Intelligent navigation method for stadium digital service

Abstract

The invention is applicable to the technical field of venue navigation, and provides an intelligent navigation method for venue digital service, which comprises the following steps: acquiring positioning information, behavior track and interaction record of a user; performing combined analysis on the user data and preset venue interest point attribute data to obtain a user feature vector containing a stay preference coefficient, movement speed analysis and a subject interest weight; extracting a topological relation between the current position and the target interest point according to the user feature vector, and calculating a constraint parameter according to the topological relation; based on the constraint parameters, constructing an initial path planning model taking the minimum expected passing time as a target function; determining a people flow density gradient map of each region of the stadium in a preset time period through the obtained historical people flow data and the user real-time positioning data; the initial path planning model is optimized by using the people flow density gradient map, and the optimal path is generated by dynamically adjusting navigation path nodes, so that the intelligence level and the user satisfaction of venue navigation are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of venue navigation, and in particular to an intelligent navigation method for venue digital services. Background Art

[0002] Currently, museums, exhibition halls, and large shopping malls are beginning to introduce indoor positioning technologies and navigation systems for densely populated and complex spaces. Traditional Bluetooth and Wi-Fi positioning, while low-cost and easy to deploy, suffer from limited accuracy due to signal susceptibility to interference, often with errors of several meters or even more than ten meters. This creates a poor experience for users who require precise guidance to specific exhibits or store locations. Most intelligent navigation technologies have been proposed to address this issue, aiming to provide the shortest navigation path and the shortest possible route. However, each visitor has unique interests, hobbies, visit purposes, and schedules, making it difficult for navigation systems to provide personalized routes based on user preferences and behavior patterns. Furthermore, when a certain area becomes crowded and congested, the system is unable to promptly reroute users based on real-time conditions, forcing them to wait for extended periods or travel slowly through congested areas, severely impacting their visitor experience. Summary of the Invention

[0003] The embodiments of the present application provide an intelligent navigation method for venue digital services, which is used to solve the problem that the lack of personalized design in venue navigation seriously affects the user's visiting experience.

[0004] A first aspect of an embodiment of the present application provides an intelligent navigation method for a venue digital service, including:

[0005] Collect user behavior data in real time through multi-source sensors, including location information, behavior trajectory, and interaction records;

[0006] The user behavior data is combined with the preset venue interest point attribute data for analysis to obtain a multi-dimensional user feature vector including the stay preference coefficient, movement speed analysis and subject interest weight;

[0007] Extracting a topological relationship between the user's current location and the target point of interest based on the multi-dimensional user feature vector, and calculating constraint parameters based on the topological relationship, wherein the constraint parameters include path length weight, historical congestion probability, and real-time crowd density distribution;

[0008] An initial path planning model is constructed based on the constraint parameters with minimization of expected travel time as the objective function, wherein the constraints of the initial path planning model include capacity constraints, interest matching constraints, physical accessibility constraints, and time sensitivity constraints;

[0009] Determine the crowd density gradient map of each area of ​​the venue within a preset time period by obtaining historical crowd flow data and user real-time location data. The crowd density gradient map includes status indicators for normal passage, slow traffic warning, and congestion prohibition;

[0010] The initial path planning model is optimized using the crowd density gradient map, and a three-dimensional evaluation path that integrates the shortest physical distance, the lowest congestion risk, and the highest interest matching degree is generated by dynamically adjusting the navigation path nodes.

[0011] Furthermore, the user behavior data is combined with the preset venue interest point attribute data for analysis to obtain a multi-dimensional user feature vector including a stay preference coefficient, movement speed analysis and subject interest weight, including:

[0012] Perform spatial density analysis on the user behavior data using the OPTICS density clustering algorithm to extract the spatial distribution pattern of user behavior;

[0013] Time series modeling is used to analyze user behavior sequences in segments, and the time dimension feature vector is determined by the time decay function and the calculated dynamic interest weight;

[0014] After fusion of spatial distribution pattern and time dimension feature vector, a multi-dimensional user feature vector including stay preference, movement speed and topic interest is generated.

[0015] Furthermore, the topological relationship between the user's current location and the target point of interest is extracted based on the multi-dimensional user feature vector, and constraint parameters are calculated based on the topological relationship. The constraint parameters include path length weight, historical congestion probability, and real-time crowd density distribution, including:

[0016] The directed weighted graph of the topological relationship between the user's current location and the target point of interest is:

[0017] Vertex set V = {v1,v2,…,v n} represents the points of interest and the intersection of paths, and the edge set E={(v i ,v j )│v i With v j Direct access}, where (v i ,v j ) is the vertex v i to v j An edge, i and j represent the starting point and end point of the edge, and the weight set W={w ij │w ij =f(d ij ,s ij ,c ij )}, where d ij For vi to v j The physical path length, s ij For v i to v j The path type weight, c ij For v i to v j The path capacity coefficient of

[0018] Path length weight C d The calculation formula is:

[0019]

[0020] Where: α is a dynamic adjustment factor, which is negatively correlated with the user's movement speed distribution; Path is a set of continuous edges from the starting point to the end point;

[0021] Historical congestion probability P e (v i ,v j ) is calculated as follows:

[0022]

[0023] Where: P e (v i ,v j ) is the path (v i ,v j ), K is the number of historical data sampling windows, is the congestion indicator function, when congestion occurs Normal γ is the time attenuation coefficient, T current ―T k The time difference between the current time and the historical record;

[0024] Real-time crowd density ρ r (R m ) is calculated as follows:

[0025]

[0026] Where: N people (R m ) is the venue area R m Real-time number of people in Area(R m ) is the venue area R m area.

[0027] Furthermore, the initial path planning model is constructed based on the constraint parameters with minimization of expected travel time as the objective function, and the constraints of the initial path planning model include capacity constraints, interest matching constraints, physical accessibility constraints, and time sensitivity constraints, including:

[0028] The expression of the objective function is as follows:

[0029]

[0030] Where: T total To minimize the expected travel time, μ v is the average moving speed of the user, η is the historical congestion impact coefficient, λ is the interest matching weight, which is dynamically calculated according to user characteristics and is used to adjust the impact of interest matching on the objective function. path is the interest matching degree of the path, which indicates the matching degree between the path and the user’s interest. m(k) For vertex v k The interest matching degree of the subject category, |Path| is the total number of path nodes, β ij is the path edge (v i ,v j )'s real-time density impact factor.

[0031] Furthermore, the initial path planning model is constructed based on the constraint parameters with minimization of expected travel time as the objective function, and the constraints of the initial path planning model include capacity constraints, interest matching constraints, physical accessibility constraints, and time sensitivity constraints, including:

[0032] The expression of capacity constraint is:

[0033]

[0034] Where: c ij is the path edge (v i ,v j )'s maximum capacity;

[0035] Expression of interest matching constraint:

[0036] TIM path ≥θ (θ=0.6~0.8)

[0037] Where: θ is the minimum threshold of interest matching;

[0038] The expression of the physical reachability constraint is:

[0039]

[0040] Where: slope(v i ,vj ) is the path edge (v i ,v j ) slope;

[0041] Expression for time sensitivity constraint:

[0042] T total ≤δ×T shortest (δ=1.2~1.5)

[0043] Where: T shortest is the travel time of the shortest physical path without considering congestion and interest matching, and δ is the time sensitivity coefficient.

[0044] Furthermore, the crowd density gradient map of each area of ​​the venue within a preset time period is determined by obtaining historical crowd flow data and user real-time location data. The crowd density gradient map includes status indicators of normal passage, slow traffic warning, and congestion prohibition, including:

[0045] Through the spatiotemporal graph convolutional network, historical crowd flow data and real-time user location data are integrated to establish a prediction model that includes time decay factors and spatial propagation coefficients.

[0046] Determine the crowd density of each area of ​​the venue within a preset time period according to the prediction model;

[0047] The crowd density is mapped into status indicators of normal passage, slow-moving warning, and congestion-prohibited passage, and a visual crowd density gradient map is generated in combination with the digital twin map.

[0048] Furthermore, the historical crowd flow data and the user's real-time location data are fused and processed through the spatiotemporal graph convolutional network to establish a crowd flow density prediction model that includes a time decay factor and a spatial propagation coefficient, including:

[0049] The calculation formula of the prediction model is:

[0050]

[0051] in: and is the input feature and output feature of the graph convolution layer, including spatial propagation information, Θ k is the trainable convolution kernel parameter of the k-th order Chebyshev polynomial, T k (·) is the kth order Chebyshev polynomial used to approximate the graph convolution operation, α t′ is the time decay factor at historical moment t′, is the characteristic of historical moment t′, T h is the total historical time, is the normalized Laplace matrix, Contains spatial propagation information, λ max is its maximum eigenvalue, I is the identity matrix;

[0052] The calculation formula of time decay factor is:

[0053]

[0054] Where: α t′ is the influence weight of historical moment t′ on the current prediction moment t, γ is the time decay rate parameter, which controls the weight decay speed of historical data;

[0055] The calculation formula of spatial propagation coefficient is:

[0056]

[0057] Among them: A pq is the spatial propagation intensity from region p to region q, d(p,q) is the Euclidean distance from region p to region q, σ is the distance attenuation coefficient, D max is the maximum impact distance.

[0058] Furthermore, determining the crowd density of each area of ​​the venue within a preset time period according to the prediction model includes:

[0059] The calculation formula for crowd density is as follows:

[0060]

[0061] in: is the predicted crowd density at the future time t, W r is the weight matrix of the regression branch, obtained through training, is the feature representation of the decoder output, Contains time decay factor α t′ Weighted historical data, including the spatial propagation coefficient A extracted by the graph convolution layer pq , b r is the bias term of the regression branch, which is learned through training.

[0062] Furthermore, the initial path planning model is optimized using the crowd density gradient map, and a three-dimensional evaluation path integrating the shortest physical distance, the lowest congestion risk, and the highest interest matching degree is generated by dynamically adjusting the navigation path nodes, including:

[0063] The candidate path set P output by the initial path planning model is n} is mapped to the corresponding grid cells of the congestion gradient map, generating the path and congestion association matrix C∈R n×m ;

[0064] Calculating the congestion risk corresponding to the candidate path set according to the path and congestion association matrix;

[0065] Determine a target path planning model using the physical distance, congestion risk, and interest matching degree corresponding to the candidate path set as evaluation parameters and minimizing the expected travel time as the target parameter;

[0066] The optimal path output by the target path planning model is mapped to a three-dimensional digital twin map to generate a personalized navigation route adapted to the user.

[0067] Furthermore, the calculating of the congestion risk corresponding to the candidate path set according to the path-congestion association matrix includes:

[0068]

[0069] Where: r i The candidate path set P = {P1, P2, ..., P n} corresponding congestion risk, C ij For path P i The congestion level of the grid where the j-th node is located, ω j is the weight coefficient of the congestion level of the jth node.

[0070] It can be seen from the above technical solutions that the embodiments of the present invention have the following advantages:

[0071] The present invention analyzes data collected in real time by multiple sensors to generate a multidimensional user feature vector containing a dwell preference coefficient, movement speed distribution, and topic interest weights. This vector accurately captures users' personalized needs and behavior patterns, providing high-quality data support for subsequent route planning. Based on the user feature vector and venue topology, a multi-objective optimization algorithm is used to generate a set of candidate routes. This algorithm not only considers physical distance and interest matching, but also introduces a dynamic weight adjustment mechanism to ensure that the generated routes strike a balance between efficiency and user experience. Historical crowd flow data is integrated with real-time positioning data to establish a prediction model that includes a time decay factor and a spatial propagation coefficient. This model then outputs a crowd density gradient map for each area within a set future time period, providing dynamic data support for route optimization. The crowd density gradient map is then used to optimize the initial route planning model, generating a three-dimensional evaluation route that combines the shortest physical distance, the lowest congestion risk, and the highest interest matching. This reduces travel time and congestion risk, while improving the interest matching of the route. The optimal route is then mapped onto a three-dimensional digital twin map to generate a personalized navigation route tailored to the user. This completes a closed loop from user behavior analysis to route planning to dynamic optimization, significantly improving the intelligence level of the navigation system and user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 This is a schematic flow chart of an embodiment of an intelligent navigation method for venue digital services in the present invention. DETAILED DESCRIPTION

[0073] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0074] The intelligent navigation method for venue digital services in this embodiment is used to improve the intelligence level of the navigation system and user satisfaction. The implementation method in this embodiment can be implemented in the system, on the server, or on the terminal, without specific limitation.

[0075] Example 1

[0076] See also Figure 1 An embodiment of an intelligent navigation method for venue digital services in the present invention includes the following steps:

[0077] S11. Collect user behavior data in real time through multi-source sensors. User behavior data includes location information, behavior trajectory, and interaction records.

[0078] In this embodiment, the multi-source sensors include UWB (ultra-wideband) positioning systems, Bluetooth beacons, or inertial measurement units (IMUs). Positioning information includes the user's real-time location coordinates, movement direction, and movement speed, which is used to track the user's location in real time and generate a behavioral trajectory. The behavioral trajectory includes the user's continuous movement path, stop locations, stop duration, and visit sequence, which is used to analyze user behavior patterns. Interaction records include the user's interaction behavior with venue facilities, interaction duration, and interaction frequency, which are used to quantify user interests and preferences.

[0079] S12. Analyze the user behavior data and the preset venue interest point attribute data to obtain a multidimensional user feature vector including the stay preference coefficient, movement speed distribution, and subject interest weight;

[0080] S121. Perform spatial density analysis on user behavior data using the OPTICS density clustering algorithm to extract the spatial distribution pattern of user behavior;

[0081] Based on the user's behavior trajectory, including the location and duration of the stay point, the stay points are denoised to remove outliers. First, the core parameters of the OPTICS density clustering algorithm are set, with a neighborhood radius of 3 meters and a minimum number of core object points of 5. The reachable distance from each object p to the core object q is calculated as follows:

[0082] reachability_dist(p,q)=max{core_dist(q),euclidean(p,q)}

[0083] Where: reachability_dist(p,q) is the reachability distance from point p to the core object q, core_dist(q) is the distance from q to its nearest neighbor with the smallest number of core objects, and euclidean(p,q) is the Euclidean distance between points p and q.

[0084] By analyzing the reachability graph and identifying steep drop-off points as cluster boundaries, the following patterns were extracted: Deep Exploration: dwell time exceeding a set threshold (here, 150 seconds) and movement speed below a set threshold (here, 0.4 m / s) meet the definition of Deep Exploration; Fast Passage: dwell time below a set threshold (here, 20 seconds) and movement speed above a set threshold (here, 1.6 m / s) meet the definition of Fast Passage; Social Aggregation: spatial overlap of multiple user trajectories greater than 70%. Outliers with reachability distances greater than two neighborhood radii were labeled as noise data. The OPTICS results were revalidated using the DBSCAN algorithm to ensure cluster consistency. Finally, the spatial distribution patterns of user behavior (e.g., Deep Exploration, Fast Passage, etc.) were generated, along with representative dwell points and movement paths for each pattern.

[0085] S122. Use time series modeling to perform segmented analysis on the user behavior sequence, and determine the time dimension feature vector using the time decay function and the calculated dynamic interest weight;

[0086] The user's behavior sequence is segmented into multiple subsequences using 30-second time windows. Key features are extracted within each time window, including movement speed, number of visits to points of interest, and duration of interactive device use. Next, the Transformer temporal attention model is employed to analyze the behavior sequence using an encoder-decoder structure. In the encoder, a sine function is used to encode the temporal position, incorporating temporal information. A multi-head attention mechanism is used to calculate the attention weight for each time window and determine the temporal dependencies of user behavior. Finally, a time decay function is designed, introducing exponentially decaying weights to strengthen the influence of recent behavior and weaken the influence of more distant behavior. Based on the attention weight matrix, the user's sustained interest components and burst interest components are extracted.

[0087] Finally, the time dimension feature vector is output, which contains the time evolution characteristics of stay preference, movement speed and topic interest.

[0088] S123. Generate a multi-dimensional user feature vector including stay preference, movement speed and topic interest by fusing the spatial distribution pattern with the time dimension feature vector.

[0089] The above spatial distribution pattern features and time dimension feature vectors are concatenated and dimensionally compressed, and the features are standardized to finally output a multi-dimensional user feature vector, including stay preferences, movement speed, and topic interests. The calculation process is as follows:

[0090] Stay preference coefficient SPC:

[0091]

[0092] Of which: SPC i is the user's preference coefficient for staying at point of interest i, and is the length of time the user stays at points of interest i and j in the tth time window, λ is the time decay factor, T is the total number of time windows, τ i and τ j are the benchmark stay time thresholds of points of interest i and j respectively, and N is the total number of points of interest in the venue.

[0093] Moving speed distribution MSD:

[0094]

[0095] Where: v (t) is the moving speed in the tth time window, μ v is the average moving speed, σ v is the velocity standard deviation, γ v is the speed deviation, and K is the total number of effective moving periods.

[0096] Subject interest weight TIW:

[0097]

[0098] Among relatives: TIW m is the user's interest weight on topic m, C m and C m′ The set of all interest points belonging to topics m and m′, A j is the user's active interaction intensity on the theme-related content, α and β are the fusion weight coefficients, and M is the total number of theme categories defined by the venue.

[0099] A j Calculated as:

[0100]

[0101] in: is the number of interactions between the user and topic m in the t-th time window, is the interaction time, d max The maximum duration threshold for a single interaction set by the system.

[0102] S13 extracts the topological relationship between the user's current location and the target point of interest based on the multidimensional user feature vector, and calculates the constraint parameters based on the topological relationship. The constraint parameters include the path length weight, the historical congestion probability, and the real-time crowd density distribution;

[0103] The directed weighted graph of the topological relationship between the user's current location and the target point of interest is:

[0104] Vertex set V = {v1,v2,…,v n} represents the points of interest and the intersection of paths, and the edge set E={(v i ,v j )│v i With v j Direct access}, where (v i ,v j ) is the vertex v i to v j An edge, i and j represent the starting point and end point of the edge, and the weight set W={w ij │w ij =f(d ij ,s ij ,c ij )}, where d ij For v i to v j The physical path length, s ij For v i to v j The path type weight, c ij For v i to v j The path capacity coefficient of

[0105] Path length weight C d The calculation formula is:

[0106]

[0107] Where: α is a dynamic adjustment factor, which is negatively correlated with the user's movement speed distribution; Path is a set of continuous edges from the starting point to the end point;

[0108] Historical congestion probability P e (v i ,v j ) is calculated as follows:

[0109]

[0110] Where: P e (v i ,v j ) is the path (v i ,v j ), K is the number of historical data sampling windows, is the congestion indicator function, when congestion occurs Normal γ is the time attenuation coefficient, T current ―T k The time difference between the current time and the historical record;

[0111] Real-time crowd density ρ r (R m ) is calculated as follows:

[0112]

[0113] Where: N people (R m ) is the venue area R m Real-time number of people in Area(R m ) is the venue area R m area.

[0114] By calculating the path length weights mentioned above, the physical distance of the path can be quantified, ensuring that the generated path is as short as possible and reducing the user's travel time. The historical congestion probability can quantify the possibility of congestion on the path in the past and predict the congestion risk that may occur in the future. The real-time crowd density distribution can quantify the crowd density in the current area and reflect the real-time congestion situation.

[0115] S14. Based on the constraint parameters, construct an initial path planning model with minimizing the expected travel time as the objective function. The constraints of the initial path planning model include capacity constraints, interest matching constraints, physical accessibility constraints, and time sensitivity constraints.

[0116] Based on the constraint parameters determined in the above steps, the initial path planning model is constructed. The objective function is expressed as follows:

[0117]

[0118] Where: T total To minimize the expected travel time, μ v is the average moving speed of the user, η is the historical congestion impact coefficient, λ is the interest matching weight, which is dynamically calculated according to the user characteristics and is used to adjust the impact of interest matching on the objective function. path is the interest matching degree of the path, which indicates the matching degree between the path and the user’s interest. m(k) For vertex vk The interest matching degree of the subject category, |Path| is the total number of path nodes, β ij is the path edge (v i ,v j )'s real-time density impact factor.

[0119] The expression of capacity constraint is:

[0120]

[0121] Where: c ij is the path edge (v i ,v j )'s maximum capacity;

[0122] Expression of interest matching constraint:

[0123] TIM path ≥θ (θ=0.6~0.8)

[0124] Where: θ is the minimum threshold of interest matching;

[0125] The expression of the physical reachability constraint is:

[0126]

[0127] Where: slope(v i ,v j ) is the path edge (v i ,v j ) slope;

[0128] Expression for time sensitivity constraint:

[0129] T total ≤δ×T shortest (δ=1.2~1.5)

[0130] Where: T shortest is the travel time of the shortest physical path without considering congestion and interest matching, and δ is the time sensitivity coefficient.

[0131] Based on the combination of the above objective function and constraints, the generated path is ensured to achieve the optimal balance between physical distance, congestion risk and interest matching.

[0132] S15. Determine the crowd density gradient map for each area of ​​the venue within a preset time period by obtaining historical crowd data and real-time user location data. The crowd density gradient map includes status indicators for normal traffic, slow traffic warning, and congestion prohibition;

[0133] S151. Use a spatiotemporal graph convolutional network to fuse historical crowd flow data with real-time user location data to establish a crowd flow density prediction model that includes a time decay factor and a spatial propagation coefficient.

[0134] Specifically, the time decay factor is used to weight the historical data to extract the time dimension features, and then the graph convolution layer is constructed through the spatial propagation coefficient to capture the crowd density distribution characteristics in the spatial dimension.

[0135] The calculation formula of the crowd density prediction model is:

[0136]

[0137] in: and is the input feature and output feature of the graph convolution layer, including spatial propagation information, Θ k is the trainable convolution kernel parameter of the k-th order Chebyshev polynomial, T k (·) is the kth order Chebyshev polynomial used to approximate the graph convolution operation, α t′ is the time decay factor at historical moment t′, is the characteristic of historical moment t′, T h is the total historical time, is the normalized Laplace matrix, Contains spatial propagation information, λ max is its maximum eigenvalue, I is the identity matrix;

[0138] The calculation formula of time decay factor is:

[0139]

[0140] Where: α t′ is the influence weight of historical moment t′ on the current prediction moment t, γ is the time decay rate parameter, which controls the weight decay speed of historical data;

[0141] The calculation formula of spatial propagation coefficient is:

[0142]

[0143] Among them: A pq is the spatial propagation intensity from region p to region q, d(p,q) is the Euclidean distance from region p to region q, σ is the distance attenuation coefficient, D max is the maximum impact distance.

[0144] S152. Determine the crowd density of each area of ​​the venue within a preset time period based on the prediction model;

[0145] The features output by the above graph convolution layer Input decoder, get The crowd density is calculated through the regression branch, and the calculation formula is as follows:

[0146]

[0147] in: is the predicted crowd density at the future time t, W r is the weight matrix of the regression branch, obtained through training, is the feature representation of the decoder output, Contains time decay factor α t′ Weighted historical data, including the spatial propagation coefficient A extracted by the graph convolution layer pq , b r is the bias term of the regression branch, which is learned through training.

[0148] S153. Map the pedestrian flow density into status indicators for normal passage, slow-moving warning, and congestion-prohibited entry, and combine it with the digital twin map to generate a visual pedestrian flow density gradient map.

[0149] The predicted crowd density Mapped to the third-level status mark: normal passage Marked in green, indicating that the area is unobstructed; slow-moving warning The yellow mark indicates that the area is becoming congested; no driving is allowed in the event of congestion. Areas marked in red indicate severe congestion. By integrating status indicators with the digital twin map, the crowd density gradient map is smoothed using Gaussian filtering to eliminate predicted noise. WebGL technology is then used to render the resulting visual crowd density gradient map in real time. This map, with dynamic color gradients and area markers, intuitively displays the real-time congestion status of each area within the venue, enabling users and operators to quickly identify congested areas and make informed decisions.

[0150] S16. Optimize the initial path planning model using the crowd density gradient map, and generate a three-dimensional evaluation path that integrates the shortest physical distance, the lowest congestion risk, and the highest interest matching by dynamically adjusting the navigation path nodes.

[0151] S161. The candidate path set P output by the initial path planning model = {P1, P2, ..., P n} is mapped to the corresponding grid cells of the congestion gradient map, generating the path and congestion association matrix C∈R n×m ;

[0152] Specifically, for P = {P1, P2, ..., P n}, each path P i Contains a node sequence of {v1,v2,…,v m}; Grid density state map, each grid cell contains the congestion level. Each path P i Node v j The corresponding grid cell R mapped to the congestion gradient map ij , record each node v j Congestion level of the grid unit. Construct matrix C∈R n×m , where element C ij Represents path P i No. v j The congestion level of the grid where the node is located is defined by the grid cell state in the congestion gradient map. Each grid cell contains a congestion level, which is used to represent the degree of congestion in the area.

[0153] S162. Calculate the congestion risk corresponding to the candidate path set based on the path and congestion association matrix;

[0154]

[0155] Where: r i The candidate path set P = {P1, P2, ..., P n} corresponding congestion risk, C ij For path P i The congestion level of the grid where the j-th node is located, ω j is the weight coefficient of the congestion level of the jth node.

[0156] S163. Determine a target path planning model using the physical distance, congestion risk, and interest matching degree corresponding to the candidate path set as evaluation parameters and minimizing the expected travel time as the target parameter;

[0157] The congestion risk corresponding to the candidate path set calculated in step S162 is incorporated into the calculation of the objective function of the initial planning model constructed in step S14 above, ensuring that the generated path is optimal between physical distance, congestion risk and interest matching, and obtaining an updated target path planning model. Based on real-time congestion data, the path planning results are dynamically adjusted to prevent users from entering congested areas. Finally, the improved A* algorithm is used to solve the objective function and generate the optimal path P * .

[0158] S164. Map the optimal path output by the target path planning model to the three-dimensional digital twin map to generate a personalized navigation route adapted to the user.

[0159] The optimal path P *The system maps the node sequence to the corresponding coordinates of a 3D digital twin map, smoothing the path turning points using Bezier curves to generate a continuous 3D path. Dynamic arrows are superimposed on the path to indicate the user's direction of travel. Voice navigation is triggered at key decision points (such as turns and congested areas). SLAM technology registers the path arrows with the real scene, and vibration intensity encodes the urgency of navigation instructions. Changes in the congestion status of path nodes are detected every Δt = 30 seconds, dynamically adjusting the navigation route.

[0160] Assume that the topology of the venue contains nodes V = {v c ,v1,v2,v3,v t}, the edge is: E={(v c ,v1),(v c ,v2),(v1,v3),(v2,v3),(v3,v t )}; Path P1 = [v c ,v1,v3,v t ], mapped to the congestion gradient map, the congestion level is [0, 1, 2], and the first row of the path and congestion association matrix C is generated as [0, 1, 2]. The congestion risk of path P1 is calculated as: r1 = 0.5 × 0 + 1.0 × 1 + 2.0 × 2 = 5.0. The expected travel time T of path P1 is calculated based on the objective function. total , generate the optimal path P through the improved A* algorithm * =[v c ,v2,v3,v t ]. The optimal path P * Mapped to a three-dimensional digital twin map, it generates dynamic arrow guidance and voice prompts, detects the congestion status of path nodes in real time, and dynamically adjusts the navigation route.

[0161] The above embodiments achieve a complete closed loop from user behavior analysis to path planning to dynamic optimization through multi-level data fusion, optimization algorithms and visualization technology, significantly improving the intelligence level of the navigation system and user satisfaction.

[0162] Those skilled in the art will appreciate that the units of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0163] In the embodiments provided herein, it should be understood that the division of units is merely a logical functional division. In actual implementation, other division methods may be employed, such as combining multiple units into one unit, splitting a unit into multiple units, or ignoring certain features. Furthermore, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically as a separate unit, or two or more units may be integrated into a single unit. These integrated units may be implemented in either hardware or software functional units.

[0164] If the integrated unit is implemented in the form of 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 the present invention, 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. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nly Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program code.

[0165] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. An intelligent navigation method for venue digital services, characterized in that: include: Collect user behavior data in real time through multi-source sensors, including location information, behavior trajectory, and interaction records; The user behavior data is combined with the preset venue interest point attribute data for analysis to obtain a multi-dimensional user feature vector including the stay preference coefficient, movement speed analysis and subject interest weight; Extracting a topological relationship between the user's current location and the target point of interest based on the multi-dimensional user feature vector, and calculating constraint parameters based on the topological relationship, wherein the constraint parameters include path length weight, historical congestion probability, and real-time crowd density distribution; An initial path planning model is constructed based on the constraint parameters with minimization of expected travel time as the objective function, wherein the constraints of the initial path planning model include capacity constraints, interest matching constraints, physical accessibility constraints, and time sensitivity constraints; Determine the crowd density gradient map of each area of ​​the venue within a preset time period by obtaining historical crowd flow data and user real-time location data. The crowd density gradient map includes status indicators for normal passage, slow traffic warning, and congestion prohibition; The initial path planning model is optimized using the crowd density gradient map, and a three-dimensional evaluation path that integrates the shortest physical distance, the lowest congestion risk, and the highest interest matching degree is generated by dynamically adjusting the navigation path nodes.

2. The intelligent navigation method for venue digital services according to claim 1, characterized in that: The user behavior data is combined with the preset venue interest point attribute data for analysis to obtain a multi-dimensional user feature vector including a stay preference coefficient, movement speed analysis, and subject interest weight, including: Perform spatial density analysis on the user behavior data using the OPTICS density clustering algorithm to extract the spatial distribution pattern of user behavior; Time series modeling is used to analyze user behavior sequences in segments, and the time dimension feature vector is determined by the time decay function and the calculated dynamic interest weight; After fusion of spatial distribution pattern and time dimension feature vector, a multi-dimensional user feature vector including stay preference, movement speed and topic interest is generated.

3. The intelligent navigation method for venue digital services according to claim 1, characterized in that: The method extracts the topological relationship between the user's current location and the target point of interest based on the multi-dimensional user feature vector, and calculates constraint parameters based on the topological relationship, wherein the constraint parameters include path length weight, historical congestion probability, and real-time crowd density distribution, including: The directed weighted graph of the topological relationship between the user's current location and the target point of interest is: Vertex set V = {v1,v2,…,v n } represents the points of interest and the intersection of paths, and the edge set E={(v i ,v j )|v i With v j Direct access}, where (v i ,v j ) is the vertex v i to v j An edge, i and j represent the starting point and end point of the edge, and the weight set W={w ij |w ij =f(d ij ,s ij ,c ij )}, where d ij For v i to v j The physical path length, s ij For v i to v j The path type weight, c ij For v i to v j The path capacity coefficient of Path length weight C d The calculation formula is: Where: α is a dynamic adjustment factor, which is negatively correlated with the user's movement speed distribution; Path is a set of continuous edges from the starting point to the end point; Historical congestion probability P e (v i ,v j ) is calculated as follows: Where: P e (v i ,v j ) is the path (v i ,v j ), K is the number of historical data sampling windows, is the congestion indicator function, when congestion occurs Normal γ is the time attenuation coefficient, T current ―T k The time difference between the current time and the historical record; Real-time crowd density ρ r (R m ) is calculated as follows: Where: N people (R m ) is the venue area R m Real-time number of people in Area(R m ) is the venue area R m area.

4. The intelligent navigation method for venue digital services according to claim 1, characterized in that: The initial path planning model is constructed based on the constraint parameters with minimization of expected travel time as the objective function, wherein the constraints of the initial path planning model include capacity constraints, interest matching constraints, physical accessibility constraints, and time sensitivity constraints, including: The expression of the objective function is as follows: Where: T total To minimize the expected travel time, μ v is the average moving speed of the user, η is the historical congestion impact coefficient, λ is the interest matching weight, which is dynamically calculated according to user characteristics and is used to adjust the impact of interest matching on the objective function. path is the interest matching degree of the path, which indicates the matching degree between the path and the user’s interest. m(k) For vertex v k The interest matching degree of the subject category, |Path| is the total number of path nodes, β ij is the path edge (v i ,v j )'s real-time density impact factor.

5. The intelligent navigation method for venue digital services according to claim 4, characterized in that: The initial path planning model is constructed based on the constraint parameters with minimization of expected travel time as the objective function, wherein the constraints of the initial path planning model include capacity constraints, interest matching constraints, physical accessibility constraints, and time sensitivity constraints, including: The expression of capacity constraint is: Where: c ij is the path edge (v i ,v j )'s maximum capacity; Expression of interest matching constraint: TEAM path ≥θ(θ=0.6~0.8) Where: θ is the minimum threshold of interest matching; The expression of the physical reachability constraint is: Where: slope(v i ,v j ) is the path edge (v i ,v j ) slope; Expression for time sensitivity constraint: T total ≤δ×T shortest (δ=1.2~1.5) Where: T shortest is the travel time of the shortest physical path without considering congestion and interest matching, and δ is the time sensitivity coefficient.

6. The intelligent navigation method for venue digital services according to claim 1, characterized in that: The crowd density gradient map of each area of ​​the venue within a preset time period is determined by obtaining historical crowd flow data and user real-time positioning data. The crowd density gradient map includes status indicators of normal passage, slow traffic warning, and congestion prohibition, including: Through the spatiotemporal graph convolutional network, historical crowd flow data and real-time user location data are integrated to establish a prediction model that includes time decay factors and spatial propagation coefficients. Determine the crowd density of each area of ​​the venue within a preset time period according to the prediction model; The crowd density is mapped into status indicators of normal passage, slow-moving warning, and congestion-prohibited passage, and a visual crowd density gradient map is generated in combination with the digital twin map.

7. The intelligent navigation method for venue digital services according to claim 6, characterized in that: The method uses a spatiotemporal graph convolutional network to fuse historical crowd flow data with real-time user location data to establish a crowd flow density prediction model that includes a time attenuation factor and a spatial propagation coefficient, including: The calculation formula of the prediction model is: in: and is the input feature and output feature of the graph convolution layer, including spatial propagation information, Θ k is the trainable convolution kernel parameter of the k-th order Chebyshev polynomial, T k (·) is the kth order Chebyshev polynomial used to approximate the graph convolution operation, α t′ is the time decay factor at historical moment t′, is the characteristic of historical moment t′, T h is the total historical time, is the normalized Laplace matrix, Contains spatial propagation information, λ max is its maximum eigenvalue, I is the identity matrix; The calculation formula of time decay factor is: Where: α t′ is the influence weight of historical moment t′ on the current prediction moment t, γ is the time decay rate parameter, which controls the weight decay speed of historical data; The calculation formula of spatial propagation coefficient is: Among them: A pq is the spatial propagation intensity from region p to region q, d(p,q) is the Euclidean distance from region p to region q, σ is the distance attenuation coefficient, D max is the maximum impact distance.

8. The intelligent navigation method for venue digital services according to claim 7, characterized in that: Determining the crowd density of each area of ​​the venue within a preset time period according to the prediction model includes: The calculation formula for crowd density is as follows: in: is the predicted crowd density at the future time t, W r is the weight matrix of the regression branch, obtained through training and learning, is the feature representation of the decoder output, Contains time decay factor α t′ Weighted historical data, including the spatial propagation coefficient A extracted by the graph convolution layer pq , b r is the bias term of the regression branch, which is learned through training.

9. The intelligent navigation method for venue digital services according to claim 1, characterized in that: The method of optimizing the initial path planning model by using the crowd density gradient map and generating a three-dimensional evaluation path integrating the shortest physical distance, the lowest congestion risk, and the highest interest matching degree by dynamically adjusting the navigation path nodes includes: The candidate path set P output by the initial path planning model is n } is mapped to the corresponding grid cells of the congestion gradient map, generating the path and congestion association matrix C∈R n×m ; Calculating the congestion risk corresponding to the candidate path set according to the path and congestion association matrix; Determine a target path planning model using the physical distance, congestion risk, and interest matching degree corresponding to the candidate path set as evaluation parameters and minimizing the expected travel time as the target parameter; The optimal path output by the target path planning model is mapped to a three-dimensional digital twin map to generate a personalized navigation route adapted to the user.

10. The intelligent navigation method for venue digital services according to claim 9, characterized in that: Calculating the congestion risk corresponding to the candidate path set according to the path and congestion association matrix includes: Where: r i The candidate path set P = {P1, P2, ..., P n } corresponding congestion risk, C ij For path P i The congestion level of the grid where the j-th node is located, ω j is the weight coefficient of the congestion level of the jth node.

Images