Intelligent tourism recommendation method and system based on AI

Abstract

The invention belongs to the technical field of smart tourism management, and relates to an AI-based smart tourism recommendation method and system, and the method comprises the steps: obtaining instantaneous aggregation density data collected by a plurality of hidden monitoring nodes in a scenic spot in real time; generating a dynamic environment grade set according to the instantaneous aggregation density data, the real-time weather elements and the traffic control information; generating a user physical exhaustion level, and generating an interest preference label through an electronic map zooming operation track; identifying space-time synchronization features between the associated mobile terminals to generate a group type identifier, and binding the group type identifier with a strategy recommendation template; performing peak shifting recommendation sorting in combination with the dynamic environment level to form an initial recommendation scheme set; and comparing the matching degree between the initial recommendation scheme and the interest preference label, triggering the secondary optimization of the recommendation content based on the group type identifier when the matching degree is lower than a set range, and injecting an emergency risk avoiding path to complete a closed-loop recommendation decision chain. The problem that in the prior art, due to the fact that the data collection dimension is single, recommendation strategies lack group adaptability is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent tourism management and relates to an AI-based intelligent tourism recommendation method and system. Background Art

[0002] Existing tourism recommendation systems generally have limitations in coping with dynamic environmental changes. Most systems rely on static historical data and basic location services, making it difficult to integrate multi-dimensional dynamic parameters such as dense crowds, sudden weather, and traffic control in real time. The recommended routes are often out of touch with the actual carrying capacity of scenic spots. Local congestion is likely to occur during peak hours in scenic spots. Traditional solutions only give warnings based on simple thresholds and lack accurate density calibration, unable to effectively avoid potential group safety hazards.

[0003] Traditional solutions usually adopt fixed preference tags and general path planning algorithms to generate recommendations by manually inputting points of interest by users or simple trajectory analysis. Although such methods can provide basic route suggestions, they ignore individual physical differences and real-time behavior characteristics, and do not analyze the deep demand signals in the interaction between mobile terminal sensor data and electronic maps. The recommended results are likely to deviate from the actual bearing capacity and potential interest orientation of users.

[0004] Due to the above problems, existing technologies have a single data collection dimension and are difficult to identify group cooperation modes such as family travel and group travel, resulting in a lack of group adaptability in recommendation strategies. Summary of the Invention

[0005] To solve the above problems, the present invention provides an AI-based intelligent tourism recommendation method and system.

[0006] In a first aspect, the present invention provides an AI-based intelligent tourism recommendation method, adopting the following technical solutions: An AI-based intelligent tourism recommendation method includes the following steps: S1. Obtain the instantaneous aggregation density data collected in real time by multiple concealed monitoring nodes in the scenic area; S2. Generate a set of dynamic environment levels according to the instantaneous aggregation density data, real-time weather elements, and traffic control information; S3. Establish a mobile terminal behavior data interaction channel, analyze the gyroscope acceleration sequence to generate the user's physical exertion level, and generate interest preference tags through the electronic map zoom operation trajectory; S4. Identify the spatio-temporal synchronization characteristics between associated mobile terminals to generate group type identifiers and bind policy recommendation templates; S5. Adjust the dynamic weight distribution of multi-dimensional recommendation parameters, match the user's physical exertion level with the scenic slope data to generate a route intensity coefficient, and combine the dynamic environment level to perform off-peak recommendation sorting to form an initial recommendation plan set; S6. Compare the matching degree between the initial recommended plan and the interest preference tags. When it is lower than the set range, trigger the secondary optimization of the recommended content based on the group type identifier, and inject the emergency avoidance path to complete the closed-loop recommendation decision chain.

[0007] A further solution of the present invention is to obtain the instantaneous aggregation density data, including the following steps: Activate the laser ranging unit of the concealed monitoring node to perform three-dimensional trajectory scanning, and generate a matrix of foot height values; Synchronously start the thermal imaging unit to perform dynamic density calibration on the laser ranging blind area, and generate a dynamic correction coefficient; Construct a data collaborative verification network between monitoring nodes to generate redundant monitoring data streams for sudden surges in the flow of people; Fuse the matrix of foot height values, the dynamic correction coefficient, and the redundant monitoring data streams to generate an instantaneous aggregation density data set including real-time credibility weights.

[0008] A further solution of the present invention is to generate a set of dynamic environment levels, including the following steps: Overlay the instantaneous aggregation density data with the temperature field interference compensation value to generate corrected density data; Perform multi-factor similarity matching between the corrected density data and the historical evolution pattern of sudden events to generate an accommodation index and a path clearance coefficient; Establish an environmental level update trigger mechanism. When the monitoring node detects an abnormal movement trajectory, shorten the sliding window time and recalculate the environmental level; Output a set of dynamic environment levels including risk classification identifiers.

[0009] A further solution of the present invention is to generate a user physical exertion level, including the following steps: Perform frequency band separation on the gyroscope three-axis acceleration data, and extract the vertical vibration component to construct a physical exertion evaluation matrix; Generate a high-frequency band energy integral value based on the number of peak mutations and the standard deviation coefficient; Generate a quantified physical exertion level according to the exceeding ratio of the high-frequency band energy integral value to the preset baseline.

[0010] A further solution of the present invention is to generate interest preference tags, including the following steps: Capture the scaling operation trajectory of the user on the electronic map to generate a visual focus weight value; Statistically analyze the hesitation characteristics of the user's stay when the path selection is not completed; Perform cosine similarity matching between the visual focus weight value and the scenic spot attribute vector to generate preference tags.

[0011] A further solution of the present invention is to generate a group type identifier, including the following steps: Construct a dynamic vector model of the movement trajectory, and calculate the direction persistence coefficient and the distribution entropy value of the stop points; Compare the Euclidean distance of the movement vectors and the direction coordination index between associated terminals; Determine the parent-child interaction mode or the team collaboration mode according to the continuous spatio-temporal synchronization characteristics, and load the corresponding strategy recommendation template.

[0012] A further solution of the present invention is to generate a route strength coefficient, including the following steps: Establish a non-linear mapping relationship between the user's physical exertion level and the slope threshold; Adopt a step function to dynamically adjust the route strength coefficient, and superimpose the attenuation factor obtained by spectrum analysis on the path with an excessive slope; Generate a standardized route feasibility parameter in combination with the real-time route smoothness coefficient.

[0013] A further solution of the present invention is to perform off-peak recommendation ranking in combination with the dynamic environment level, including the following steps: Screen the candidate list above the safety threshold according to the tolerance index; Establish a dynamic weight mapping for the candidate scenic spots by superimposing the time decay factor; Generate a priority recommendation queue through the secondary matching of the route strength coefficient and the preference label.

[0014] A further solution of the present invention is to inject an emergency evacuation path, including the following steps: Perform geofence comparison through a Gaussian mixture model to mark the area where the instantaneous aggregation density exceeds the limit; Generate a three-dimensional shortest obstacle avoidance path integrating the physical exertion level; Embed the emergency evacuation path into the recommendation scheme set and monitor the user response to trigger feedback optimization.

[0015] In a second aspect, the present invention provides an AI-based intelligent tourism recommendation system, adopting the following technical solution: A concealed monitoring network module, configured to deploy monitoring nodes with eco-friendly camouflage, and integrate a laser ranging unit and a thermal imaging unit to generate instantaneous aggregation density data; An environment perception calculation module, configured to fuse weather elements, traffic control information, and density data to generate a set of dynamic environment levels; A terminal behavior analysis module, configured to parse the gyroscope features through a low-power Bluetooth link to generate the physical exertion level, and synchronously process the map zooming trajectory to generate interest preference labels; A group identification module, configured to calculate the spatio-temporal synchronization characteristics of mobile terminals, generate group type identifiers and associate strategy recommendation templates; A dynamic recommendation engine module, configured to match the user's physical exertion level with the scenic spot slope data to generate a route intensity coefficient, and combine the dynamic environment level to perform off-peak recommendation sorting to form an initial recommendation plan set; An emergency path planning module, configured to load three-dimensional terrain data to generate the shortest evacuation path including slope and control constraints.

[0016] In summary, the present invention includes the following beneficial technical effects: 1. Real-time collection of instantaneous aggregation density data in the scenic area and combination with weather and traffic control information to generate a dynamic environment level set, enabling the recommendation system to accurately perceive environmental changes. This dynamic adjustment mechanism can effectively identify congested areas and potential risk points, timely trigger off-peak recommendation strategies, significantly reduce the risk of tourists being stranded in case of emergencies, and improve the overall safety management ability of the scenic area; 2. Using the mobile terminal gyroscope to generate the user's physical exertion level, and extracting interest preference tags by analyzing the electronic map zooming trajectory to achieve personalized adaptation of the travel route. This method comprehensively considers the user's physical endurance range and interest characteristics, avoids recommending overloaded or irrelevant scenic spots, optimizes the user experience while reducing resource waste caused by ineffective route planning; 3. When the initial recommendation matching degree is insufficient, trigger a secondary optimization mechanism based on the group type identifier, integrate the instantaneous aggregation density and physical exertion level to generate a three-dimensional evacuation path, and form a closed-loop decision chain by dynamically superimposing emergency path parameters to ensure the path safety under extreme weather or emergencies, significantly improving the disaster tolerance response efficiency of the tourism system. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. The drawings are used to provide a further understanding of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 Disclosed is a schematic flowchart of an AI-based intelligent tourism recommendation method.

[0019] Figure 2 Disclosed is a curve graph of performing off-peak recommendation sorting in the application embodiment.

[0020] Figure 3 Disclosed is a schematic structural diagram of an AI-based intelligent tourism recommendation system. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0022] The following is combined with Figures 1 - 3 The preferred embodiments of the present invention are described in detail.

[0023] See attached Figure 1 , the present invention proposes an AI-based intelligent tourism recommendation method, comprising the following steps: S1. Obtain instantaneous aggregation density data collected in real time by multiple hidden monitoring nodes in the scenic area; S2. generating a dynamic environment level set according to the instantaneous gathering density data, real-time weather elements, and traffic control information; S3, establish a mobile terminal behavior data interaction channel, analyze the gyroscope acceleration sequence to generate the user's physical exertion level, and generate interest preference tags through the electronic map zoom operation trajectory; S4, identifying the spatiotemporal synchronization features between associated mobile terminals to generate a group type identifier and bind it to a policy recommendation template; S5. Adjust the dynamic weight distribution of multi-dimensional recommendation parameters, match the user's physical exertion level with the scenic spot slope data to generate a route intensity coefficient, and perform peak-shift recommendation sorting in combination with the dynamic environment level to form an initial recommendation solution set; S6. Compare the matching degree of the initial recommendation plan with the interest preference label. When it is lower than the set range, trigger the secondary optimization of the recommended content based on the group type identification, and inject the emergency risk avoidance path to complete the closed-loop recommendation decision chain.

[0024] In one embodiment of the present invention, step S1 includes the following steps: Deploy covert monitoring nodes with multimodal data collection capabilities; Specifically, the top of the street light poles and the side walls of the trash cans in the scenic area are pre-buried with environmental sensing modules. The shell of the environmental sensing module adopts an ecological cracked heat dissipation grille design and integrates a micro servo motor drive system. The installation density of each monitoring node is determined according to the heat map of the flow of tourists during the historical peak hours, and the installation spacing on both sides of the passages that tourists must pass through does not exceed rice; Among them, the concealed monitoring node is a data acquisition device that realizes visual camouflage fusion between the outer surface and the mounting carrier equipment. The ecological cracked heat dissipation grille is a heat dissipation structure made of basalt texture patches and internal honeycomb metal grids. The micro servo motor drive system is a miniaturized multi-axis steering device using a closed-loop control mode.

[0025] Activate the laser ranging unit of the monitoring node to complete three-dimensional trajectory scanning; Specifically, the laser transmitter is controlled by a servo motor to perform vertical rotation scanning at a rate of 180 degrees per minute, and the scanning angle is adjusted to a downward tilt range of 45 to 70 degrees. During each complete rotation, a pulsed laser beam with a wavelength of 905 nanometers is continuously emitted; after receiving the reflected signal, the time difference of each laser point is calculated to generate a foot height value matrix, and when the motion vector change of the same coordinate point exceeds a preset threshold within two consecutive rotation cycles, it is marked as valid foot displacement data; The motion vector change is the composite parameter of the displacement direction and velocity of the detection point coordinates in a continuous time series. The vertical plane rotation scanning of the laser ranging unit is a technical means to change the laser emission direction through motor drive to cover the three-dimensional space of the monitoring area. The foot height value matrix is ​​a three-dimensional space data set corresponding to the two-dimensional plane coordinates and height values ​​established based on the laser ranging principle. The generation of the foot height value matrix needs to exclude the reflection interference of fixed obstacles.

[0026] Synchronously perform dynamic density calibration of the thermal imaging unit; Specifically, when the laser ranging unit rotates to the 90-degree reference position, the thermal imaging sensor is activated to take panoramic shots, collect electromagnetic radiation energy in the infrared wavelength range of 8-14 microns and convert it into a temperature distribution matrix; the boundary area where the adjacent temperature difference exceeds 3°C in the temperature matrix is ​​defined as the human activity area, and the ratio of the area that meets the typical thermal radiation characteristics of the human body to the coverage range of the laser ranging data is calculated as a dynamic correction coefficient; Among them, the dynamic correction coefficient is an important compensation parameter to eliminate the influence of the laser ranging blind area. The typical thermal radiation characteristics of the human body are the imaging area with a surface temperature of 34-38 degrees Celsius and a clear body contour. Dynamic density calibration is a technical solution to compensate for the systematic deviation of the laser ranging blind area in personnel counting through thermal imaging data. The calculation of the dynamic correction coefficient must meet the thermal imaging recognition criteria of ergonomic characteristics.

[0027] Build a data collaborative verification network between monitoring nodes; Specifically, any node detects that the displacement data increment per unit time exceeds the preset safety value. When it is time, send a collaborative wake-up instruction to adjacent nodes within a radius of 50 meters; the woken nodes adjust the laser scanning angle to be tilted downward by 30 degrees to 60 degrees, and the sampling frequency of the thermal imaging sensor is increased to 2 frames per second, forming a redundant monitoring data stream with superimposed coverage; Among them, the collaborative wake-up instruction refers to an encrypted data packet containing the coordinates of the burst point and the ID of the service node. The redundant monitoring data stream refers to the spatio-temporal alignment detection result formed by the heterogeneous sensor data of multiple nodes. The burst response mechanism of the collaborative verification network is a method for quickly capturing instantaneous population surge events through the time synchronization comparison of heterogeneous sensor data. The spatio-temporal alignment of the redundant monitoring data stream needs to achieve a timestamp matching accuracy of 10 milliseconds.

[0028] Output the fusion density parameter of the environmental perception network; Specifically, after the foot displacement data collected by each node is superimposed with the dynamic correction coefficient, calculate the product of the number of effective displacement points per unit area and the displacement rate to generate a spatial distribution density value; perform a weighted summation process on the spatial distribution density value and the dynamic correction coefficient to generate an instantaneous aggregation density data set containing real-time credibility weights and complete encrypted storage at the network edge node; Among them, the real-time credibility weight represents a data reliability index determined based on the collaborative verification result. Encrypted storage refers to a local storage method that encrypts end-side data using the existing national cryptography SM4 algorithm.

[0029] Calculate the product of the number of effective displacement points per unit area and the displacement rate to generate a spatial distribution density value, which satisfies the following formula:

[0030] Among them, is the spatial distribution density value, and "person" is regarded as a dimensionless counting unit; is the number of effective displacement points, obtained through the determination criterion that the motion vector change exceeds the preset threshold; is the rate value of a single displacement point, calculated based on the laser point time difference method, ; represents the displacement distance change amount, that is, the spatial displacement difference between adjacent laser ranging points; represents the time difference, that is, the sampling interval time of laser ranging; is the position weight factor, and when the displacement point is within 3 meters of the center line of the channel, the value is , and in other areas, take ; is the area of the sub-region where the monitoring node is located.

[0031] The fusion calculation of the instantaneous aggregation density satisfies the following formula:

[0032] Among them, is the instantaneous aggregation density data of the output, with the unit of standard density unit; is the dynamic correction coefficient, which is calculated through the area ratio of the thermal imaging region; represents the spatial distribution density value, and "person" is regarded as a dimensionless counting unit; represents the credibility weight of the laser data, which is dynamically assigned according to the matching degree of the collaborative verification network; represents the thermal imaging calibration weight, which linearly increases according to the temperature boundary clarity.

[0033] represents normalized linear superposition, which satisfies the following formula:

[0034] Among them, and are both dimensionless processed (through , standardized weight conversion), ensuring that the final output is in a unified standard unit.

[0035] Exemplarily, for the acquisition of the instantaneous aggregation density data in step S1, monitoring nodes with imitation bark textures are deployed at the tops of the lamp posts on both sides of the road section. When the monitoring nodes are activated, the micro servo motor drives the pulsed laser to perform vertical plane scanning at a 45-degree tilt angle, completing 30 rotations per second and generating a foot height matrix with an accuracy of 0.2 m. At the same time, the thermal imaging module automatically calibrates when the temperature change exceeds , and when the temperature of the stone step area rises to at noon due to sunlight, dynamic compensation will be performed on the error area of the laser ranging blind area.

[0036] When three adjacent nodes form a collaborative verification network, if a certain node detects a sudden increase in the personnel displacement vector within 10 seconds , it will immediately wake up the adjacent 5 nodes to increase the thermal imaging sampling rate to , forming an overlapping redundant data stream, and finally taking the position weight factor the weighting coefficient of the central area of the channel when calculating the density value per unit area through the formula.

[0037] In one embodiment of the present invention, step S2 includes the following steps: Obtain multi-source data and align the timestamps; Specifically, the Internet of Things platform in the scenic area retrieves real-time weather elements, which are three types of parameters directly collected by meteorological sensors: precipitation, wind speed, and temperature rapid change rate. These three types of parameters are converted into five-level risk values. Traffic control information is read from the interface of the traffic management center, and the traffic control information is the road closure period and the passing capacity coefficient of the alternative route. Instantaneous aggregation density data output in step S1 is synchronously received, and timestamp synchronization processing is performed on the above data with a 30-second period to eliminate clock errors between different devices. Among them, the passing capacity coefficient of the alternative route in the traffic control information is the ratio of the maximum number of people who can pass to the current number of people.

[0038] Construct a dynamic environment assessment sliding window. Specifically, the sliding window algorithm is used to analyze the continuous time series data before the current moment. The sliding window algorithm is a dynamic matching method that divides data in the time dimension. The window length is 15 minutes, and it slides forward every 30 seconds. The evolution patterns of historical emergencies within the window include the crowd aggregation rate, the conduction path during equipment failures, and the evolution characteristics of evacuation behaviors caused by sudden weather changes. Among them, the crowd aggregation rate represents the statistical proportion that the density increase of the monitoring nodes in the same area within a unit time exceeds . The equipment failure conduction path represents the percentage decrease in the coverage rate of the emergency monitoring scope that adjacent nodes need to bear after the current monitoring node fails.

[0039] Calculate the dynamic correction coefficient and generate the environment level. Specifically, the instantaneous aggregation density data is superimposed with the temperature field interference compensation value. The temperature field interference compensation value is the temperature drift error correction amount caused by the direct sunlight on the thermal imaging unit, and this value is calibrated in real time by the temperature sensor. The corrected density data is matched with the historical emergency evolution pattern. The similarity matching method is a weighted evaluation model of Euclidean distance and time fluctuation characteristics. The weight distribution scheme: the crowd aggregation characteristic accounts for , the equipment anomaly characteristic accounts for , and the sudden weather change characteristic accounts for ; The tolerance index and the path smoothness coefficient are generated according to the matching result. Among them, the tolerance index is the percentage of the remaining accommodation capacity obtained by subtracting the current corrected density from the maximum carrying capacity of the scenic area, and then superimposing the combined parameter of the weather risk level and the traffic control coefficient. The combination method is the percentage of the remaining accommodation capacity multiplied by the weather risk weight. The path smoothness coefficient is a decay function of the ratio of the passing time of the optimal path to the average congestion duration.

[0040] The similarity matching method is a weighted evaluation model of Euclidean distance and time fluctuation characteristics, which satisfies the following formula:

[0041] Among them, is the matching score between the environmental state and historical events, a dimensionless scalar; is the normalized value of the Euclidean distance between the current crowd aggregation rate and historical events. The Euclidean distance is mapped to the range by the maximum-minimum method; is the correlation degree of device abnormal features, normalized according to the percentage loss of the failure coverage rate of monitoring nodes in historical events to ; is the deviation degree of weather mutation behavior, which is converted into a coefficient according to the difference in the fluctuation amplitude of rainfall and wind speed data compared with similar events.

[0042] , , are all processed by normalization into dimensionless numerical values. The sum of the weight coefficients (0.6, 0.25, 0.15) is 1, which is the result obtained through the experience of experimental personnel to ensure that the superimposed effectively reflects the synergistic effect of multiple factors and avoids directly operating with mixed dimensional parameters.

[0043] The calculation of the path smoothness coefficient satisfies the following formula:

[0044] Among them, represents the path smoothness coefficient, dimensionless; represents the shortest passing time of the path in the non-congested state, in minutes; represents the current average congestion duration, in minutes, and the data of the previous 1 hour is statistically analyzed by the moving average method; is the time decay factor, obtained by regression analysis of historical data; is the time interval from the nearest control measure, in minutes. The units of both the numerator and the denominator are minutes, so the ratio is dimensionless, and the time term in the independent variable of the exponential function can eliminate the unit to ensure that all operations conform to the principle of dimensional consistency.

[0045] The calculation of the tolerance index, the remaining tolerance percentage and the weather risk level are combined using a limited amplitude linear model, satisfying the following formula:

[0046] Among them, represents the tolerance index, and the limited output result does not exceed 100% to avoid overload error. Among them, the weather risk level is an integer

[0047] Perform dynamic loop monitoring on the set of execution environment levels; Specifically, establish an environmental level update trigger mechanism. When the thermal imaging temperature change amplitude of any monitoring node exceeds or an abnormal movement trajectory is detected in the laser ranging data, forcibly shorten the sliding window time to 5 minutes for recalculation; record the transfer sequence of the historical state after each level update, which is used to optimize the weight allocation strategy for subsequent similarity matching; Among them, the abnormal movement trajectory is the monitoring data where the tourist's foot trajectory makes more than three round trips within 10 seconds, and this feature corresponds to sudden aggregation or panic escape behavior. The historical state transfer sequence is a multi-parameter change trajectory database when the environmental level transitions from low risk to high risk.

[0048] In one embodiment of the present invention, step S3 includes the following steps: Construct a mobile terminal behavior data interaction channel; Specifically, on the premise of tourist authorization, establish a low-power Bluetooth transmission link through the sensor open interface provided by the mobile operating system. The transmission link periodically obtains the original gyroscope sampling data in the background and intercepts the touch event stream of the electronic map application layer at the same time; Among them, the mobile terminal behavior data interaction channel represents a data acquisition architecture based on permission hierarchical management. Register an acceleration listener through the interface of the Android system, set the sampling frequency to 20 times per second, and capture the three-axis data at each sampling. The low-power Bluetooth transmission link adopts the GATT characteristic value notification mechanism of the BLE protocol stack to ensure continuous data reception in the state where the terminal screen is off.

[0049] Analyze the gyroscope three-axis acceleration sequence to generate a physical exertion level; Specifically, perform differential preprocessing on the continuously collected axial acceleration data, extract the number of peak mutations and the standard deviation coefficient in the time-domain waveform through the sliding window algorithm, separate the horizontal movement component and the vertical vibration component in frequency bands, and construct a physical exertion evaluation matrix; Among them, differential preprocessing means calculating the change rate of adjacent sampling points using the three-point central difference method. The sliding window algorithm sets the window length to 5 seconds, and each window contains at least 100 groups of three-axis data points. The frequency band separation is achieved through a Butterworth filter. The horizontal movement component corresponds to the low-frequency band of and the vertical vibration component corresponds to

[0050] The calculation of the high-frequency band energy integral value satisfies the following formula:

[0051] where, is the high-frequency band energy integral value; is the frequency-domain amplitude of the vertical vibration component, obtained through fast Fourier transform; is the frequency resolution, related to the sampling rate; The integral result of is The total vibration energy of the frequency band, used to quantify the user's exercise intensity.

[0052] Process the operation trajectory of the electronic map zooming to generate interest preference labels; Specifically, capture the user's touch operation on the electronic map, record the duration of each two-finger zoom action, the coordinates of the zoom center point, and the zoom ratio difference. Generate a visual focus weight value based on the operation of focusing on the same geographical area three or more times in a row, and synchronously count the number of times the operation interval exceeds the set duration but the path selection is not completed as the stay hesitation degree feature; where, the two-finger zoom action is a touch screen gesture including the starting contact distance and the ending contact distance The zoom ratio difference is obtained by calculating to obtain the absolute value; the visual focus weight value is the product of the number of times the target area is continuously magnified and the full-screen display duration of the map; the stay hesitation degree feature is the difference between the number of times the user clicks on the same map marker point within the preset time threshold and the number of times the navigation instruction is not finally executed.

[0053] The zoom ratio difference is obtained by calculating to obtain the absolute value, satisfying the following formula:

[0054] where, represents a dimensionless zoom level parameter; represents the distance between the starting contacts of the two fingers; represents the distance between the ending contacts of the two fingers. Both are length values of the same dimension, The ratio of can cancel the dimension, and the logarithmic operation result meets the requirements of mathematical definitions.

[0055] Establish the mapping relationship between behavior characteristics and decision-making parameters; Specifically, the physical exertion level is converted into the slope adaptation coefficient of the route planning system, the visual focus weight value is matched with the label attributes in the scenic spot portrait database by cosine similarity, and the stay hesitation degree feature is used to dynamically adjust the exploration tendency weight of the recommendation strategy; Among them, the slope adaptation coefficient represents an inverse proportional function of the physical exertion level and the preset slope threshold range. Cosine similarity matching means performing spatial projection calculation on the user's visual focus weight vector and the scenic spot attribute vector. The exploration tendency weight is implemented by a piecewise function where the greater the stay hesitation eigenvalue, the higher the recommendation diversity.

[0056] The calculation of the visual focus weight value satisfies the following formula:

[0057] Where is the magnification times in three consecutive zoom operations of the target area, a dimensionless count; is the cumulative stay time of this area in the full-screen display state, in seconds; is the total duration of the current user using the map, in seconds.

[0058] As a probabilistic weight, it needs to be normalized and converted into a dimensionless value as the output of representing the visual focus weight value.

[0059] The exploration tendency weight piecewise function satisfies the following formula:

[0060] Where is the probability given to the diversity strategy by the exploration tendency weight when the recommendation system selects candidate scenic spots. The greater the weight value, the higher the proportion of unknown interest points recommended; is the stay hesitation eigenvalue, which is the difference between the number of clicks of the user on the same location and the number of times the navigation is not finally executed, reflecting the contradiction intensity in the user's behavior.

[0061] Basis for setting the exploration tendency weight piecewise function: The low hesitation interval is ; The initial slope of the linear growth mechanism: When the user has occasional hesitation, by default, maintain the existing preference recommendation of The basic weight corresponds to exploration, but as increases, gradually introduce niche scenic spots. For each unit increase in when force a fundamental change in the recommendation strategy.

[0062] The medium hesitation platform interval is ; Constant weight That is Tendency to explore: When the number of times the user hesitates significantly exceeds the number of times the navigation is executed It is determined that the user is dissatisfied with the current recommendation mode, and it is necessary to forcibly enable a high exploration strategy to avoid user loss caused by repeated recommendation of similar scenic spots. The platform interval setting prevents the weight from fluctuating violently near the critical point.

[0063] The high hesitation incentive interval, that is ; Quadratic linear growth, that is, the slope drops to For users with extremely contradictory behaviors (such as frequent clicks but never confirming the path), On the basis of the exploration weight, the diversity is additionally enhanced, but the growth rate is slowed down to avoid the recommended content deviating too much from the user's core interests. The slope adjustment here prevents the system from overreacting.

[0064] Encapsulate the behavior data output interface; Specifically, the physical exertion level, visual focus weight value, and stay hesitation degree characteristics are structured and encapsulated in the format, and after adding the timestamp verification and the data integrity hash value, it is pushed to the recommendation decision engine through long connection; Among them, the data structured encapsulation is to quantify the floating-point physical value into level integer identifier, the visual focus weight value is reserved to two decimal places and normalized, the timestamp verification uses the server synchronization mechanism to ensure the data time series consistency of multiple terminals, and the data integrity hash value uses the existing algorithm to generate a message digest for verification by the receiving end.

[0065] In one embodiment of the present invention, step S4 includes the following steps: Obtain the trajectory dynamic data sequence of the associated mobile terminal; Specifically, the positioning module and the motion sensor preset in the tourist mobile terminal collect the physical position coordinates and the moving direction angle of the target terminal at each fixed time interval. The position coordinates are the fusion of module and the mixed positioning result of the signal fingerprint, and the moving direction angle is the triaxial motion synthesis vector direction angle calculated by the built-in gyroscope and accelerometer of the terminal; Among them, the trajectory dynamic data sequence is an ordered set containing the timestamp, longitude and latitude deviation value, and vector angle offset. The longitude and latitude deviation value of the position coordinates is the calibration data for compensating the drift using base station triangulation.

[0066] Construct a mobile trajectory dynamic vector model; Specifically, for the trajectory data continuously collected in each time period, taking the starting position as the reference point, the terminal movement trajectory is transformed into a three-dimensional vector with time and space attributes. The vector elements include the moving distance vector modulus, the direction persistence coefficient, and the stay point distribution entropy value. The random positioning noise signals are removed through the sliding window mechanism. Among them, the direction persistence coefficient is an index of the direction angle change trend calculated by the differential exponential smoothing algorithm, which is used to quantify the non-randomness degree of the terminal movement path. The differential exponential smoothing is a dynamic filtering method that combines the sliding window and gradient calculation, and adjusts the smoothing intensity according to the time weight of the data points, highlighting the waveform change trend when denoising.

[0067] The stay point distribution entropy value is a spatial dispersion index constructed based on the Shannon entropy theory, which measures the distribution randomness of the stay positions of the terminal in a specific area and is used to judge the difference between sightseeing behavior and emergency detention.

[0068] The direction persistence coefficient, which represents an index of the direction angle change trend calculated by the differential exponential smoothing algorithm, satisfies the following formula:

[0069] Among them, is the direction persistence coefficient of the current time slot, defined as the quantization value of the trend intensity of the change in the terminal movement direction angle, and is a dimensionless scalar after normalization processing; is a dynamic adjustment factor, defined as the forgetting factor of the sliding window data; is the original direction angle sequence, and is a dimensionless scalar after normalization processing; and represent the time slot number.

[0070] The Shannon entropy calculation formula of the stay point distribution entropy value satisfies the following formula:

[0071] Among them, is the proportion of the stay duration of the terminal in the th sub-block of the preset geographical grid.

[0072] Calculate the spatio-temporal synchronization characteristics between associated terminals; Specifically, select a set of mobile terminals with overlapping time windows within the same scenic area, and perform the following operations on every two terminals among them: 1. Based on the vector space projection of the dynamic vector model, calculate the Euclidean distance of the movement vectors of two terminals in the same time slice, and eliminate the instantaneous positioning error through the moving average of the time series; calculate the Euclidean distance of the movement vectors of two terminals in the same time slice, which satisfies the following formula:

[0073] where and are the plane rectangular coordinates after projection, with the unit of meter.

[0074] 2. Extract the continuous direction persistence coefficient sequences of the two terminals, and calculate the cosine similarity within the sliding time window as the direction collaboration index; 3. Set the distance difference threshold and the direction deviation threshold, and count the cumulative time length that continuously meets the threshold conditions. When it exceeds the preset threshold, trigger the synchronization event marking.

[0075] Generate the group type identifier and bind the policy template; Specifically, make a classification decision according to the combination mode of the synchronization feature calculation results: 1. Synchronization mode matching. If the difference in movement distance continuously remains below 2 meters and the direction cosine similarity is higher than 0.85, and the cumulative compliance duration exceeds 30 minutes, then output the parent-child interaction mode identifier; if the difference in movement distance is below 5 meters but there is a periodic position alternating leading feature, and the cumulative compliance duration in a single day exceeds 4 hours, then output the team collaboration mode identifier; 2. Policy template matching. According to the classification results, call the pre-stored policy library. The parent-child interaction mode is associated with the route interval optimization template and the emergency contact notification template, and the team collaboration mode is bound to the multi-person task recommendation template and the resource collaborative allocation model; Among them, the multi-person task recommendation template is a collaborative action suggestion plan generated according to the skill tags of team members, device compatibility, and task geographical distribution constraints, and includes the cross-terminal data synchronization interface and the conflict resolution rule library.

[0076] 3. Identifier dynamic update. When the spatio-temporal synchronization features of group members are detected to be broken in subsequent time periods, trigger the identifier weight attenuation mechanism until the cumulative deviation duration exceeds the tolerance threshold, and then update it to the independent tourist mode.

[0077] In one embodiment of the present invention, step S5 includes the following steps: Establish a dynamic weight mapping relationship library; Specifically, the parameter configuration unit of the server pre-sets a matching rule library for the user's physical exertion level and the scenic spot slope data. The user's physical exertion level is a quantization parameter obtained by analyzing the acceleration amplitude fluctuation frequency in the gyroscope waveform of the mobile terminal. The scenic spot slope data is the quantization value of the vertical climbing angle marked in the scenic area geographic information system. The rule library contains multiple key-value pairs corresponding to the relationship between the physical exertion level threshold and the slope threshold; Among them, the construction method of the matching rule library: divide the physical exertion level in the user's historical movement data into three intervals of low consumption, medium consumption, and high consumption according to the standard deviation of the step frequency change per minute. At the same time, extract the slope data recorded by the scenic area trail slope detector to generate a continuous piecewise function. When a tourist's current physical exertion level falls into the low consumption interval, the upper limit of the slope threshold allowed to be matched is degrees, corresponding to degrees in the medium consumption interval, and corresponding to degrees in the high consumption interval.

[0078] Generate the route intensity coefficient; Specifically, call the real-time physical exertion level data of the tourist and the slope data of the target scenic spot for real-time matching. If the current slope exceeds the upper limit of the threshold corresponding to the physical level in the matching rule library, mark the scenic spot as a high-risk section, and the route intensity coefficient is dynamically adjusted according to the step function. The route intensity coefficient is a standardized parameter representing the feasibility of the path. When the slope exceeds the limit, the coefficient decays by for each additional degree; Among them, the route intensity coefficient is dynamically adjusted according to the step function as follows: Initialize the coefficient to 1, calculate the difference between the current slope and the upper limit of the threshold, and each unit difference corresponds to linear decay factor, but when the difference exceeds 3 degrees, enable the exponential decay mode, and at the same time superimpose the continuous change rate of the user's current physical exertion level as a correction parameter, and the correction parameter is obtained from the spectral analysis result of the gyroscope waveform in the recent 5 minutes.

[0079] The route intensity coefficient is dynamically adjusted according to the step function and satisfies the following formula:

[0080] Among them, represents the final route intensity coefficient; represents the initial intensity coefficient; represents the physical exertion correction coefficient, which is calculated from the variance of the gyroscope acceleration spectrum; represents the physical exertion change rate, which is calculated by the derivative of the step frequency fluctuation amplitude collected by the mobile terminal within 5 minutes; all variables in the formula are dimensionless through standardization or ratio conversion to ensure the legality of addition and subtraction operations. represents the slope decay factor, by dividing the actual slope difference by the upper limit of the slope threshold Perform dimensionless processing to satisfy the following formula:

[0081] Activate the environmental capacity feedback channel; Specifically, receive the tolerance index in the dynamic environmental level set. The tolerance index is a scenic spot instantaneous carrying capacity parameter predicted by combining the spatial distribution density value of monitoring nodes and historical pressure-bearing data. Screen the candidate list with a tolerance index higher than the safety threshold among scenic spots of the same type through a sliding window; Among them, the method for screening scenic spots of the same type is: calculate the semantic similarity of the theme attribute labels of scenic spots, and combine the visual focus weight value in the user interest preference label. The cosine similarity of the labels of two scenic spots exceeds and the difference in visual focus weights is less than to be determined as of the same type.

[0082] Refer to Appendix Figure 2 , and execute off-peak recommendation sorting; Specifically, the scenic spots in the candidate list are sorted in descending order according to their tolerance indices. Insert the scenic spots with a route intensity coefficient higher than into the priority recommendation queue, and add a time decay factor to each scenic spot in the queue. The time decay factor is a weight parameter dynamically adjusted based on the remaining opening duration from the recommended time point. The weight of the scenic spot with a remaining duration of less than 1 hour is automatically increased ; Among them, the dynamically adjusted weight parameter is: the route intensity coefficient is the reference value. When the tolerance index exceeds the pressure-bearing safety line, for every exceeded, an additional gain coefficient is added, and a multiplication operation is performed with the time decay factor to generate the final sorting weight.

[0083] The dynamically adjusted weight parameter satisfies the following formula:

[0084] Among them, is the final sorting weight; is the tolerance gain coefficient; is the tolerance index of the current scenic spot; is the safety threshold preset by the system; is the time decay rate, which is dynamically adjusted according to the remaining opening duration of the scenic spot. When , otherwise ; The exponential term realizes time-sensitive weighting, and the shorter the opening time, the slower the weight decay. All variables are dimensionless parameters, and the exponential operation conforms to mathematical specifications.

[0085] Generate an initial set of recommended solutions; Specifically, the top five scenic spots with sorting weights in the priority recommendation queue are secondarily matched with the user interest preference tags. When the hesitation degree eigenvalue in the preference tags is higher than the preset warning line, the artificially preset alternative library expansion mechanism is triggered. The alternative library expansion mechanism is to call the hidden path data of non-popular scenic spots in the scenic area to reduce the congestion risk; Among them, the activation condition of the hidden path data is: when the average accommodation index of candidate scenic spots of the same type is lower than the safety threshold and the fluctuation range of the user's hesitation degree eigenvalue exceeds within three consecutive scan cycles then, the alternative path is automatically loaded into the recommended solution set and marked as a low-density priority option.

[0086] In one embodiment of the present invention, step S6 includes the following steps: Match degree threshold comparison; Specifically, calculate the cosine similarity between the feature vectors of each scenic spot in the initial recommended solution set and the feature vectors of the user interest preference tags. When the similarity calculation results of three consecutive recommended items are lower than the dynamically trained baseline, the secondary optimization threshold is triggered; Among them, the dynamic baseline represents an adaptive judgment criterion constructed based on the user's historical behavior data set, which is formed by statistically mapping the actual click conversion rate of the recommended content under the same environmental level conditions. During the training process, the backpropagation algorithm in the prior art is used to optimize the weight distribution coefficients of different feature dimensions.

[0087] Dynamic matching of strategy templates; Specifically, call the strategy recommendation template library corresponding to the group type identifier, and correct the template parameters according to the accommodation index in the current dynamic environmental level set. When the recognized group type is the parent-child interaction mode, preferentially load the template group containing the weight addition item of educational scenic spots; Among them, the strategy recommendation template library is a preset structural configuration file, and each template contains a scenic spot type preference matrix, a time allocation ratio threshold, and a safety factor boundary condition.

[0088] Template parameter correction, using the gradient descent method to iteratively calculate the response critical values of each parameter to the current environmental parameters.

[0089] Multi-source data cross-coupling; Specifically, compare the recommended solution set after secondary optimization with the dynamically updated dynamic environmental level using a geographical fence. If the highest matching scenic spot in the recommended solution is in an area where the instantaneous aggregation density exceeds the limit, the deep preference compensation mechanism is activated; Among them, the geographical fence comparison is to perform spatial coordinate conversion calculation on the regional heat distribution map generated by the Gaussian mixture model and the coordinates of the recommended scenic spots. When the coordinates of the center point of the scenic spot fall into the red warning area of the heat map and exceed the coverage area, it is determined as over-limit.

[0090] Emergency evacuation route injection; Specifically, extract alternative safe routes from the emergency evacuation priority data layer of the dynamic environment level set, generate a tree structure based on the shortest geometric distance between the user's current location coordinates and the entrance of the target scenic spot, and synchronously integrate real-time traffic control information for reachability verification; Among them, the shortest geometric distance generation tree is the output result of the improved algorithm. The improvement lies in introducing the user's physical exertion level as a non-linear correction factor for the path slope weight. Before the algorithm execution, it is necessary to load the three-dimensional terrain elevation data to construct the vertex set.

[0091] The shortest geometric distance generation tree is the output result of the improved algorithm and satisfies the following formula:

[0092] Among them, is the original geometric distance from path node to , which is normalized to value; is the absolute value of the slope at path node ; is the mapping coefficient of the user's physical exertion level. The physical level is divided into 5 levels, taking 0.15, 0.3, 0.45, 0.6, and 0.75; is the real-time traffic control correction function. When the path contains traffic control points, the output value is set to 1000 to force the path to be deprecated, otherwise it is 1; Perform linear superposition on the parameters after normalization processing (all dimensions are unified into dimensionless parameters).

[0093] Closed-loop feedback decision verification; Specifically, after outputting the final recommendation instruction, continuously monitor the actual response delay time of the user to the recommendation instruction. When the response time exceeds twice the standard deviation of the historical average of this group type, automatically roll back to the initial recommendation plan and recalculate the parameter weights; Among them, the response delay time is the time interval from receiving the recommendation instruction by the user terminal to generating the map zoom interaction or navigation confirmation operation. The abnormal response trigger mechanism realizes the dynamic evaluation of the state transition probability by constructing a Markov chain model of behavioral events within a sliding window.

[0094] Referring to the appended Figure 3 , the present invention also proposes an AI-based intelligent tourism recommendation system, including the following modules: A concealed monitoring network module configured to deploy monitoring nodes with ecological camouflage and integrate a laser ranging unit and a thermal imaging unit to generate instantaneous aggregation density data; An environmental perception computing module configured to fuse weather elements, traffic control information, and density data to generate a set of dynamic environmental levels; A terminal behavior analysis module configured to parse gyroscope features through a low-power Bluetooth link to generate a physical exertion level and synchronously process map zooming trajectories to generate interest preference tags; A group identification module configured to calculate the spatio-temporal synchronization features of mobile terminals, generate group type identifiers, and associate policy recommendation templates; A dynamic recommendation engine module configured to match the user's physical exertion level with scenic slope data to generate a route intensity coefficient, and combine the dynamic environmental level to perform off-peak recommendation sorting to form an initial set of recommendation schemes; An emergency path planning module configured to load three-dimensional terrain data to generate the shortest evacuation path including slope and control constraints.

[0095] Each of the above modules can be implemented in whole or in part by software, hardware, and their combination, supporting hardware form embedded in or independent of the processor in a computer device, and also supporting software form stored in the memory of the computer device for the processor to call and execute the operations corresponding to each of the above modules.

[0096] It should be noted that the human body information (including but not limited to human device information and personal information, etc.) and data (including but not limited to data for analysis, stored data, and displayed data, etc.) involved in the present invention are all information and data authorized by the human body or fully authorized by all parties. The collection, use, and processing of relevant data require relevant legal standards.

[0097] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An AI-based intelligent tourism recommendation method, characterized in that, It includes the following steps: S1. Obtain the instantaneous aggregation density data collected in real time by multiple concealed monitoring nodes in the scenic area; S2. Generate a set of dynamic environment levels based on the instantaneous aggregation density data, real-time weather elements, and traffic control information; S3. Establish a data interaction channel for mobile terminal behavior data, analyze the gyroscope acceleration sequence to generate the user's physical exertion level, and generate interest preference tags through the electronic map zoom operation trajectory; S4. Identify the spatio-temporal synchronization characteristics between associated mobile terminals to generate a group type identifier and bind a policy recommendation template; S5. Adjust the dynamic weight allocation of multi-dimensional recommendation parameters, match the user's physical exertion level with the scenic slope data to generate a route intensity coefficient, and combine the dynamic environment level to perform off-peak recommendation sorting to form an initial recommendation plan set; S6. Compare the matching degree between the initial recommendation plan and the interest preference tags. When it is lower than the set range, trigger secondary optimization of the recommended content based on the group type identifier, and inject an emergency evacuation route to complete the closed-loop recommendation decision chain.

2. The intelligent tourism recommendation method based on AI according to claim 1, wherein To obtain the instantaneous aggregation density data, it includes the following steps: Activate the laser ranging unit of the concealed monitoring node to perform three-dimensional trajectory scanning to generate a matrix of foot height values; Synchronously start the thermal imaging unit to perform dynamic density calibration on the laser ranging blind area to generate a dynamic correction coefficient; Construct a data collaborative verification network between monitoring nodes to generate redundant monitoring data streams for sudden surges in the number of people; Fuse the matrix of foot height values, the dynamic correction coefficient, and the redundant monitoring data streams to generate an instantaneous aggregation density data set including real-time credibility weights.

3. The intelligent tourism recommendation method based on AI according to claim 2, wherein, To generate a set of dynamic environment levels, it includes the following steps: Overlay the instantaneous aggregation density data with a temperature field interference compensation value to generate corrected density data; Perform multi-factor similarity matching between the corrected density data and the historical evolution pattern of emergencies to generate an accommodation index and a path smoothness coefficient; Establish an environmental level update trigger mechanism. When the monitoring node detects an abnormal movement trajectory, shorten the sliding window time and recalculate the environmental level; Output a set of dynamic environment levels including risk grading identifiers.

4. The intelligent tourism recommendation method based on AI according to claim 3, wherein To generate the user's physical exertion level, it includes the following steps: Perform frequency band separation on the gyroscope three-axis acceleration data, extract the vertical vibration component to construct a physical exertion evaluation matrix; Generate a high-frequency band energy integral value based on the number of peak mutations and the standard deviation coefficient; Generate a quantified physical exertion level according to the exceeding ratio of the high-frequency band energy integral value to the preset baseline.

5. An AI-based intelligent tourism recommendation method according to claim 3, characterized in that, To generate interest preference tags, it includes the following steps: Capture the user's zoom operation trajectory on the electronic map to generate a visual focus weight value; Statistically analyze the hesitation characteristics of the user's stay when the path selection is not completed; Perform cosine similarity matching between the visual focus weight value and the scenic spot attribute vector to generate preference tags.

6. The intelligent tourism recommendation method based on AI according to claim 1, characterized in that, To generate a group type identifier, it includes the following steps: Construct a dynamic vector model of the movement trajectory, calculate the direction persistence coefficient and the entropy value of the distribution of stop points; Compare the Euclidean distance of the movement vectors and the direction coordination index between associated terminals; Determine the parent-child interaction mode or the team collaboration mode according to the continuous spatio-temporal synchronization characteristics, and load the corresponding policy recommendation template.

7. An AI-based intelligent tourism recommendation method according to claim 4, characterized in that, To generate a route intensity coefficient, it includes the following steps: Establish a non-linear mapping relationship between the user's physical exertion level and the slope threshold; Adopt a step function to dynamically adjust the route intensity coefficient, and superimpose the attenuation factor obtained by spectral analysis on the path with over-limit slope; Generate a standardized route feasibility parameter by combining the real-time path smoothness coefficient.

8. An AI-based intelligent tourism recommendation method according to claim 4, wherein, Perform off-peak recommendation ranking in combination with the dynamic environment level, including the following steps: Screen the candidate list above the safety threshold according to the tolerance index; Superimpose the time decay factor on the candidate scenic spots to establish a dynamic weight mapping; Generate a priority recommendation queue through the secondary matching of the route intensity coefficient and the preference label.

9. The intelligent tourism recommendation method based on AI according to claim 8, wherein, Inject an emergency evacuation path, including the following steps: Perform geofence comparison through a Gaussian mixture model to mark the area with over-limit instantaneous aggregation density; Generate a three-dimensional shortest obstacle avoidance path integrating the physical exertion level; Embed the emergency evacuation path into the recommendation scheme set and monitor the user response to trigger feedback optimization.

10. An AI-based intelligent tourism recommendation system, characterized in that, Including the following modules: Covert monitoring network module, configured to deploy monitoring nodes with ecological camouflage, and integrate a laser ranging unit and a thermal imaging unit to generate instantaneous aggregation density data; Environmental perception calculation module, configured to generate a set of dynamic environment levels by integrating weather elements, traffic control information, and density data; Terminal behavior analysis module, configured to generate the physical exertion level by parsing the gyroscope features through a low-power Bluetooth link, and synchronously process the map zooming trajectory to generate interest preference labels; Group identification module, configured to calculate the spatio-temporal synchronization features of mobile terminals, generate group type identifiers, and associate policy recommendation templates; Dynamic recommendation engine module, configured to match the user's physical exertion level with the scenic spot slope data to generate a route intensity coefficient, and perform off-peak recommendation ranking in combination with the dynamic environment level to form an initial recommendation scheme set; Emergency path planning module, configured to load three-dimensional terrain data to generate the shortest evacuation path including slope and control constraints.

Images