Hotspot area recreation stress assessment and early warning method based on spatiotemporal trajectory of tourists

Abstract

The application discloses a kind of based on the hot spot area recreation stress evaluation and early warning method of tourist space-time track, by processing tourist space-time track and geographic information data, dynamic hot spot area and its space-time characteristics are identified;Further, the real-time recreation stress index is evaluated by fusing multidimensional index, and the space-time graph network is constructed to model the stress propagation relationship;Using pre-trained space-time graph prediction model, the predicted stress index and propagation path of future period can be obtained;Finally, based on the prediction result, generate graded early warning and macroscopic diversion plan, and publish personalized guide strategy through path induction algorithm in combination with individual location information.The application realizes the dynamic and accurate evaluation and prediction of scenic spot recreation stress, can prospectively identify stress conduction risk, and generate coordinated control strategy from macro plan to individual guidance, effectively improve the initiative and fine level of scenic spot management.

Description

Technical Field

[0001] This invention relates to the field of tourism management and data analysis technology, and in particular to a method for assessing and warning of recreational pressure in hotspot areas based on tourists' spatiotemporal trajectories. Background Technology

[0002] In the fields of tourism management and smart scenic area construction, effective monitoring and intervention of tourist distribution and flow within scenic areas are crucial for enhancing tourist experience and ensuring scenic area safety and ecological sustainability. With the widespread adoption of location technology and mobile internet, understanding tourist behavior patterns by collecting and analyzing spatiotemporal trajectory data has become an important technological tool in the industry. Existing solutions typically focus on statistical analysis and pattern mining of historical and real-time visitor flow data to predict future visitor numbers in specific areas. These methods can, to some extent, reveal tourist aggregation trends and migration patterns, providing a reference for macro-level decision-making by management.

[0003] However, this technological approach, centered on "number prediction" of visitor flow, still has limitations in practical, refined management and proactive control. First, its assessment dimensions are relatively singular, primarily focusing on the number of people, making it difficult to comprehensively reflect the multidimensional and complex pressures caused by tourist gatherings in a specific area, such as the combined impact on spatial comfort, facility capacity, the ecological environment, and the smoothness of the visitor experience. Second, existing methods often treat each area as an independent analytical unit or only consider static transfer probabilities, lacking effective modeling of the dynamic correlation and transmission effects of pressure generated by tourist flow between areas. This leads to management responses often lagging behind the actual development of the problem, and the resulting diversion measures are mostly general, broad-based announcements or restrictions, unable to accurately trace the source and provide personalized guidance in the early stages of pressure formation. Therefore, it is difficult to fundamentally optimize the spatiotemporal distribution of tourists and achieve proactive management to prevent problems before they arise. Summary of the Invention

[0004] In view of this, the purpose of this invention is to propose a method for assessing and warning of recreational pressure in hotspot areas based on the spatiotemporal trajectory of tourists. By constructing a model that integrates multidimensional pressure assessment and spatiotemporal network prediction, the method realizes the transformation from passive tourist flow prediction to active pressure control, and solves the problem in scenic area management of the difficulty in accurately assessing, predicting and disseminating dynamic recreational pressure and implementing personalized guidance.

[0005] To achieve the above-mentioned technical objectives, the technical solution adopted by this invention is: a method for assessing and issuing early warnings of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories, comprising:

[0006] Acquire spatiotemporal trajectory data of tourists within the target area and geographic information data of the target area. The spatiotemporal trajectory data includes a sequence of tourist locations with timestamps.

[0007] Spatiotemporal clustering and behavioral pattern recognition are performed on spatiotemporal trajectory data to generate a dynamic set of hotspot areas and their spatiotemporal characteristics. The spatiotemporal characteristics include the real-time tourist density, tourist stay duration and tourist transfer flow of each hotspot area within the target area.

[0008] Based on spatiotemporal characteristics and geographic information data, a multi-dimensional recreational pressure assessment model is constructed. The multi-dimensional recreational pressure assessment model calculates the real-time recreational pressure index of each hotspot area by integrating indicators such as spatial congestion, facility load, ecological disturbance, and experience smoothness.

[0009] Based on the real-time recreational stress index and its historical sequence, a spatiotemporal network of graphs representing the relationship between inter-regional population flow and stress transmission is constructed.

[0010] The current state of the spatiotemporal graph network is input into the pre-trained spatiotemporal graph prediction model to obtain the predicted recreational pressure index and pressure propagation path of each hotspot area in the future preset time period.

[0011] Based on the comparison between the predicted recreational stress index and the preset warning threshold, a graded warning information is generated, and a macro-level diversion plan is generated by matching the stress propagation path with the contingency plan knowledge base.

[0012] Based on the macro-level traffic management plan and real-time individual tourist location information, personalized tour guide strategies are generated and released through path guidance algorithms to adjust the spatial and temporal distribution of tourists.

[0013] In some embodiments, spatiotemporal clustering and behavioral pattern recognition are performed on spatiotemporal trajectory data to generate a dynamic hotspot region set and its spatiotemporal characteristics, including:

[0014] Preprocessing the spatiotemporal trajectory data yields a standardized set of tourist spatiotemporal location points;

[0015] Based on the spatiotemporal location set of tourists, a spatiotemporal density clustering algorithm is used for cluster analysis to identify tourist clusters with spatiotemporal continuity, and the spatiotemporal boundary of each tourist cluster is defined as a candidate hotspot region.

[0016] Behavioral pattern analysis is performed on the trajectory point sequence within each candidate hotspot area. The behavioral pattern analysis includes calculating the average moving speed, dwell determination, and consistency of moving direction of the trajectory points.

[0017] Based on the results of behavioral pattern analysis, candidate hotspot areas are semantically classified. Areas where dwelling behavior is dominant and the duration exceeds a preset duration threshold are classified as static hotspot areas, while areas where the movement speed drops sharply or where there are obvious convergence inflow characteristics are classified as dynamic hotspot areas. Static hotspot areas and dynamic hotspot areas together constitute a set of dynamic hotspot areas.

[0018] Extract the spatiotemporal features of each region in the dynamic hotspot region set;

[0019] Among them, the real-time tourist density is calculated by the ratio of the number of trajectory points in each region of the dynamic hotspot area set to the area of ​​the region; the tourist stay duration is obtained by statistically analyzing the time span of the trajectory points identified as staying in each region of the dynamic hotspot area set; and the tourist transfer flow direction is determined by analyzing the region that the trajectory points leaving the current region enter at the next moment.

[0020] In some embodiments, a multi-dimensional recreational stress assessment model is constructed based on spatiotemporal characteristics and geographic information data, including:

[0021] Based on geographic information data, we define assessment dimensions and basic parameters for each hotspot area. The assessment dimensions include spatial congestion, facility load, ecological disturbance, and experience smoothness. The basic parameters include spatial safety capacity, service capacity of key service facilities, boundary of ecologically sensitive area, and free flow speed of path, which correspond to each dimension.

[0022] Based on spatiotemporal characteristics, quantitative calculation sub-models for each evaluation dimension are constructed. These sub-models include a spatial congestion sub-model, a facility load sub-model, an ecological disturbance sub-model, and an experience smoothness sub-model.

[0023] Among them, the spatial congestion sub-model is constructed based on the ratio of real-time tourist density to spatial safety capacity; the facility load sub-model is constructed based on the identification results of queuing behavior around the facility and the supply and demand relationship of service capacity; the ecological disturbance sub-model is constructed based on the frequency of trajectory intrusion into the boundary of ecologically sensitive area; and the experience smoothness sub-model is constructed based on the ratio of average tourist movement speed to free flow speed.

[0024] Define the output normalization function of the quantitative calculation sub-model for each evaluation dimension, so that the output of the quantitative calculation sub-model for each evaluation dimension is normalized dimension index.

[0025] Define a weighted aggregation function that integrates various normalized dimension indicators to generate a comprehensive stress value;

[0026] By integrating the assessment dimensions, basic parameters, quantitative calculation sub-models for each assessment dimension, output normalization function, and weighted aggregation function, a multi-dimensional recreational stress assessment model is obtained.

[0027] In some embodiments, the real-time recreational stress index of each hotspot area is calculated, including:

[0028] The real-time tourist density in the spatiotemporal characteristics is input into the spatial congestion sub-model to calculate the spatial congestion index value.

[0029] By inputting the tourist stay distribution and the location of key service facilities in the spatiotemporal characteristics into the facility load sub-model, the facility load index value is calculated.

[0030] The spatiotemporal trajectory data and the boundary of the ecologically sensitive area are input into the ecological disturbance degree sub-model to calculate the ecological disturbance degree index value.

[0031] The average moving speed of tourists and the free flow speed of the path in the spatiotemporal characteristics are input into the experience smoothness sub-model to calculate the experience smoothness index value.

[0032] The space congestion index, facility load index, ecological disturbance index, and experience smoothness index are processed according to the output normalization function to obtain the index values ​​of each normalized dimension.

[0033] The normalized dimensional index values ​​are input into a weighted aggregation function for calculation, and the real-time recreational pressure index of each hotspot area is output.

[0034] In some embodiments, a spatiotemporal graph network characterizing the relationship between inter-regional population flow and stress propagation is constructed based on the real-time recreational stress index and its historical sequence, including:

[0035] Each hotspot region in the dynamic hotspot region set is abstracted as a node in a spatiotemporal graph network;

[0036] Based on the tourist transfer flow direction in the spatiotemporal characteristics, the tourist transfer intensity between any two hotspot areas within a preset time window is calculated, and the tourist transfer intensity is used as the weight of the edge connecting the corresponding two nodes.

[0037] Based on spatiotemporal trajectory data, the time spent by tourists transferring between popular areas is analyzed, and the transfer time is used as the time delay attribute of the corresponding edge.

[0038] The real-time recreational pressure index and its historical sequence of each hotspot area are used as the state feature vector of the corresponding node.

[0039] External influencing factor nodes are incorporated into the spatiotemporal graph network. These nodes include transportation hub status nodes and weather event nodes. Connection edges between external influencing factor nodes and relevant hotspot area nodes are established based on causal relationships.

[0040] In some embodiments, the current state of the spatiotemporal graph network is input into a pre-trained spatiotemporal graph prediction model to obtain the predicted recreational stress index and stress propagation path for each hotspot area within a preset future time period, including:

[0041] The state feature vectors of all nodes in the spatiotemporal graph network, the weight attributes and time delay attributes of the edges are used together as the input of the spatiotemporal graph prediction model.

[0042] By using graph convolutional layers in the spatiotemporal graph prediction model, spatial dependency modeling is performed on the topology and node characteristics of the spatiotemporal graph network to capture the pressure correlation between hotspot regions.

[0043] By using the temporal modeling layer in the spatiotemporal graph prediction model, the historical sequence of node feature vectors is modeled for time dependence, so as to learn the trend and periodic pattern of recreational pressure evolution over time.

[0044] By utilizing the attention mechanism in the spatiotemporal graph prediction model, the importance weights of different nodes and different historical moments to the current prediction task are dynamically calculated.

[0045] Based on spatial dependence modeling, temporal dependence modeling, and importance weights, a predicted recreational pressure index for each hotspot area is generated within a preset future time period.

[0046] Meanwhile, based on the results of spatial dependence modeling, the spatiotemporal graph prediction model outputs pressure propagation paths that characterize the probability and direction of pressure diffusion among nodes in the spatiotemporal graph network.

[0047] In some embodiments, the attention mechanism in the spatiotemporal graph prediction model is used to dynamically calculate the importance weights of different nodes and different historical moments to the current prediction task, including:

[0048] Based on the topology and node characteristics of the spatiotemporal graph network, the spatial association strength weight between any two nodes is calculated through the spatial attention module. The spatial association strength weight is used to reflect the contribution of the state change of one node to the future pressure of another node.

[0049] Based on the historical sequence of node feature vectors, the temporal correlation strength weight of different historical moments to the prediction of future target moments is calculated through the time attention module. The temporal correlation strength weight reflects the contribution of the state of the historical moment to the prediction of future trends.

[0050] Spatial correlation strength weights and temporal correlation strength weights are fused to generate importance weights for feature aggregation in guiding graph convolutional layers and temporal modeling layers.

[0051] In some embodiments, a tiered early warning system is generated by comparing a predicted recreational stress index with a preset warning threshold, including:

[0052] The predicted recreational stress index is compared with the preset multi-level stress thresholds, which correspond to different stress levels.

[0053] Based on the comparison results, the pressure level that each hotspot area will reach in the future within a preset time period is determined;

[0054] If the pressure level exceeds the preset warning trigger level, a warning event will be generated that includes the warning area, warning level, estimated arrival time, and pressure propagation path.

[0055] Based on the warning level of the warning event, a corresponding warning release strategy is matched. The warning release strategy includes the release channels for warning information, the release content template, and the release priority.

[0056] Tiered early warning information is generated based on the early warning release strategy.

[0057] In some embodiments, a macro-level diversion plan is generated by matching the pressure propagation path with the contingency plan knowledge base, including:

[0058] The warning area, warning level and pressure propagation path in the warning event are matched with the triggering conditions of the contingency plan in the contingency plan knowledge base. The contingency plan knowledge base has standard diversion plans for different pressure scenarios.

[0059] Retrieve at least one standard evacuation plan that best matches the current warning event from the contingency plan knowledge base as the basic contingency plan;

[0060] Based on the real-time tourist distribution and tourist transfer flow in the spatiotemporal characteristics, the diversion parameters in the basic plan are dynamically optimized. The diversion parameters include the diversion direction, the activation time of the diversion path, and the recommended diversion rate.

[0061] Based on the optimized diversion parameters, an executable macro-level diversion plan is generated for the current early warning event. The macro-level diversion plan includes specific control measures, resource allocation suggestions, and expected diversion targets.

[0062] In some embodiments, based on the macro-level traffic management plan and real-time individual visitor location information, a personalized tour guide strategy is generated and published using a route guidance algorithm, including:

[0063] Obtain real-time individual tourist location information and the target areas and recommended routes defined in the macro-level traffic control plan;

[0064] With the target area for guidance as a constraint and the goal of minimizing the comprehensive recreational pressure index of the areas that individual tourists are expected to pass through in the future within a preset time period, a path guidance optimization model is constructed.

[0065] Real-time individual tourist location information, individual travel preferences, and route guidance optimization model are input into the route guidance algorithm. The route guidance algorithm is based on a reinforcement learning framework and calculates personalized recommended routes that meet the guidance of the macro-level traffic management plan.

[0066] The system compares the personalized recommended route with the current real-time route congestion status to generate navigation tips that include route change suggestions, alternative attraction recommendations, and the expected improvement in the experience.

[0067] Through the tourist terminal application, real-time guidance information is pushed to individual tourists who are located in or about to enter the warning area to guide them to adjust their tour routes.

[0068] By adopting the above technical solution, the present invention has the following beneficial effects compared with the prior art: It identifies dynamic hotspot areas and their spatiotemporal characteristics by processing tourist spatiotemporal trajectories and geographic information data; furthermore, it integrates multi-dimensional indicators to evaluate the real-time recreational pressure index and constructs a spatiotemporal graph network to model the pressure propagation relationship; using a pre-trained spatiotemporal graph prediction model, it obtains the predicted pressure index and propagation path for future periods; finally, it generates tiered early warning and macro-level guidance plans based on the prediction results, and releases personalized tour guide strategies through path guidance algorithms combined with individual location information. This invention achieves dynamic and accurate assessment and prediction of recreational pressure in scenic areas, proactively identifies the risk of pressure transmission, and generates coordinated control strategies from macro-level plans to individual guidance, effectively improving the initiative and refinement of scenic area management. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 This is a schematic diagram of steps S101 to S107 of the method described in the specific implementation embodiment;

[0071] Figure 2 This is a schematic diagram of steps S201 to S205 of the method described in the specific implementation embodiment;

[0072] Figure 3 This is a schematic diagram of steps S301 to S305 of the method described in the specific implementation. Detailed Implementation

[0073] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be particularly noted that the following embodiments are for illustrative purposes only and do not limit the scope of the invention. Similarly, the following embodiments are only some, not all, embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0074] Please see Figure 1This embodiment provides a method for assessing and issuing early warnings of recreational pressure in hotspot areas based on tourists' spatiotemporal trajectories, including:

[0075] S101. Obtain the spatiotemporal trajectory data of tourists within the target area and the geographic information data of the target area. The spatiotemporal trajectory data includes a sequence of tourist locations with timestamps.

[0076] S102. Perform spatiotemporal clustering and behavioral pattern recognition on the spatiotemporal trajectory data to generate a dynamic hotspot area set and its spatiotemporal characteristics. The spatiotemporal characteristics include the real-time tourist density, tourist stay duration and tourist transfer flow of each hotspot area within the target area.

[0077] S103. Based on spatiotemporal characteristics and geographic information data, a multi-dimensional recreational pressure assessment model is constructed. The multi-dimensional recreational pressure assessment model calculates the real-time recreational pressure index of each hotspot area by integrating spatial congestion, facility load, ecological disturbance and experience smoothness indicators.

[0078] S104. Based on the real-time recreational stress index and its historical sequence, construct a spatiotemporal network representing the relationship between inter-regional population flow and stress transmission;

[0079] S105. Input the current state of the spatiotemporal graph network into the pre-trained spatiotemporal graph prediction model to obtain the predicted recreational pressure index and pressure propagation path of each hotspot area in the future preset time period.

[0080] S106. Based on the comparison between the predicted recreational stress index and the preset warning threshold, generate graded warning information, and match it with the contingency plan knowledge base according to the stress propagation path to generate a macro-level diversion plan.

[0081] S107. Based on the macro-level traffic management plan and real-time individual tourist location information, generate and release personalized tour guide strategies through path guidance algorithms to adjust the spatial and temporal distribution of tourists.

[0082] In step S101, spatiotemporal trajectory data is continuously collected through positioning infrastructure deployed in the target area, forming a sequence of latitude and longitude coordinates with precise timestamps. Geographic information data is extracted from digital maps and planning documents of the target area, covering the spatial distribution and attribute information of geographic elements such as roads, buildings, vegetation, and water bodies. The two types of data acquired in this step together constitute the dynamic behavior and static environment foundation required for subsequent analysis.

[0083] In step S102, the spatiotemporal clustering process aims to identify areas where tourists continuously gather in both spatial and temporal dimensions. This process is achieved by analyzing the density distribution of location sequences in a spatiotemporal coordinate system, distinguishing between instantaneous waypoints and stable activity areas. Behavioral pattern recognition further analyzes the activity characteristics of tourists within these areas. For example, it distinguishes between movement and stillness by analyzing the time intervals and distance changes between adjacent trajectory points, and analyzes their relocation intentions by tracking their destinations after leaving a certain area. Finally, this step outputs a series of dynamically changing areas and their quantitative characteristics, which directly reflect the density of tourist distribution, the continuity of activity, and the tendency for spatial movement.

[0084] In step S103, the key to constructing the multi-dimensional recreational stress assessment model lies in defining and quantifying several independent stress influencing factors. Spatial congestion reflects the matching relationship between the number of tourists and available physical space; facility load focuses on the ratio of tourists' demand for public service facilities such as rest, sanitation, and transportation to the supply capacity; ecological disturbance assesses the impact of tourist activity trajectories on ecologically fragile areas; and experience smoothness characterizes the ease of passage felt by tourists moving within the area. This model determines the calculation benchmarks for each dimension based on geographic information data and designs algorithms to map the spatiotemporal characteristics obtained in step S102 into index values ​​for each dimension. These values ​​are then aggregated into a comprehensive real-time recreational stress index through weighted fusion and other methods. This step realizes the transformation from descriptive behavioral characteristics to comprehensive load assessment.

[0085] In step S104, the spatiotemporal graph network is constructed using the hotspot areas identified in step S102 as basic units. Each area is considered a node in the network, and the state of a node is characterized by the real-time recreational stress index of that area and its time-varying sequence. If there is a significant visitor transfer relationship between two areas (i.e., a visitor leaves one area and enters another), a connection edge is established between the corresponding nodes. In this way, discrete stress assessment points are integrated into a network topology that reflects the interaction of people flow between areas and the potential transmission of stress.

[0086] In step S105, the pre-trained spatiotemporal graph prediction model is a machine learning model capable of handling both graph-structured spatial relationships and time-series features. This model is trained using historical data and can learn the evolution and propagation patterns of recreational stress over time within the network structure. By inputting the spatiotemporal graph network, which includes the current node state and edge connections, into this model, the trend of stress index changes for each node (i.e., hotspot areas) within a specific future time period can be inferred, and the most likely path sequence for stress to spread within the network can be identified.

[0087] In step S106, a preset early warning threshold is defined based on the safety management standards and historical experience data of the target area, used to classify continuous predicted pressure indices into different risk levels. Through comparison, the system automatically determines the risk level of each hotspot area in the future and generates early warning information including elements such as region, level, and time. The contingency plan knowledge base pre-stores standardized management response plans for different pressure scenarios. Based on the risk diffusion direction revealed by the early warning information, especially the pressure propagation path, the system matches and retrieves applicable contingency plan frameworks from the knowledge base, and then generates a macro-level guidance plan containing specific measures and suggestions.

[0088] In step S107, the path guidance algorithm optimizes based on the visitor flow control targets set in the macro-level guidance plan generated in step S106. This algorithm combines the real-time, precise location of each visitor to calculate a tour route with lower overall recreational pressure within the expected timeframe. The generated personalized tour guide strategy is pushed out in real-time and precisely to the smart terminals held by visitors, thereby guiding individual visitor behavior and collaboratively achieving the visitor flow distribution adjustment effect desired by the macro-level plan at the micro-level.

[0089] It should be noted that the scope of the target area in this method can be set according to actual management needs, ranging from a single scenic spot to a city area containing multiple attractions. Hotspot areas are dynamically identified core areas of tourist activity within the target area; these could be a single attraction or a functional area within an attraction. The macro-level guidance plan is a holistic management strategy developed for one or more related hotspot areas within the target area; personalized guidance strategies are action suggestions generated for individual tourists within specific affected hotspot areas, based on the regulatory guidance of the macro-level guidance plan. Therefore, this solution achieves a seamless transition from overall regional management to individual behavioral guidance.

[0090] This embodiment transforms discrete tourist trajectory data into quantifiable and correlated regional pressure situations by constructing a multi-dimensional pressure assessment model and a spatiotemporal graph network. It then uses a graph prediction model to predict future pressure and its transmission paths. Based on this, the solution not only generates early warnings and contingency plans for managers but also decomposes macro-management objectives into actionable individual guidance strategies through path guidance, thus forming a complete closed loop from situational awareness and early warning to precise intervention. This embodiment changes the traditional management model that relies on lagging statistics and experience-based judgment, achieving proactive, refined, and pre-emptive control of recreational pressure.

[0091] Please see Figure 2 In some embodiments, spatiotemporal clustering and behavioral pattern recognition are performed on spatiotemporal trajectory data to generate a dynamic set of hotspot regions and their spatiotemporal characteristics, including:

[0092] S201. Preprocess the spatiotemporal trajectory data to obtain a standardized set of tourist spatiotemporal location points;

[0093] S202. Based on the set of tourist spatiotemporal location points, a spatiotemporal density clustering algorithm is used for cluster analysis to identify tourist clusters with spatiotemporal continuity, and the spatiotemporal boundary of each tourist cluster is defined as a candidate hotspot region. The cluster analysis process of the spatiotemporal density clustering algorithm includes:

[0094] Define a spatiotemporal neighborhood radius parameter, which includes a spatial neighborhood radius and a temporal neighborhood radius;

[0095] Based on the spatial neighborhood radius and the temporal neighborhood radius, a spatiotemporal distance metric function is constructed to calculate the spatiotemporal distance between any two tourist spatiotemporal locations.

[0096] Based on the spatiotemporal distance metric function and the preset minimum neighborhood point threshold, density accessibility analysis is performed on the tourist spatiotemporal location point set, and spatiotemporal location points that meet the density accessibility condition are grouped into the same tourist cluster.

[0097] The set of tourist spatiotemporal location points is traversed to identify all tourist clusters that meet the density accessibility analysis conditions, and noise clusters with fewer spatiotemporal location points than the preset clustering size threshold are removed.

[0098] Based on the spatial coordinates and timestamps of all spatiotemporal location points within each tourist cluster, their spatial and temporal boundaries are calculated respectively. The spatial boundary is determined by the convex hull algorithm or the minimum bounding rectangle algorithm, and the temporal boundary is determined by the earliest and latest timestamps within the cluster.

[0099] The spatial boundary and the temporal boundary are jointly defined as the spatiotemporal boundary of the tourist cluster, i.e., the candidate hotspot area;

[0100] S203. Perform behavioral pattern analysis on the trajectory point sequence within each candidate hotspot area. The behavioral pattern analysis includes calculating the average moving speed, dwell determination, and consistency of moving direction of the trajectory points.

[0101] S204. Based on the results of behavioral pattern analysis, the candidate hotspot areas are semantically classified. Among them, areas where dwelling behavior is dominant and the duration exceeds the preset duration threshold are classified as static hotspot areas, and areas where the movement speed drops sharply or the convergence inflow characteristics are clearly classified as dynamic hotspot areas. Static hotspot areas and dynamic hotspot areas together constitute a set of dynamic hotspot areas.

[0102] S205. Extract the spatiotemporal features of each region in the dynamic hotspot region set;

[0103] Among them, the real-time tourist density is calculated by the ratio of the number of trajectory points in each region of the dynamic hotspot area set to the area of ​​the region; the tourist stay duration is obtained by statistically analyzing the time span of the trajectory points identified as staying in each region of the dynamic hotspot area set; and the tourist transfer flow direction is determined by analyzing the region that the trajectory points leaving the current region enter at the next moment.

[0104] In step S201, the preprocessing operations include data cleaning, coordinate system one, and timestamp alignment. Data cleaning filters out abnormal trajectory points that significantly deviate from the road network or temporal logic based on preset spatial and temporal reasonableness thresholds; coordinate system one converts data from different positioning devices or coordinate systems to a unified geographic reference system, such as the WGS-84 coordinate system; timestamp alignment calibrates the time information of all data to the same time base and unifies the time format and accuracy. After the above processing, a standardized set of tourist spatiotemporal location points is formed, ensuring the quality and consistency of the input data for subsequent algorithms.

[0105] In step S202, the implementation of the spatiotemporal density clustering algorithm relies on the definition of spatiotemporal neighborhood. The spatial neighborhood radius is typically set based on spatial characteristics such as road width and plaza scale of the target area, while the temporal neighborhood radius is determined based on the data sampling frequency and reasonable time intervals for continuous tourist activities. The spatiotemporal distance metric function combines spatial Euclidean distance and time difference in a weighted manner, with its weight coefficients reflecting the relative importance of spatial proximity and temporal continuity in clustering. The minimum neighborhood point threshold is used to define the minimum number of tourists required to form an effective cluster, and this threshold can be dynamically adjusted according to the overall tourist density of the area. Density accessibility analysis is based on the above parameters and can effectively identify tourist groups that are physically dense and temporally coherent. For each identified tourist cluster, its spatial boundary is defined by calculating the convex hull or minimum bounding rectangle of all points within the cluster, thus obtaining a polygonal region; the temporal boundary is determined by the extreme timestamp values ​​of the points within the cluster, resulting in a time interval. Together, these define a clear spatiotemporal range for candidate hotspot areas.

[0106] In step S203, behavioral pattern analysis calculates multiple indicators based on the trajectory point sequence. Average movement speed is obtained by statistically averaging the displacement and time difference between consecutive trajectory points within the region. Dwell time determination is achieved by setting speed thresholds and minimum duration thresholds: when the instantaneous speeds of multiple consecutive trajectory points are all below the speed threshold, and their total duration exceeds the minimum duration threshold, the trajectory segment is determined to be a dwell time. Consistency of movement direction is quantified by calculating the standard deviation or variance of the direction angles of consecutive trajectory segments; low variance values ​​indicate concentrated movement directions, while high variance values ​​indicate dispersed directions. These quantitative indicators provide an objective basis for the semantic classification of candidate hotspot regions.

[0107] In step S204, semantic classification is performed based on the quantitative indicators obtained in step S203. The criteria for determining static hotspot areas are: the proportion of trajectory points identified as stationary within the area exceeds a preset percentage threshold, and the total stationary time exceeds a preset duration threshold. The criteria for determining dynamic hotspot areas focus on abrupt changes or convergence characteristics in movement states. For example, the average moving speed of trajectory points within the area decreases by a percentage exceeding a preset threshold compared to their speed before entering the area, or streamline analysis reveals that trajectories from multiple directions converge towards the area. The classification process assigns candidate hotspot areas with different behavioral characteristics to either static or dynamic categories, collectively forming a dynamic hotspot area set. This classification result directly affects the setting of weights for different dimensions in the subsequent stress assessment model.

[0108] In step S205, the extraction of spatiotemporal features is based on the classified set of dynamic hotspot areas and their internal trajectory data. The calculation of real-time visitor density first requires determining the statistical time granularity (e.g., per minute), counting the number of independent visitor trajectory points appearing within the area during that time period, and then dividing by the area area. For the statistics of visitor dwell time, for each trajectory segment determined as a dwelling point, the difference between its start and end timestamps is calculated, and the durations of all dwelling segments within the area are summed or averaged. The analysis of visitor transfer flows requires establishing time-series correlations between trajectory points. For trajectory points leaving the spatial boundary of a certain area, the new area they first entered is searched among the trajectory points at subsequent times, thus establishing transfer relationship pairs from the source area to the target area. This step ultimately outputs a set of feature vectors with clear physical meaning, bound to each hotspot area.

[0109] This embodiment addresses the problem of traditional spatial clustering neglecting temporal continuity by introducing a spatiotemporal density clustering algorithm, enabling more accurate identification of continuously existing core areas of tourist activity. Furthermore, by designing quantified behavioral pattern analysis indicators and semantic classification rules, it achieves automatic discrimination of hotspot area functions, distinguishing between static areas primarily characterized by prolonged stays and dynamic areas primarily characterized by changes in visitor flow. This fine-grained identification and classification provides a precise and semantically rich data foundation for subsequently constructing a realistic, multi-dimensional stress assessment model, a prerequisite for achieving high-precision early warning and personalized guidance.

[0110] Please see Figure 3 In some embodiments, a multi-dimensional recreational stress assessment model is constructed based on spatiotemporal characteristics and geographic information data, including:

[0111] S301. Based on geographic information data, define the assessment dimensions and basic parameters for each hotspot area. The assessment dimensions include spatial congestion, facility load, ecological disturbance, and experience smoothness. The basic parameters include spatial safety capacity, service capacity of key service facilities, boundary of ecologically sensitive area, and free flow speed of path, corresponding to each dimension.

[0112] S302. Based on spatiotemporal characteristics, construct quantitative calculation sub-models for each evaluation dimension. The quantitative calculation sub-models for each evaluation dimension include spatial congestion sub-model, facility load sub-model, ecological disturbance sub-model, and experience smoothness sub-model.

[0113] Among them, the spatial congestion sub-model is constructed based on the ratio of real-time tourist density to spatial safety capacity; the facility load sub-model is constructed based on the identification results of queuing behavior around the facility and the supply and demand relationship of service capacity; the ecological disturbance sub-model is constructed based on the frequency of trajectory intrusion into the boundary of ecologically sensitive area; and the experience smoothness sub-model is constructed based on the ratio of average tourist movement speed to free flow speed.

[0114] S303. Define the output normalization function of the quantitative calculation sub-model for each evaluation dimension, so that the output of the quantitative calculation sub-model for each evaluation dimension is normalized dimension index.

[0115] S304. Define a weighted aggregation function that integrates the normalized dimension indicators to generate a comprehensive stress value.

[0116] S305. The assessment dimensions, basic parameters, quantitative calculation sub-models for each assessment dimension, output normalization function, and weighted aggregation function are integrated to obtain a multi-dimensional recreational stress assessment model.

[0117] In step S301, assessment dimensions and basic parameters are defined based on geographic information data. Spatial safety capacity is determined comprehensively based on the physical area, terrain complexity, and safety evacuation standards of each hotspot area, for example, by multiplying the maximum allowable number of people per unit area by the area. The service capacity of critical service facilities refers to the maximum instantaneous number of people served or the throughput per unit time for facilities such as sightseeing vehicles, cable cars, and restrooms within the area; this data is derived from the facility's design parameters or operational statistics. The boundary of ecologically sensitive areas is obtained from environmental protection planning data and defined in the form of polygonal geofences. Path free-flow speed refers to the typical walking speed of tourists on the path under conditions of no congestion or interference; this can be obtained through statistical analysis of historical trajectory data during low-density periods or by setting empirical values ​​based on path type (e.g., steps, flat roads). These basic parameters provide an objective benchmark for subsequent quantitative calculations.

[0118] In step S302, quantitative calculation sub-models for each evaluation dimension are constructed. The spatial congestion sub-model obtains a ratio representing the space occupancy rate by dividing the real-time visitor density by the spatial safety capacity. The facility load sub-model first identifies queuing behavior around the facility, which can be achieved by analyzing the clustering density, dwell time, and consistency of movement direction of trajectory points near the facility entrance; the identified number of people queuing (or estimated waiting time) is compared with the service capacity of the facility to calculate the load rate. The ecological disturbance sub-model is constructed by counting the number of times visitor trajectory points fall within the boundary polygon of the ecologically sensitive area per unit time; the higher the frequency, the stronger the disturbance. The experience smoothness sub-model is constructed by calculating the ratio of the average movement speed of visitors to the free flow speed of the path; the closer the ratio is to 1, the higher the smoothness; the smaller the ratio, the more severe the congestion. Each sub-model takes the corresponding spatiotemporal characteristics and geographic information parameters as input and outputs a raw index value reflecting the pressure level of that dimension.

[0119] In step S303, the original index values ​​output by each dimension's sub-model have different dimensions and ranges, making direct comprehensive comparison and weighting impossible. Therefore, a normalization function is used to map these heterogeneous indices to a unified numerical range. For example, a minimum-maximum normalization method can be used to linearly transform the original values ​​to the target range. For the space congestion ratio, a piecewise function can also be used for normalization; when the ratio is less than 1, it is mapped to a lower value, and when it exceeds 1, the mapped value increases rapidly to reflect the nonlinear effect of overload. The design of the normalization function needs to consider the actual physical meaning of each dimension's index and its sensitivity to pressure growth.

[0120] In step S304, the weighted aggregation function is used to merge the multiple normalized dimensional indicators obtained in step S303 into a comprehensive stress value, namely the real-time recreational stress index. Preferably, the weighted aggregation adopts the form of linear weighted sum, that is, each normalized dimensional indicator is assigned a weight coefficient, and the weighted sum of all indicators is used as the comprehensive output. The weight coefficients can be determined based on expert experience, analytic hierarchy process (AHP), or regression analysis based on historical stress event data to reflect the differences in the importance of different dimensions to the overall stress. For example, in ecologically fragile scenic areas, the weight of ecological disturbance may be higher; while in theme parks, the weights of facility load and experience smoothness may be more critical.

[0121] In step S305, the evaluation dimensions, basic parameters, various quantitative calculation sub-models, output normalization functions, and weighted aggregation functions defined in the preceding steps are integrated and encapsulated through software code or model configuration files to form a complete and executable multi-dimensional recreational stress assessment model. This model receives spatiotemporal features and geographic information data as input, and after calculation, normalization, and weighted aggregation processes by each sub-model, it finally outputs the real-time recreational stress index for each hotspot area.

[0122] This embodiment constructs a multi-dimensional recreational stress assessment model. It designs four mutually independent assessment dimensions with clear physical meaning and establishes a complete calculation chain from raw data to indicators of each dimension, and finally to a comprehensive index. By introducing normalization and weighted aggregation mechanisms, it effectively solves the problem of the fusion of multi-source heterogeneous indicators, enabling the final stress index to scientifically and rationally reflect the complex load conditions of space, facilities, ecology, and experience. This model provides a customizable and calculable core tool for accurately quantifying the stress of scenic areas.

[0123] In some embodiments, the real-time recreational stress index of each hotspot area is calculated, including:

[0124] The real-time tourist density in the spatiotemporal characteristics is input into the spatial congestion sub-model to calculate the spatial congestion index value.

[0125] By inputting the tourist stay distribution and the location of key service facilities in the spatiotemporal characteristics into the facility load sub-model, the facility load index value is calculated.

[0126] The spatiotemporal trajectory data and the boundary of the ecologically sensitive area are input into the ecological disturbance degree sub-model to calculate the ecological disturbance degree index value.

[0127] The average moving speed of tourists and the free flow speed of the path in the spatiotemporal characteristics are input into the experience smoothness sub-model to calculate the experience smoothness index value.

[0128] The space congestion index, facility load index, ecological disturbance index, and experience smoothness index are processed according to the output normalization function to obtain the index values ​​of each normalized dimension.

[0129] The normalized dimensional index values ​​are input into a weighted aggregation function for calculation, and the real-time recreational pressure index of each hotspot area is output.

[0130] In this embodiment, the process of calculating the real-time recreational pressure index is a specific process of executing the constructed multi-dimensional recreational pressure assessment model. The real-time visitor density in the spatiotemporal characteristics is used as input and substituted into the spatial congestion sub-model. The sub-model obtains the spatial congestion index value by performing the division operation according to its internally defined calculation rules, such as dividing the real-time visitor density by the pre-set spatial safety capacity of the area. This value is usually a number greater than or equal to zero.

[0131] The calculation of facility load index requires combining visitor dwell distribution and key service facility location information. Visitor dwell distribution data identifies the locations where visitors stayed and the duration of their stays. Through spatial correlation analysis, dwell points clustered within specific ranges around key service facilities are identified; these dwell points can be inferred to be visitors queuing or waiting. The number of visitors staying within this range or their total waiting time is estimated, and this value is then compared with the corresponding facility's service capacity (e.g., maximum number of people served per unit time). The facility load index is calculated through division or proportion.

[0132] The calculation of the ecological disturbance index depends on spatiotemporal trajectory data and the boundaries of ecologically sensitive areas. The system iterates through the input trajectory points, determines their geospatial relationships (such as whether a point is within a polygon), and counts the total number of times the trajectory points fall within the boundary of ecologically sensitive areas within a specified time window. This count is the raw output of the ecological disturbance index; a higher count indicates more frequent human encroachment on the ecological area.

[0133] The experience smoothness index is calculated by comparing the average tourist movement speed in the spatiotemporal characteristics with the free-flow speed of the corresponding path. This sub-model divides the average tourist movement speed by the free-flow speed of the path to obtain a ratio. The closer this ratio is to 1, the smoother the tourist movement; the smaller the ratio, the more severe the congestion, thus obtaining the experience smoothness index value.

[0134] After obtaining the four original indicator values, predefined output normalization functions are called to process them respectively. For example, for the space congestion index value, if its original value is the occupancy ratio, a piecewise linear function can be used to map it to the interval between 0 and 100: when the ratio is less than 1, it is linearly mapped to a lower score interval; when the ratio is greater than 1, the mapped score increases more rapidly. After each indicator value is processed by its respective normalization function, it is transformed into a normalized dimensional indicator value with uniform dimensions, consistent range, and comparability.

[0135] Using all normalized dimensional index values ​​as input, a weighted aggregation function is called to sum the values ​​of each normalized dimensional index according to a preset weight coefficient for each dimension. The weight coefficient reflects the contribution of different dimensional pressures to the overall recreational pressure index, and their sum is usually 1. The result of the weighted sum is the real-time recreational pressure index of the hotspot area. This index is a comprehensive scalar value, and its magnitude directly reflects the overall recreational pressure level currently experienced by the area.

[0136] This embodiment utilizes a pre-constructed multi-dimensional assessment model to calculate a real-time recreational stress index. By sequentially calling the sub-models of each dimension, the normalization function, and the weighted aggregation function, the original, heterogeneous tourist behavior characteristics and geographical environmental parameters are systematically transformed into a standardized and comparable comprehensive stress index. This ensures the scientific nature and operability of the stress assessment and provides direct and quantitative input data for subsequent spatiotemporal network construction and prediction and early warning.

[0137] In some embodiments, a spatiotemporal graph network characterizing the relationship between inter-regional population flow and stress propagation is constructed based on the real-time recreational stress index and its historical sequence, including:

[0138] Each hotspot region in the dynamic hotspot region set is abstracted as a node in a spatiotemporal graph network;

[0139] Based on the tourist transfer flow direction in the spatiotemporal characteristics, the tourist transfer intensity between any two hotspot areas within a preset time window is calculated, and the tourist transfer intensity is used as the weight of the edge connecting the corresponding two nodes.

[0140] Based on spatiotemporal trajectory data, the time spent by tourists transferring between popular areas is analyzed, and the transfer time is used as the time delay attribute of the corresponding edge.

[0141] The real-time recreational pressure index and its historical sequence of each hotspot area are used as the state feature vector of the corresponding node.

[0142] External influencing factor nodes are incorporated into the spatiotemporal graph network. These nodes include transportation hub status nodes and weather event nodes. Connection edges between external influencing factor nodes and relevant hotspot area nodes are established based on causal relationships.

[0143] In this embodiment, the spatiotemporal graph network is constructed by first mapping each independent region in the dynamic hotspot region set to a node in the network. Each node has a unique identifier to represent that geographical region.

[0144] The weights of the edges connecting nodes are determined by calculating the intensity of tourist migration. This calculation relies on tourist migration flow data recorded in the spatiotemporal features. For a preset time window (e.g., the past 15 minutes), the number of independent tourists leaving the first area and entering the second area is counted. This number of migrating tourists is divided by the length of the time window to obtain the average number of migrations per unit time, which is used as the weight of the edge connecting the nodes of the two areas. The larger the weight value, the stronger the connection between the two areas.

[0145] The connecting edges also have a time lag attribute, which reflects the typical time it takes for tourists to move between the two areas. The time lag attribute is determined based on spatiotemporal trajectory data. This is done by tracking the trajectories of tourists who have completed the transfer from the first area to the second area, calculating the difference between the timestamp of their departure from the first area and the timestamp of their arrival in the second area, and statistically analyzing the time taken for all such transfers (e.g., taking the median or average). The result is used as the time lag attribute value for the corresponding edge.

[0146] The state feature vector of each node is composed of the real-time recreational stress index of the hotspot area corresponding to that node and its historical sequence. This vector not only contains the stress value at the current moment, but also includes the stress values ​​of several consecutive moments in chronological order, thus forming a feature sequence that can characterize the evolution of the stress state of the node over time.

[0147] To more comprehensively model the influencing factors of pressure propagation, external influencing factor nodes are also introduced into the spatiotemporal graph network. For example, transportation hub status nodes can represent the traffic status of the main entrances or connecting stations of a scenic area (e.g., smooth, congested, closed), and weather event nodes can represent whether the current weather conditions are special, such as rainfall or high temperatures. Directed connections are established between these external nodes and the hotspot area nodes affected by them. The direction of the edge indicates the direction of the influence, and the weight of the edge can be set based on the influence intensity obtained from historical data analysis. For example, heavy rain has a significant positive impact on the pressure of outdoor viewing areas.

[0148] This embodiment constructs a structured spatiotemporal graph network through the steps described above. This network not only accurately characterizes the spatial relationships and interaction strength between hotspot areas based on actual pedestrian flow through nodes and weighted edges, but also records the dynamic changes in pressure through the temporal state vectors of nodes, and enhances the model's ability to characterize the impact of complex environments by introducing external factor nodes. This networked and structured representation method lays an essential data foundation for subsequent accurate prediction of pressure propagation using advanced models such as graph neural networks.

[0149] In some embodiments, the current state of the spatiotemporal graph network is input into a pre-trained spatiotemporal graph prediction model to obtain the predicted recreational stress index and stress propagation path for each hotspot area within a preset future time period, including:

[0150] The state feature vectors of all nodes in the spatiotemporal graph network, the weight attributes and time delay attributes of the edges are used together as the input of the spatiotemporal graph prediction model.

[0151] By using graph convolutional layers in the spatiotemporal graph prediction model, spatial dependency modeling is performed on the topology and node characteristics of the spatiotemporal graph network to capture the pressure correlation between hotspot regions.

[0152] By using the temporal modeling layer in the spatiotemporal graph prediction model, the historical sequence of node feature vectors is modeled for time dependence, so as to learn the trend and periodic pattern of recreational pressure evolution over time.

[0153] By utilizing the attention mechanism in the spatiotemporal graph prediction model, the importance weights of different nodes and different historical moments to the current prediction task are dynamically calculated.

[0154] Based on spatial dependence modeling, temporal dependence modeling, and importance weights, a predicted recreational pressure index for each hotspot area is generated within a preset future time period.

[0155] Meanwhile, based on the results of spatial dependence modeling, the spatiotemporal graph prediction model outputs pressure propagation paths that characterize the probability and direction of pressure diffusion among nodes in the spatiotemporal graph network.

[0156] In this embodiment, the pre-trained spatiotemporal graph prediction model is a machine learning model that integrates graph neural networks and sequence learning capabilities. The training process of this model utilizes a sequence of spatiotemporal graph networks constructed within a historical time period as training samples. Specifically, the spatiotemporal graph network (including node states, edge weights, and time delays) from multiple consecutive historical moments is used as input features, and the actual recreational pressure index of each node at a future moment is used as the training label. Through supervised learning, the model learns to predict future node states from historical network states. The training process optimizes the model's internal parameters using the backpropagation algorithm to minimize the error between predicted and actual values.

[0157] When the current state of the spatiotemporal graph network is input into the trained model, the model first processes it through its internal graph convolutional layers. These layers aggregate and update the state feature vector of each node based on the connections and weights between nodes in the network. This process ensures that each node's new features not only include its own historical stress information but also incorporate stress information from neighboring nodes with which it has passenger flow connections, thus enabling the modeling of stress relationships on the spatial network.

[0158] Meanwhile, the temporal modeling layer in the model (e.g., using a recurrent neural network or a temporal convolutional network structure) analyzes the historical sequence of the state feature vector of each node. This layer learns the pattern of stress index changes over time, such as upward trends, downward trends, and daytime peak patterns, thereby capturing the time dependence of stress evolution.

[0159] The model also integrates an attention mechanism, enabling dynamic evaluation of which other nodes in the network have a more critical impact on predicting the future pressure of a specific node, and which past states in the node's own historical sequence are more relevant. Based on these evaluations, different importance weights are assigned to different nodes and different historical moments.

[0160] Finally, the model integrates the spatial correlation features processed by the graph convolutional layer, the temporal evolution features extracted by the temporal modeling layer, and the dynamic weights provided by the attention mechanism. Through structures such as fully connected layers, it performs information fusion and transformation, and outputs an estimate of the recreational pressure index for each hotspot area in the future preset time period.

[0161] Furthermore, based on the spatial dependencies learned by the graph convolutional layers, the model can infer the diffusion path of pressure in the network. For example, by analyzing the strength and direction of feature propagation between nodes, the model can output a pressure propagation probability matrix, which indicates the likelihood and main direction of pressure spreading from one node to other nodes, thus forming the pressure propagation path.

[0162] This embodiment achieves accurate and quantitative predictions of future recreational stress and its propagation paths by introducing a pre-trained spatiotemporal graph prediction model. This model organically integrates the processing capabilities of graph-structured data and time-series data, capturing the transmission effects of stress on spatial networks and learning its evolution over time. Furthermore, an attention mechanism enhances the utilization of key information, significantly improving the predictive foresight and accuracy. This provides a core decision-making basis for implementing prediction-based proactive early warning and mitigation measures.

[0163] In some embodiments, the attention mechanism in the spatiotemporal graph prediction model is used to dynamically calculate the importance weights of different nodes and different historical moments to the current prediction task, including:

[0164] Based on the topology and node characteristics of the spatiotemporal graph network, the spatial association strength weight between any two nodes is calculated through the spatial attention module. The spatial association strength weight is used to reflect the contribution of the state change of one node to the future pressure of another node.

[0165] Based on the historical sequence of node feature vectors, the temporal correlation strength weight of different historical moments to the prediction of future target moments is calculated through the time attention module. The temporal correlation strength weight reflects the contribution of the state of the historical moment to the prediction of future trends.

[0166] Spatial correlation strength weights and temporal correlation strength weights are fused to generate importance weights for feature aggregation in guiding graph convolutional layers and temporal modeling layers.

[0167] In this embodiment, an attention mechanism is introduced to enhance the feature selection and fusion capabilities of the spatiotemporal graph prediction model. Specifically, the spatial attention module evaluates the dynamic correlation strength between any two nodes in the network in the spatial dimension. This module receives the current spatiotemporal graph network topology (i.e., node connection relationships) and the current state feature vectors of all nodes as input. Its calculation process does not rely on the original weights of the edges, but rather uses a learnable neural network to dynamically generate a spatial correlation strength weight based on the current feature similarity between the two nodes, the inherent properties of the connecting edges (such as weights and time delays), and other global information. This weight is a numerical value, representing the reference value of the current state of the source node when predicting the future state of the target node. For example, even if there is a direct tourist transfer between two areas, if one area experiences an abnormal stress state due to a special event during prediction, the spatial attention mechanism can reduce the contribution of this abnormal node to the prediction of other nodes.

[0168] The Time Attention module focuses on filtering key historical information from a temporal perspective. For the node to be predicted, this module analyzes the historical sequence of its state feature vectors. Through another learnable neural network, it evaluates the importance of the state features of each past moment in the sequence for predicting the state at the future target moment, and calculates a temporal correlation strength weight for each historical moment. This mechanism allows the model to adaptively focus on historical periods most relevant to future trends, such as paying more attention to recent points of rapid change or identifying periodically similar historical periods, while downplaying the influence of irrelevant or noisy moments.

[0169] The generated spatial association strength weights and temporal association strength weights need to be fused to form the final importance weights that guide the feature aggregation of the core layer of the model. A preferred fusion method is to multiply or perform a weighted sum. For example, when aggregating neighbor node information through a graph convolutional layer, the spatial association strength weight calculated for a specific connection edge can be combined with the temporal attention weight of that neighbor node at key historical moments to jointly adjust the contribution ratio of that neighbor node's features in the aggregation process. Similarly, when processing historical sequences in the temporal modeling layer, the temporal attention weights are directly used for weighted summation to highlight features at important moments.

[0170] This embodiment constructs separate spatial and temporal attention modules, enabling the dynamic and differentiated use of spatial correlation information and historical temporal information in the network, rather than applying equal weighting. This dynamic weight allocation capability allows the model to more accurately capture key influence paths and critical historical moments in pressure propagation, effectively improving the robustness and accuracy of predictions. It is a key component for enhancing the model's intelligent decision-making capabilities.

[0171] In some embodiments, a tiered early warning system is generated by comparing a predicted recreational stress index with a preset warning threshold, including:

[0172] The predicted recreational stress index is compared with the preset multi-level stress thresholds, which correspond to different stress levels.

[0173] Based on the comparison results, the pressure level that each hotspot area will reach in the future within a preset time period is determined;

[0174] If the pressure level exceeds the preset warning trigger level, a warning event will be generated that includes the warning area, warning level, estimated arrival time, and pressure propagation path.

[0175] Based on the warning level of the warning event, a corresponding warning release strategy is matched. The warning release strategy includes the release channels for warning information, the release content template, and the release priority.

[0176] Tiered early warning information is generated based on the early warning release strategy.

[0177] In this embodiment, the preset multi-level pressure thresholds are pre-set based on the safety management regulations of the target area, the distribution of historical pressure data, and the management response requirements under different pressure levels. For example, three levels of thresholds can be set: the first level threshold corresponds to the comfortable pressure level, the second level threshold corresponds to the level requiring attention for mild crowding, and the third level threshold corresponds to the level requiring immediate intervention for severe crowding. These thresholds divide the continuous range of predicted recreational pressure indexes into discrete pressure levels.

[0178] The predicted recreational stress index for each hotspot area is compared sequentially with the aforementioned multi-level stress thresholds. By determining which threshold range the index falls into, the expected stress level for that area within a preset future time period is determined. For example, if the predicted index is higher than the second-level threshold but lower than the third-level threshold, its stress level is determined to be level two.

[0179] The system presets a warning trigger level, which typically corresponds to the pressure level requiring a management response. When a region is determined to reach a pressure level equal to or higher than this warning trigger level, the warning event generation process is triggered. The generated warning event is a structured data object that must contain the following core fields: warning area (i.e., the hotspot area identifier reaching the warning level), warning level (i.e., the determined pressure level), estimated arrival time (i.e., the predicted time when the pressure index exceeds the threshold), and pressure propagation path information related to the region, output from the prediction model, indicating the direction in which the pressure may spread.

[0180] The early warning release strategy defines the specific rules for information release under different warning levels. Release channels specify the recipients and methods of information delivery. For example, a Level 1 warning may only be sent to the scenic area management backend, a Level 2 warning may also be pushed to on-site management personnel terminals, and a Level 3 warning may be further released to the public through public broadcasts, visitor apps, etc. Release content templates are predefined text, graphic, or audio templates for different levels, with reserved placeholders for dynamic information such as the warning area, level, and time. Release priority specifies the order in which information is processed when multiple warning events occur simultaneously; generally, the higher the warning level, the higher the priority.

[0181] Finally, based on the matched early warning release strategy, the dynamic information in the early warning event is filled into the corresponding release content template, and the final executable hierarchical early warning information is generated according to the specified release channels and priorities, completing the conversion from predictive data to operable instructions.

[0182] This embodiment transforms the predicted quantitative stress index into early warning levels and events with clear management implications by comparing it with preset thresholds. It further links these levels to specific information dissemination strategies, thus bridging the gap between data analysis and management actions. This ensures the timeliness, accuracy, and operability of early warning information and provides clear triggering conditions and action guidelines for subsequent evacuation plans and tourist guidance.

[0183] In some embodiments, a macro-level diversion plan is generated by matching the pressure propagation path with the contingency plan knowledge base, including:

[0184] The warning area, warning level and pressure propagation path in the warning event are matched with the triggering conditions of the contingency plan in the contingency plan knowledge base. The contingency plan knowledge base has standard diversion plans for different pressure scenarios.

[0185] Retrieve at least one standard evacuation plan that best matches the current warning event from the contingency plan knowledge base as the basic contingency plan;

[0186] Based on the real-time tourist distribution and tourist transfer flow in the spatiotemporal characteristics, the diversion parameters in the basic plan are dynamically optimized. The diversion parameters include the diversion direction, the activation time of the diversion path, and the recommended diversion rate.

[0187] Based on the optimized diversion parameters, an executable macro-level diversion plan is generated for the current early warning event. The macro-level diversion plan includes specific control measures, resource allocation suggestions, and expected diversion targets.

[0188] In this embodiment, the contingency plan knowledge base is a structured database storing various standard evacuation plans. Each standard evacuation plan is associated with a set of triggering conditions, which typically include: the applicable warning area type (e.g., entrances / exits, core observation decks), the warning level range, and the main direction in which pressure may spread. The matching process compares the specific attributes of the current warning event with the triggering conditions of each plan in the knowledge base and calculates the matching degree. The matching degree calculation can be based on rule matching or similarity calculation. For example, when the warning area type and warning level both meet the conditions of a certain plan, and the main diffusion direction indicated by the pressure propagation path is consistent with the preset response direction of the plan, the matching degree is considered high.

[0189] One or more standard congestion mitigation plans with the highest matching degree are retrieved from the contingency plan knowledge base to serve as the basis for generating the final macro-level congestion mitigation plan. These basic plans provide a proven and universal response framework for similar stress scenarios. For example, the "Core Observation Deck Level 3 Congestion Mitigation Plan" may include suggested measures such as activating backup evacuation routes, increasing on-site guidance personnel, and suspending access to certain areas.

[0190] After obtaining the basic contingency plan, the diversion parameters are dynamically optimized based on the current real-time tourist distribution and tourist transfer flow data. The diversion direction refers to the recommended alternative areas to which tourists should be guided from the warning area. This requires considering the real-time tourist distribution and selecting alternative areas with lower current carrying capacity that are confirmed to be accessible through transfer flow analysis. The timing of activating the diversion path is determined by comprehensively considering the predicted pressure arrival time along the pressure propagation path and the path's own preparation time (such as opening gates and deploying guidance signs). The recommended diversion rate is estimated based on the target path's capacity, the current tourist density along the path, and the expected diversion completion time.

[0191] Based on the optimized diversion parameters, the specific control measures, resource scheduling suggestions (such as the number of security guards and shuttle buses to be mobilized) and expected diversion targets (such as how long to reduce the tourist density in the warning area to a certain level) in the basic plan are refined and quantified, thereby generating a macro-level diversion plan document that can be deployed and implemented immediately for the current specific warning event.

[0192] This embodiment achieves a unified approach to standardization and personalization in emergency response by combining knowledge base matching with real-time data dynamic optimization. It leverages an effective contingency plan framework built upon historical experience, ensuring the plan's scientific rigor and reliability, while incorporating real-time situational information for parameter optimization, ensuring the plan's adaptability and operability to specific scenarios. This significantly improves the efficiency and effectiveness of macro-level guidance and decision-making.

[0193] In some embodiments, based on the macro-level traffic management plan and real-time individual visitor location information, a personalized tour guide strategy is generated and published using a route guidance algorithm, including:

[0194] Obtain real-time individual tourist location information and the target areas and recommended routes defined in the macro-level traffic control plan;

[0195] With the target area for guidance as a constraint and the goal of minimizing the comprehensive recreational pressure index of the areas that individual tourists are expected to pass through in the future within a preset time period, a path guidance optimization model is constructed.

[0196] Real-time individual tourist location information, individual travel preferences, and route guidance optimization model are input into the route guidance algorithm. The route guidance algorithm is based on a reinforcement learning framework and calculates personalized recommended routes that meet the guidance of the macro-level traffic management plan.

[0197] The system compares the personalized recommended route with the current real-time route congestion status to generate navigation tips that include route change suggestions, alternative attraction recommendations, and the expected improvement in the experience.

[0198] Through the tourist terminal application, real-time guidance information is pushed to individual tourists who are located in or about to enter the warning area to guide them to adjust their tour routes.

[0199] In this embodiment, the path guidance algorithm is built on a reinforcement learning framework. Its training process takes place in a simulated environment that mimics the path network of the target area, tourist movement rules, and dynamically changing recreational pressure distribution. The agent in the algorithm (representing a virtual tourist) learns a strategy by trying different path choices and based on the "rewards" received for each choice. The reward function is designed with the macro-level relief goals in mind; for example, a positive reward is given when the agent chooses to go to a lower-pressure area or follow a recommended path, while a negative reward is given when choosing to enter a high-pressure area. After extensive simulated training, the algorithm learns a strategy that can output an action suggestion (which area to go to next) based on the current state (tourist location, surrounding pressure distribution). This suggestion maximizes the cumulative reward in the long run, achieving an optimal balance between individual experience and overall relief goals.

[0200] When constructing the path-guided optimization model, the target areas are treated as a hard constraint, meaning that the paths planned for tourists must ultimately guide them away from or around these areas. The optimization objective is set to minimize the sum of the comprehensive recreational pressure indices of all areas that an individual tourist is expected to traverse within a predetermined timeframe. This model transforms the path planning problem into a sequential decision-making problem, considering the areas that may be entered after each step of movement and their predicted pressure.

[0201] During algorithm execution, it receives real-time individual tourist location information, individual travel preferences (such as a preference for quiet areas or popular attractions, which can be obtained through historical behavior or initial settings), and the aforementioned optimization model. Based on a reinforcement learning framework, the algorithm calculates the optimal next step or several future steps' movement sequence from the learned strategies according to the current tourist state, forming a personalized recommended path. This path, while meeting macro-level guidance constraints, strives to closely match the tourist's individual preferences.

[0202] After generating a personalized recommended route, it is compared with the real-time congestion status of the tourist's current route or the route they are about to enter. If the recommended route is significantly better than the current route in terms of expected pressure or congestion, navigation tips are generated. These tips include specific route change suggestions, alternative attraction recommendations, and a quantitative estimate of the degree of experience improvement.

[0203] Finally, through a dedicated application or mini-program on tourists' mobile phones, the system pushes the aforementioned guidance information in real time and accurately to individual tourists who are located within the warning area or who are judged to be about to enter the warning area based on their movement trajectory. After receiving the information, tourists can decide whether to adopt the suggestions, thereby responding to the macro-level guidance goals at the practical behavior level and achieving closed-loop regulation from overall management to individual behavior.

[0204] This embodiment uses a reinforcement learning path guidance algorithm to transform macro-level traffic management plans into executable individual guidance strategies. It not only considers the dynamic optimization of real-time pressure distribution but also takes into account individual tourist preferences, making the generated suggestions more acceptable. Through real-time comparison and precise push, it achieves effective and flexible intervention of management intentions on individual tourist behavior, ultimately improving the individual tour experience while synergistically achieving the macro-level goal of optimizing the overall regional passenger flow distribution.

[0205] By adopting the above technical solutions, this invention differs from existing technologies and possesses the following beneficial effects: By constructing a multi-dimensional recreational pressure assessment model, tourist spatiotemporal trajectory data is transformed into a real-time recreational pressure index that comprehensively reflects various aspects of the load, such as spatial congestion, facility load, ecological disturbance, and experience smoothness, thus achieving a precise quantitative assessment of scenic area pressure. Furthermore, by constructing a spatiotemporal graph network representing the relationship between inter-regional pedestrian flow and pressure propagation, and utilizing a pre-trained spatiotemporal graph prediction model, the predicted recreational pressure index and pressure propagation path of each hotspot area within a preset time period can be predicted proactively, thereby elevating management decisions from passive response to proactive early warning. Based on the prediction results, the solution automatically generates tiered early warning information and generates macro-level guidance plans by matching the pressure propagation path with the contingency plan knowledge base, ensuring the standardization and timeliness of emergency response. Finally, based on the macro-level guidance plan and real-time individual tourist location information, personalized tour guide strategies are generated and released through a path guidance algorithm, achieving closed-loop control from overall regional management to individual behavior guidance. The above-mentioned technical solutions effectively solve the problems of traditional scenic area management relying on lagging statistics and the difficulty in assessing, predicting and accurately alleviating dynamic and complex pressures. They enable proactive and refined control of recreational pressures, significantly improving the scientific nature of scenic area management, the comfort of tourists' experiences and the safety of the regional ecology.

[0206] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0207] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods of various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0208] The above description is only a part of the embodiments of the present invention and does not limit the scope of protection of the present invention. Any equivalent device or equivalent process transformation made based on the content of the present invention specification and drawings, or direct or indirect application in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourists' spatiotemporal trajectories, characterized in that, include: Acquire spatiotemporal trajectory data of tourists within the target area and geographic information data of the target area, wherein the spatiotemporal trajectory data includes a sequence of tourist locations with timestamps; Spatiotemporal clustering and behavioral pattern recognition are performed on the spatiotemporal trajectory data to generate a dynamic hotspot area set and its spatiotemporal characteristics. The spatiotemporal characteristics include the real-time tourist density, tourist stay duration and tourist transfer flow of each hotspot area within the target area. Based on the spatiotemporal characteristics and geographic information data, a multi-dimensional recreational pressure assessment model is constructed. The multi-dimensional recreational pressure assessment model calculates the real-time recreational pressure index of each hotspot area by integrating spatial congestion, facility load, ecological disturbance and experience smoothness indicators. Based on the real-time recreational stress index and its historical sequence, a spatiotemporal graph network representing the relationship between inter-regional population flow and stress propagation is constructed; The current state of the spatiotemporal graph network is input into the pre-trained spatiotemporal graph prediction model to obtain the predicted recreational pressure index and pressure propagation path of each hotspot area within a preset time period in the future. Based on the comparison between the predicted recreational stress index and the preset warning threshold, a graded warning information is generated, and a macro-level diversion plan is generated by matching the stress propagation path with the contingency plan knowledge base. Based on the aforementioned macro-level traffic management plan and real-time individual tourist location information, personalized tour guide strategies are generated and released through a path guidance algorithm to adjust the spatiotemporal distribution of tourists. The process includes performing spatiotemporal clustering and behavioral pattern recognition on the spatiotemporal trajectory data to generate a dynamic hotspot region set and its spatiotemporal characteristics, including: The spatiotemporal trajectory data is preprocessed to obtain a standardized set of tourist spatiotemporal location points; Based on the set of tourist spatiotemporal location points, a spatiotemporal density clustering algorithm is used for cluster analysis to identify tourist clusters with spatiotemporal continuity, and the spatiotemporal boundary of each tourist cluster is defined as a candidate hotspot region. Behavioral pattern analysis is performed on the trajectory point sequence within each candidate hotspot area. The behavioral pattern analysis includes calculating the average moving speed, dwell determination, and consistency of moving direction of the trajectory points. Based on the results of the behavioral pattern analysis, the candidate hotspot areas are semantically classified. Areas where dwelling behavior is dominant and the duration exceeds a preset duration threshold are classified as static hotspot areas, and areas where the movement speed drops sharply or where there is a clear convergence inflow characteristic are classified as dynamic hotspot areas. The static hotspot areas and dynamic hotspot areas together constitute the set of dynamic hotspot areas. Extract the spatiotemporal features of each region in the dynamic hotspot region set; The real-time tourist density is calculated by the ratio of the number of trajectory points in each region of the dynamic hotspot area set to the area of ​​the region. The tourist stay duration is obtained by statistically analyzing the time span of the trajectory points identified as staying in each region of the dynamic hotspot area set. The tourist transfer flow direction is determined by analyzing the region that the trajectory points leaving the current region enter at the next moment.

2. The method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories according to claim 1, characterized in that, Based on the aforementioned spatiotemporal characteristics and geographic information data, a multi-dimensional recreational stress assessment model is constructed, including: Based on the geographic information data, assessment dimensions and basic parameters for each hotspot area are defined. The assessment dimensions include spatial congestion, facility load, ecological disturbance, and user experience smoothness. The basic parameters include spatial safety capacity, service capacity of key service facilities, boundary of ecologically sensitive areas, and path free flow speed corresponding to each dimension. Based on the aforementioned spatiotemporal characteristics, quantitative calculation sub-models for each evaluation dimension are constructed. These sub-models include a spatial congestion sub-model, a facility load sub-model, an ecological disturbance sub-model, and an experience smoothness sub-model. The spatial congestion sub-model is constructed based on the ratio of real-time tourist density to the spatial safety capacity; the facility load sub-model is constructed based on the identification results of queuing behavior around the facility and the supply and demand relationship of the service capacity; the ecological disturbance sub-model is constructed based on the frequency of trajectory intrusion into the boundary of the ecologically sensitive area; and the experience smoothness sub-model is constructed based on the ratio of the average tourist movement speed to the free flow speed. Define the output normalization function of the quantitative calculation sub-model for each evaluation dimension, so that the output of the quantitative calculation sub-model for each evaluation dimension is normalized dimension index. Define a weighted aggregation function that integrates various normalized dimension indicators to generate a comprehensive stress value; The multi-dimensional recreational stress assessment model is obtained by integrating the assessment dimensions, basic parameters, quantitative calculation sub-models for each assessment dimension, output normalization function, and weighted aggregation function.

3. The method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories according to claim 2, characterized in that, The real-time recreational stress index for each hotspot area was calculated, including: The real-time tourist density in the spatiotemporal features is input into the spatial congestion sub-model to calculate the spatial congestion index value. The tourist stay distribution in the spatiotemporal features and the location of the key service facilities are input into the facility load sub-model to calculate the facility load index value. The spatiotemporal trajectory data and the boundary of the ecologically sensitive area are input into the ecological disturbance degree sub-model to calculate the ecological disturbance degree index value; The average tourist movement speed in the spatiotemporal features and the free-flow speed along the path are input into the experience smoothness sub-model to calculate the experience smoothness index value. The space congestion index, facility load index, ecological disturbance index, and experience smoothness index are processed according to the output normalization function to obtain the normalized dimension index values. The normalized dimensional index values ​​are input into the weighted aggregation function for calculation, and the real-time recreational pressure index of each hotspot area is output.

4. The method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories according to claim 1, characterized in that, Based on the real-time recreational stress index and its historical sequence, a spatiotemporal graph network representing the relationship between inter-regional population flow and stress propagation is constructed, including: Each hotspot region in the dynamic hotspot region set is abstracted as a node of the spatiotemporal graph network; Based on the tourist transfer flow direction in the aforementioned spatiotemporal features, the tourist transfer intensity between any two hotspot areas within a preset time window is calculated, and the tourist transfer intensity is used as the weight of the edge connecting the corresponding two nodes. Based on the spatiotemporal trajectory data, the time spent by tourists transferring between hotspot areas is analyzed, and the transfer time is used as the time delay attribute of the corresponding edge. The real-time recreational pressure index and its historical sequence of each hotspot area are used as the state feature vector of the corresponding node. External influencing factor nodes are incorporated into the spatiotemporal graph network. These external influencing factor nodes include transportation hub status nodes and weather event nodes. Connection edges are established between the external influencing factor nodes and relevant hotspot area nodes based on causal relationships.

5. The method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories according to claim 1, characterized in that, The current state of the spatiotemporal graph network is input into a pre-trained spatiotemporal graph prediction model to obtain the predicted recreational stress index and stress propagation path for each hotspot area within a preset future time period, including: The state feature vectors of all nodes in the spatiotemporal graph network, the weight attributes and time delay attributes of the edges are used together as the input of the spatiotemporal graph prediction model. The spatiotemporal graph prediction model uses graph convolutional layers to model the spatial dependencies of the spatiotemporal graph network's topology and node features, thereby capturing the pressure correlations between hotspot regions. The temporal modeling layer in the spatiotemporal graph prediction model is used to model the historical sequence of node feature vectors in terms of time dependence, so as to learn the trend and periodic pattern of recreational pressure evolution over time. By utilizing the attention mechanism in the spatiotemporal graph prediction model, the importance weights of different nodes and different historical moments to the current prediction task are dynamically calculated. Based on the spatial dependence modeling, temporal dependence modeling, and importance weights, the predicted recreational pressure index for each hotspot area within a preset future time period is generated. Meanwhile, based on the results of the spatial dependency modeling, the spatiotemporal graph prediction model outputs the pressure propagation path, which represents the probability and direction of pressure diffusion among nodes in the spatiotemporal graph network.

6. The method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories according to claim 5, characterized in that, Utilizing the attention mechanism in the spatiotemporal graph prediction model, the importance weights of different nodes and different historical moments to the current prediction task are dynamically calculated, including: Based on the topology and node characteristics of the spatiotemporal graph network, the spatial association strength weight between any two nodes is calculated through the spatial attention module. The spatial association strength weight is used to reflect the contribution of the state change of one node to the future pressure of another node. Based on the historical sequence of node feature vectors, the temporal correlation strength weight of different historical moments to the prediction of future target moments is calculated through the time attention module. The temporal correlation strength weight reflects the contribution of the state of the historical moment to the prediction of future trends. The spatial correlation strength weight and the temporal correlation strength weight are fused to generate the importance weight used to guide the feature aggregation of the graph convolutional layer and the temporal modeling layer.

7. The method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories according to claim 1, characterized in that, Based on a comparison between the predicted recreational stress index and a preset warning threshold, tiered warning information is generated, including: The predicted recreational stress index is compared with a preset multi-level stress threshold, which corresponds to different stress levels. Based on the comparison results, the pressure level that each hotspot area will reach in the future within a preset time period is determined; If the pressure level exceeds the preset warning trigger level, a warning event is generated that includes the warning area, warning level, estimated arrival time, and the pressure propagation path. Based on the warning level of the warning event, a corresponding warning release strategy is matched, and the warning release strategy includes the release channel of the warning information, the release content template and the release priority; Tiered early warning information is generated based on the aforementioned early warning release strategy.

8. The method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories according to claim 7, characterized in that, Based on the pressure propagation path, a macro-level evacuation plan is generated by matching it with the contingency plan knowledge base, including: The warning area, warning level and pressure propagation path in the warning event are matched with the contingency plan triggering conditions in the contingency plan knowledge base, which contains standard evacuation plans for different pressure scenarios. Retrieve at least one standard evacuation plan that best matches the current early warning event from the contingency plan knowledge base as the basic contingency plan; Based on the real-time tourist distribution and tourist transfer direction in the spatiotemporal characteristics, the diversion parameters in the basic plan are dynamically optimized. The diversion parameters include the diversion direction, the activation timing of the diversion path, and the recommended diversion rate. Based on the optimized diversion parameters, an executable macro-level diversion plan is generated for the current early warning event. The macro-level diversion plan includes specific control measures, resource allocation suggestions, and expected diversion targets.

9. The method for assessing and issuing early warning of recreational pressure in hotspot areas based on tourist spatiotemporal trajectories according to claim 8, characterized in that, Based on the aforementioned macro-level traffic management plan and real-time individual visitor location information, personalized tour guide strategies are generated and published using a path guidance algorithm, including: Obtain the real-time individual tourist location information and the target areas and recommended routes defined in the macro-level traffic control plan; Using the target area for guidance as a constraint, and aiming to minimize the comprehensive recreational pressure index of the areas that individual tourists are expected to pass through in a future preset time period, a path guidance optimization model is constructed. The real-time individual tourist location information, individual travel preferences, and the path guidance optimization model are input into the path guidance algorithm. The path guidance algorithm is based on a reinforcement learning framework and calculates a personalized recommended path that meets the guidance of the macro-level traffic management plan. The personalized recommended route is compared with the current real-time route congestion status to generate navigation prompts that include route change suggestions, alternative attraction recommendations, and the expected degree of experience improvement. Through the tourist terminal application, the guide prompts are pushed in real time to individual tourists who are located or about to enter the warning area, so as to guide them to adjust their tour routes.

Images