence data; The output of LSTM is represented as It contains the comprehensive features of node i at different time steps; the output of the LSTM network is mapped through the fully connected layer to obtain the predicted value of tourist flow for each node in the future time step: in: y i is the predicted value of tourist flow at node i in the future time step; W y is the weight matrix of the fully connected layer, which is learned through training data; b y is the bias vector of the fully connected layer, which is learned through training data.

2. A tourism management optimization method based on tourist behavior analysis according to claim 1, characterized in that: Dynamic allocation of service personnel to each node in future time steps includes: using a priority allocation strategy: allocating service personnel to nodes with larger predicted tourist flow to ensure that high-flow nodes have sufficient service support; or, a balanced allocation strategy: while meeting the needs of high-flow nodes, balancing the number of service personnel at each node to avoid having too many or too few service personnel at some nodes; or, a flexible allocation strategy: dynamically adjusting the allocation of service personnel according to real-time tourist flow and emergencies to increase flexibility and response capabilities; making adjustments to service personnel.

3. A tourism management optimization system based on tourist behavior analysis, characterized by The system includes the following modules: The acquisition module is used to periodically acquire the characteristic data of each node in the scenic area, the characteristic data including the number of tourists and node classification, the node classification including ticket points, entrances, scenic area exits, scenic spots, service points, transportation stations and toilets; acquire the edge characteristic data of tourists moving between different nodes in the scenic area, the edge characteristic data including the number of tourists moving, average stay time and moving direction; The data processing module is used to clean, fill in missing values ​​and normalize the collected node feature data and edge feature data; A convolution module is used to perform a convolution operation on the feature data of the node in each data collection cycle, and use the edge feature data to perform weighted aggregation on the features of neighboring nodes, and fuse the aggregation result with the features of the current node; use an activation function to perform a nonlinear transformation on the fused features to obtain a convolution output; The prediction module is used to input the convolution output of each acquisition cycle as time series data into the LSTM network; use the LSTM network to process these time series data to capture the temporal dynamic changes of tourist flow; map the output of the LSTM network layer through the fully connected layer to predict the tourist flow of each node in the future time step; The adjustment module is used to dynamically allocate service personnel and resources for each node in the future time step according to the tourist flow prediction results of each node in the future time step, so as to optimize the scenic area management and improve the tourist experience; At the beginning of each data collection cycle, the feature data of each node includes the current number of tourists, node classification information, and auxiliary features. These initial features are represented as a vector: Based on the data of nodes and edges, a graph structure is constructed, where nodes represent various locations in the scenic area and edges represent the movement paths of tourists between these locations; Use edge features to perform weighted aggregation on neighbor node features: in: is the feature representation of node i at the k+1th layer; is the feature representation of node i at layer k; is the feature representation of node j at layer k; is the set of neighbor nodes of node i; c ij is the normalization coefficient between node i and node j, and its value is the number of neighbor nodes; W k is the weight matrix of the kth layer, which is learned from the training data through the back propagation algorithm σ is the activation function; Merge the aggregated neighbor node features with the features of the current node: in: It is the fused feature representation of node i at the k+1th layer; α ij is the weight of edge (i, j), reflecting the influence of edge features on aggregation; α ij = f(number of tourists moving ij , average residence time ij , moving direction ij ) Where f is the mapping function; b k is the bias vector of the kth layer; Transform the fused node features through a nonlinear activation function; Use a two-layer neural network to implement the nonlinear mapping function to get α ij : z ij =tanh(W1[number of tourists moving ij , average residence time ij , moving direction ij ]+b1) Among them, z ij is the first layer nonlinear mapping function, which is used to represent the trend of tourist flow; α ij =σ(W2z ij +b2) W1 and W2 are the weight matrices of the first and second layers respectively; b1 and b2 are bias vectors; tanh is the hyperbolic tangent activation function; σ is the Sigmoid activation function; The convolution output of each node is arranged in chronological order to form the time series data of the node features. For node i, its convolution output in different acquisition cycles is arranged in sequence as follows: in Indicates that node i is in the tth k Convolution output of acquisition cycles; The convolution output time series data of each node is divided into several time windows of fixed length. Assuming the time window length is W, for node i, its time series data is divided into several subsequences Each subsequence contains a fixed number of time steps of data; the formatted time series data is input into the LSTM network; in the LSTM, the input of each time step is combined with the hidden state of the previous time step, and the hidden state is gradually updated through the memory and forgetting mechanism of the LSTM unit, thereby capturing the temporal dynamic changes in the sequence data; The output of LSTM is represented as It contains the comprehensive features of node i at different time steps; the output of the LSTM network is mapped through the fully connected layer to obtain the predicted value of tourist flow for each node in the future time step: in: y i is the predicted value of tourist flow at node i in the future time step; W y is the weight matrix of the fully connected layer, which is learned through training data; b y is the bias vector of the fully connected layer, which is learned through training data.

4. A tourism management optimization system based on tourist behavior analysis according to claim 3, characterized in that: Dynamic allocation of service personnel to each node in future time steps includes: using a priority allocation strategy: allocating service personnel to nodes with larger predicted tourist flow to ensure that high-flow nodes have sufficient service support; or, a balanced allocation strategy: while meeting the needs of high-flow nodes, balancing the number of service personnel at each node to avoid having too many or too few service personnel at some nodes; or, a flexible allocation strategy: dynamically adjusting the allocation of service personnel according to real-time tourist flow and emergencies to increase flexibility and response capabilities; making adjustments to service personnel.

Images