Knowledge graph-based human tile road condition analysis method

Abstract

The present application relates to the technical field of pedestrian brick monitoring, and discloses a pedestrian brick road condition analysis method based on a knowledge graph. The method first acquires spatial topology data and historical maintenance records of road infrastructure, fuses real-time road condition monitoring data and environmental parameters collected by multiple source sensors, and constructs a pedestrian brick state knowledge graph structure. Then, the method analyzes data correlation by a graph entity alignment algorithm to generate road condition evaluation features, combines real-time data and environmental parameters, and updates graph node weights by a dynamic path optimization algorithm. Subsequently, based on the updated node weights and evaluation features, the method runs an agent decision engine to output a brick area risk evaluation result. Finally, the method collects full-process data to optimize the knowledge graph structure by an incremental graph updating algorithm. The method realizes multi-source data fusion and dynamic analysis, and helps fine management of pedestrian brick road conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of pedestrian paving brick monitoring, in particular to a pedestrian paving brick road condition analysis method based on a knowledge graph. BACKGROUND

[0002] Currently, as an important part of road infrastructure, the road condition of urban pedestrian paving bricks is directly related to the safety of pedestrian traffic and the experience of urban public space use. With the acceleration of urbanization, the paving range of pedestrian paving bricks continues to expand, and the types, paving environment and use frequency of the paving bricks show significant differences, leading to increasing difficulty in road condition management. Traditional pedestrian paving brick road condition analysis relies mainly on manual inspection mode, and the staff needs to check the paving bricks on site for damage, loosening, water accumulation and other conditions, and manually record the relevant information. This method not only consumes a large amount of manpower and time cost, but also has the problems of long inspection cycle and limited coverage, and it is difficult to capture the changes in the state of the paving bricks in real time, especially in bad weather or densely populated areas, the safety and efficiency of manual inspection are further reduced.

[0003] Some areas have tried to introduce a single sensor for road condition monitoring, such as using a vibration sensor to detect the loosening degree of the paving bricks, or using a humidity sensor to monitor the water accumulation around the paving bricks. However, this kind of monitoring method can only obtain data in a single dimension, and cannot realize the effective fusion of multi-source information. For example, when the paving bricks are loose, relying only on the data of the vibration sensor cannot determine whether the loosening is caused by the foundation settlement due to the change in the humidity of the surrounding environment, and it is also difficult to correlate the repair effect of similar paving bricks in the historical maintenance records, resulting in a lack of comprehensiveness and accuracy of the road condition analysis results.

[0004] Existing road condition evaluation methods mostly use static analysis mode, which cannot dynamically adjust the evaluation indicators according to real-time monitoring data. When the city encounters extreme weather (such as heavy rain, high temperature), the damage risk of the paving bricks will increase significantly, and the static evaluation method is difficult to respond quickly to such sudden situations, which may cause the risk warning to lag. At the same time, in terms of path optimization, the existing technology usually only considers the current damage state of the paving bricks, ignoring the use frequency and future damage trend of the paving bricks in different areas, resulting in a lack of long-term applicability of the optimized path.

[0005] In recent years, knowledge graph technology has been gradually applied in the field of infrastructure management, but its application in the analysis of pedestrian paving brick road conditions still has many shortcomings. Existing road condition analysis methods based on knowledge graphs mostly focus on data storage and display, and fail to fully utilize the correlation analysis capabilities of the knowledge graph to analyze the internal relationships between spatial topology data, historical maintenance records and real-time monitoring data. They also lack effective algorithm models to dynamically update and incrementally optimize the weights of the nodes in the knowledge graph, resulting in the application value of the knowledge graph not being fully realized, and making it difficult to meet the needs of fine management of pedestrian paving brick road conditions. SUMMARY

[0006] The present application aims to provide a pedestrian tile road condition analysis method based on a knowledge graph to solve the problems raised in the background art.

[0007] To achieve the above-mentioned purpose, the present application provides a pedestrian tile road condition analysis method based on a knowledge graph, which comprises:

[0008] acquiring spatial topology data and historical maintenance records of road infrastructure; fusing real-time road condition monitoring data and environmental parameters collected by multiple source sensors to construct a pedestrian tile state knowledge graph structure;

[0009] analyzing the correlation between the spatial topology data and the historical maintenance records by a graph entity alignment algorithm to generate road condition evaluation features; combining the real-time road condition monitoring data and the environmental parameters to execute a dynamic path optimization algorithm to update the knowledge graph node weights;

[0010] based on the updated knowledge graph node weights and road condition evaluation features, running an agent decision engine to output the risk assessment results of the tile area; collecting the evaluation results and dynamic path optimization process data of the whole process to execute an incremental graph update algorithm to optimize the knowledge graph structure.

[0011] Preferably, the specific implementation of the graph entity alignment algorithm comprises:

[0012] extracting geographic coordinate features in the spatial topology data and time series features in the historical maintenance records; matching the geographic coordinate features and the time series features through a semantic similarity calculation model; establishing the correlation between the road infrastructure entities and the maintenance event entities according to the matching results; and generating road condition evaluation features based on the correlation between the geographic coordinate features and the time series features.

[0013] Preferably, the dynamic path optimization algorithm comprises:

[0014] analyzing the crack distribution features in the real-time road condition monitoring data and the temperature and humidity gradient in the environmental parameters; locating the nodes to be updated in the knowledge graph according to the crack distribution features; calculating the node weight adjustment amount based on the temperature and humidity gradient; selecting the adjacent node propagation path using a heuristic rule set; and iteratively updating the node weights along the selected propagation path until the convergence condition is met.

[0015] Preferably, the running of the agent decision engine comprises:

[0016] acquiring the node weights and road condition evaluation features of the target tile area in the knowledge graph; calculating the deviation amount of the current tile state from the preset safety threshold through a state transition model; generating a risk assessment parameter adjustment instruction according to the deviation amount; optimizing the risk assessment parameter adjustment instruction using a policy gradient algorithm; and outputting the optimized risk assessment results.

[0017] Preferably, the calculating the deviation amount of the current tile state from the preset safety threshold through the state transition model comprises:

[0018] The crack density index and the flatness index in the road condition evaluation feature are extracted, the environmental influence factor in the node weight is fused to calculate a comprehensive deterioration coefficient, and a difference between the comprehensive deterioration coefficient and the preset safety threshold is taken as the deviation amount.

[0019] Preferably, the incremental graph updating algorithm comprises:

[0020] The node weight change trend in the dynamic path optimization process data is compared with the risk assessment result to identify a structure conflict area needing to be optimized in the knowledge graph, and the entity relationship is updated through a subgraph reorganization operation based on the structure conflict area positioning result.

[0021] Preferably, the identifying the structure conflict area needing to be optimized in the knowledge graph comprises:

[0022] Abnormal fluctuation features in the node weight change trend are extracted, the abnormal fluctuation features are mapped to a space area in the knowledge graph, the logical consistency of the entity relationship in the space area is detected, and the entity relationship with logical inconsistency is marked as the structure conflict area.

[0023] Preferably, the constructing the pedestrian tile state knowledge graph structure comprises:

[0024] Three types of nodes including a road infrastructure entity, an environmental parameter entity and a maintenance event entity are defined, a spatial adjacency relationship and a time causal relationship between the nodes are established, and the node weight is initialized as a reference state value.

[0025] Preferably, the initializing the node weight comprises:

[0026] Reference maintenance cycle parameters in historical maintenance records are extracted, typical climate data in the environmental parameter entity are associated, the reference maintenance cycle parameters are corrected according to the typical climate data, and the corrected reference maintenance cycle parameters are converted into the node weight reference state value.

[0027] Preferably, the multi-source sensor comprises a crack detection sensor and a flatness detection sensor.

[0028] The crack detection sensor collects crack width features and crack density features.

[0029] The flatness detection sensor collects tile surface undulation features and joint misalignment features.

[0030] Compared with the prior art, the present application has the following beneficial effects:

[0031] By obtaining the spatial topology data of road infrastructure and historical maintenance records, and fusing real-time road condition monitoring data and environmental parameters collected by multiple sensors to construct a pedestrian paving stone state knowledge graph structure, the problem of data silos in traditional monitoring methods is broken. The effective fusion of multi-source data enables road condition analysis to go beyond single-dimensional information, providing a comprehensive picture of the actual condition of pedestrian paving stones from multiple perspectives such as spatial location, historical maintenance, real-time status, and environmental impact, making the analysis process more systematic.

[0032] At the data correlation analysis level, the method uses a graph entity alignment algorithm to mine the correlation between spatial topology data and historical maintenance records, generating road condition evaluation features. This correlation analysis capability can reveal the potential connections between different data, such as correlating the current damage state of paving stones in a certain area with the repair period and method of similar paving stones in historical maintenance, thereby more accurately determining the causes and trends of paving stone damage and providing more targeted basis for subsequent risk assessment, avoiding the judgment bias caused by data fragmentation in traditional analysis.

[0033] The introduction of a dynamic path optimization algorithm, combined with real-time road condition monitoring data and environmental parameter updates to the knowledge graph node weights, enables road condition evaluation to keep pace with changes in paving stone state. When external conditions change, such as increased risk of slippery paving stones due to rainfall or increased wear and tear on paving stones in a certain area due to sudden increases in pedestrian flow, the algorithm can quickly adjust the weights of the corresponding knowledge graph nodes, thereby updating the road condition evaluation results in a timely manner. This dynamic adjustment mechanism can effectively respond to sudden situations, ensuring the timeliness of road condition evaluation and avoiding the lag in risk warning of static evaluation methods.

[0034] The intelligent agent decision engine outputs the risk assessment results for paving stone areas based on updated knowledge graph node weights and road condition evaluation features, achieving automation and intelligentization of risk assessment using intelligent decision-making capabilities. Compared to traditional manual evaluation, the intelligent agent decision engine can quickly process massive amounts of data and complete risk assessment of large areas of pedestrian paving stones in a short time, and the evaluation process is not affected by human subjective factors, maintaining the objectivity and consistency of the evaluation results and improving the efficiency and quality of road condition management.

[0035] The application of the incremental graph update algorithm optimizes the knowledge graph structure by integrating the evaluation results of the entire process and the data of the dynamic path optimization process, enabling the knowledge graph to continuously improve over time. As analysis data continues to accumulate, the node associations in the knowledge graph become more closely related and the weight settings become more reasonable, further enhancing its support for road condition analysis. This self-optimization capability enables the method to adapt to long-term changes in urban pedestrian paving stone road conditions, meeting the management needs of different stages without the need for frequent replacement of analysis models or reconstruction of data systems, reducing long-term management costs. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 The working principle diagram of the knowledge graph-based pedestrian tile road condition analysis method of the present application;

[0037] Figure 2 The flowchart of the graph entity alignment algorithm;

[0038] Figure 3 The flowchart of the running of the agent decision engine;

[0039] Figure 4 The flowchart of the identification of the structural conflict area;

[0040] Figure 5 The flowchart of the initialization of the node weight. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0042] Please refer to Figure 1 The present application provides a knowledge graph-based pedestrian tile road condition analysis method, which comprises:

[0043] Obtain the spatial topology data and historical maintenance records of road infrastructure, which contain structured data such as geographic coordinate information, construction time, material properties, etc.; fuse real-time road condition monitoring data collected from crack detection sensors, flatness detection sensors and other devices, and environmental parameters collected by temperature and humidity sensors; construct a pedestrian tile state knowledge graph structure containing three types of nodes of road infrastructure entities, environmental parameter entities and maintenance event entities, and initialize the weights of each node as the baseline state value; analyze the correlation between the spatial topology data and the historical maintenance records by the graph entity alignment algorithm, generate road condition evaluation features containing crack density, flatness and other indicators; combine the real-time collected crack distribution characteristics and environmental temperature and humidity gradient, execute the dynamic path optimization algorithm to update the knowledge graph node weight; based on the updated node weight and road condition evaluation features, run the agent decision engine to calculate the deviation of the tile state and the safety threshold, and output the risk assessment result; finally, collect the evaluation results and optimization process data of the whole process, and continuously optimize the knowledge graph structure by the incremental graph update algorithm.

[0044] Embodiment 1: refer to Figure 2In implementing the graph entity alignment algorithm, the system first obtains the spatial topology data of the pedestrian tile area from the municipal road management database. These data are stored in GIS format, containing the precise geographic coordinates of each tile block, the boundary polygon vertex sequence, and the topological connection relationship with adjacent blocks. At the same time, historical maintenance records are retrieved from the maintenance management system, which are stored in time series form, and each record contains fields such as maintenance date, operation type, material used, construction personnel, etc. The system establishes a data cleaning pipeline to perform coordinate system unification processing on the spatial topology data, converting geographic coordinates from different sources to a unified WGS84 coordinate system; and performs time standardization processing on the historical maintenance records, converting all timestamps to ISO8601 format.

[0045] The construction of the semantic similarity calculation model adopts a double-tower neural network architecture. The spatial feature processing tower receives geographic coordinate feature input and extracts spatial pattern features, including block shape features, adjacency relationship features, etc., through a convolutional neural network. The time feature processing tower receives maintenance record input and extracts time series features, including maintenance cycle regularity, operation type distribution, etc., through a long short-term memory network. The outputs of the two feature towers are calculated through an attention mechanism in a shared vector space. The spatial feature vector and the time feature vector interact through a multi-head attention layer to calculate the cross-modal similarity score. When the spatial features of a certain tile block have a similarity greater than 0.85 with the maintenance features in a specific time period, the system automatically creates a "experienced maintenance" relationship between the block entity and the maintenance event entity.

[0046] The system maintains an association confidence matrix, which records the association strength between each pair of entities. After the initial association is established, the system continuously monitors whether subsequent data supports the association. When a new maintenance record is detected that has a higher similarity with the existing spatial features, the system automatically updates the association relationship and marks the old relationship as a historical version. All association relationships are accompanied by a time validity label to ensure that the knowledge graph can trace the relationship status in different periods.

[0047] The generation of road condition evaluation features adopts a feature fusion strategy. The geographic coordinate features in the spatial topology data are processed to extract geometric features such as block area, perimeter, shape complexity, and topological features such as adjacent block material consistency. The time series features in the historical maintenance records are analyzed to extract time series features such as the interval between the last three maintenances and the distribution of main operation types. The system combines spatial features and time features through a feature cross layer to generate evaluation features containing spatial-temporal joint information. For example, combining block shape complexity with maintenance interval to generate the "complex shape maintenance demand" feature, and combining adjacent material consistency with operation type to generate the "joint maintenance priority" feature.

[0048] The implementation of the dynamic path optimization algorithm starts with obtaining real-time data streams from the sensor network. Crack detection sensors are deployed in key monitoring areas and collect images of the tile surface every 15 minutes. The image processing unit uses a multi-scale feature extraction method. First, it identifies cracks of different sizes through Gaussian pyramid decomposition. Then, it calculates the length of the cracks using a skeleton extraction algorithm. Finally, it measures the crack width through morphological operations. The flatness detection sensor uses a laser ranging array to scan the tile surface at a 5mm interval, generating point cloud data of the elevation. The point cloud processing unit calculates the surface curvature rate of change, identifies protruding and recessed areas, and measures the height difference at the joints.

[0049] The environmental parameter monitoring system consists of distributed weather stations that record data such as ground temperature, relative humidity, and ultraviolet index every 10 minutes. The temperature and humidity gradient calculation module analyzes the data changes in the last 6 hours to calculate the temperature change rate and humidity diffusion coefficient. These parameters are connected to the material performance database to query the expansion coefficient and corrosion rate of the tile material under the current temperature and humidity conditions.

[0050] The system first identifies hotspots in the crack distribution characteristics and locates the corresponding nodes in the knowledge graph. Based on the crack width and density, it calculates the initial weight adjustment amount, with the node weights in severe crack areas increasing significantly. Then, it calculates the environmental influence factor based on the temperature and humidity gradient data and applies a correction to the initial adjustment amount. The weight update propagates along the spatial topological relationship, and adjacent nodes receive part of the weight change according to the distance decay function. The system sets a maximum propagation depth limit to prevent excessive diffusion of the update range. After each weight update, the global energy function of the knowledge graph is checked, and the iteration is stopped when the energy change tends to be stable.

[0051] The spatial proximity rule prioritizes updating nodes adjacent to the physical location, and the temporal proximity rule prioritizes updating nodes that have recently undergone maintenance. The system dynamically selects the dominant rule based on the current environmental conditions, such as prioritizing the spatial proximity rule during rainy seasons with high humidity and prioritizing the temporal proximity rule during extreme temperature weather. The rule selector is trained based on a reinforcement learning framework and automatically adjusts the rule weights based on historical update effects.

[0052] The system monitors the weight change amount for three consecutive iterations and starts convergence detection when the average change rate is below the preset threshold. It also checks the stability of key nodes in the knowledge graph to ensure that the weight update of important areas reaches equilibrium. In special weather conditions, the system extends the convergence detection period to prevent premature stopping of the update. An update report is generated after each convergence, recording the final weight values and change trajectories of each node.

[0053] Example 2: see Figure 3The operation of the agent decision engine begins with the invocation of the knowledge graph data interface. The system selects a certain commercial pedestrian street area as the evaluation target, extracts the latest weight data of all tile nodes in the area from the knowledge graph. These weight data contain dynamically calculated environmental impact factors, such as the current seasonal temperature and humidity action coefficient, material aging acceleration factor, etc. At the same time, the road condition evaluation feature set of the area is obtained, including the crack density distribution map generated by the last scan, the flatness thermal map, and other spatial feature data. The system establishes a data preprocessing channel to convert the node weight data into a state vector and the road condition feature data into an observation matrix, forming the input data stream of the decision engine.

[0054] According to the characteristics of the pedestrian tiles, four core states are defined: the perfect state corresponds to no visible cracks and the flatness meets the standard; the slight damage state corresponds to a crack width less than 2mm or a local fluctuation of no more than 3mm; the moderate damage state corresponds to a crack width of 2-5mm or a fluctuation of 3-5mm; the severe damage state corresponds to a crack width greater than 5mm or a risk of collapse. The model sets up a state transition probability matrix, where the transition probability between states is affected by environmental parameters and material properties. For example, in a frequent freeze-thaw area, the transition probability from the perfect state to the slight damage state will significantly increase.

[0055] The system first analyzes the road condition evaluation features of the target area, extracting key indicators: the crack density index is obtained by calculating the total length of cracks per unit area, and the flatness index is calculated by the root mean square value of the surface elevation data. At the same time, environmental impact factors are extracted from the node weight data, including the material expansion coefficient under the current temperature and humidity conditions, the intensity of ultraviolet radiation, etc. The calculation of the degradation coefficient uses a weighted fusion method, with the crack density index assigned a weight of 0.6 and the flatness index assigned a weight of 0.4, then multiplied by the adjustment coefficient of the environmental impact factor. The calculated result is compared with the preset safety threshold, which is dynamically adjusted according to the regional function: commercial areas use stricter standards (threshold 0.2), and residential areas use regular standards (threshold 0.35). The deviation obtained by calculation is quantified as a risk index, with a larger value indicating a greater deviation from the safe state.

[0056] The initial instruction generator contains an expert rule library, such as generating an "emergency repair" instruction when the risk index exceeds 0.5, and generating a "planned repair" instruction in the 0.3-0.5 interval. Instruction parameters include specific treatment measure priority, recommended repair time window, resource allocation suggestion, etc. The system records a case of a certain cultural plaza area: the risk index of the northwest corner tile group is detected to be 0.68, and a red alert instruction is initially generated, suggesting that the repair should be closed within 48 hours.

[0057] The system constructs a digital twin model, inputs initial instructions into the simulated environment, and predicts the state evolution path within the next 30 days. The value assessment function calculates the long-term returns of different instructions, considering factors such as maintenance costs, pedestrian safety risks, and traffic impact. In the case of Cultural Square, the algorithm finds that immediate closure for maintenance will disrupt the holiday activities, while using temporary reinforcement measures can reduce risks while avoiding the cancellation of activities. After multiple rounds of strategy iteration, the optimized instructions are adjusted to: immediately set up warning fences, complete temporary reinforcement within 72 hours, and implement full replacement after the activities are over.

[0058] The decision output module integrates the optimization results to generate the final evaluation report, which includes a three-dimensional visualization interface showing the risk levels of different areas with color gradients. Each risk point is labeled with specific parameters: for example, the entrance of a certain dining area shows an orange alert with a crack density of 0.35 / cm², a flatness deviation of 4.2mm, and a comprehensive risk index of 0.41. The disposal recommendations are presented in a hierarchical manner: the primary measure is to set up anti-slip warning signs, the suggested measure is to repair locally within 7 days, and the long-term suggestion is to monitor quarterly. The system also generates a resource scheduling scheme, automatically calculating the required materials, manpower, and equipment configuration based on the risk distribution.

[0059] After each decision is executed, the system collects actual maintenance effect data through the sensor network, compares the differences between the predicted state and the actual state, and adjusts the related parameters in the state transition matrix when it finds that the maintenance effect of a certain type of material consistently deviates from the prediction. In the case of Historical Block monitoring, the system found that the actual degradation rate of traditional stone tiles was 20% faster than the prediction model, and after three months of data accumulation, it automatically corrected the transition probability parameters for this type of material.

[0060] When detecting abnormal sensor data or conflicts in the knowledge graph nodes, the decision engine automatically switches to a conservative mode. It uses the latest reliable data copy for calculation and marks the data confidence in the report. During a heavy rain, the temperature and humidity sensors failed, and the system automatically used the regional data released by the weather station instead, and noted the data source change in the evaluation report. All decision-making processes generate traceable log files, recording input data fingerprints, intermediate values, and final output results, supporting subsequent audit analysis.

[0061] Example 3: In the implementation process of calculating the deviation amount in the state transition model, the system analyzes the east entrance area of a commercial pedestrian street. The road condition evaluation feature extraction module first calls the latest sensor scan data: in the crack distribution map collected by the crack detection sensor with an accuracy of 0.2 mm, three main cracks are marked, and the total crack length per unit area is calculated to be 83 cm / m² through image processing; the 1024 elevation points obtained by the flatness detection sensor through laser triangulation show that the maximum fluctuation is 7.3 mm, and the root mean square value is 4.1 mm. The node weight database returns the environmental influence factor set of this area, including the temperature and humidity action time records in the past 30 days (the average daily high temperature is 35°C for 18 days), and the extreme weather event count (3 times of heavy rain).

[0062] The system assigns dynamic weight coefficients to different indicators: the crack density indicator weight is set to 0.55, the flatness indicator weight is 0.35, and the remaining 10% weight is allocated to environmental factors. The environmental adjustment coefficient is calculated through temperature and humidity time series data, and the current value is 1.18, reflecting the high-temperature accelerated degradation effect. The safety threshold is set to 0.25 according to the "Urban Sidewalk Maintenance Specification", and this value is calibrated according to the regional attribute (commercial area pedestrian density coefficient 1.3). The deviation amount calculation formula is as follows:

[0063] ;

[0064] Among them: represents the deviation amount, is the safety threshold (0.25), is the crack density indicator (0.83), represents the flatness indicator (0.41), is the environmental factor adjustment value (1.18), , , correspond to the crack density weight (0.55), the flatness weight (0.35), and the environmental factor weight (0.10), respectively. Substituting the calculation gives a deviation amount of +0.37, triggering an orange risk alert.

[0065] The start of the incremental graph update algorithm is based on the analysis of historical data in this area. The system retrieves the dynamic path optimization records in the past three months and finds that the node weight change curve presents an anomaly: the weight suddenly increases by 0.22 after the heavy rain event on June 15, but the risk assessment result during the same period only shows moderate risk. Comparing the current structure of the knowledge graph, the spatial correlation strength between this node and the adjacent drainage outlet node is only 0.3, while the meteorological database shows that the drainage efficiency in this area is low. The system identifies that there is a spatial topology conflict here: the environmental sensitivity reflected by the weight change is not fully expressed in the knowledge graph.

[0066] The node weight time series is Fourier transformed to extract frequency domain features, identifying an abnormal fluctuation cluster with a peak on June 15th. The time cluster is converted to a geographic area through a spatial mapping algorithm, defining the conflict area as an 8-meter radius around the entrance. When performing entity relationship verification in this area, two contradictions are found: maintenance records show that the tiles were replaced on July 1st, but the sensor detects that the crack density in the replacement area has increased by 15%; environmental parameter nodes show that the correlation coefficient between ultraviolet intensity and crack development is only 0.21, which is lower than the threshold value of 0.35.

[0067] The subgraph reorganization operation is implemented in three stages. First, reconstruct the spatial adjacency relationship: based on the drainage network GIS data, recalculate the node spatial correlation, increase the drainage node correlation strength to 0.7, add new rainwater grate nodes and establish a bidirectional influence relationship. Second, adjust the temporal causal relationship: correct the maintenance event node attributes, add the "material adaptability" field (detection shows that the water absorption rate of the new tiles exceeds the standard), and establish a new causal chain "heavy rain event → material expansion → accelerated cracking". Finally, optimize the relationship strength parameters: lower the ultraviolet influence coefficient from 0.15 to 0.08, and add a new rainfall penetration influence coefficient of 0.23.

[0068] The system simulates a continuous heavy rain scenario for reasoning verification: when the input is 72 hours of cumulative rainfall of 150mm, the updated graph accurately predicts that three new cracks will appear in the southwest corner of the entrance area (consistent with actual monitoring); the resource scheduling module generates a preventive plan based on the new graph and deploys a tarpaulin covering measure before the heavy rain arrives. The reorganization process generates a structure change log, recording the complete operation trajectory of deleting 2 outdated associations, adding 5 entity relationships, and adjusting 13 node attribute fields.

[0069] The system assigns three labels to the conflict area of the commercial street entrance: spatial conflict label SP-04 (missing drainage correlation), temporal conflict label TM-12 (maintenance timing misalignment), and causal conflict label CS-08 (environmental factor action mechanism deviation). Each type of label comes with a solution suggestion: SP class conflict suggests supplementing infrastructure topology check, TM class conflict requires calibrating event time window parameters, and CS class conflict triggers material performance experiment verification. These classification information is stored in the knowledge graph metadata database, providing feature samples for subsequent conflict prediction.

[0070] After each subgraph reorganization, the system automatically generates a test case set: including 20 historical scenario reproduction cases and 10 extreme working condition deduction cases. In the entrance area reorganization case, the test found that the new structure has a 40% increase in response sensitivity to high temperature and high humidity working conditions, but the prediction deviation for freeze-thaw cycle working conditions has increased. The system automatically creates an optimization task item accordingly, triggering material low-temperature performance data collection instructions, and arranging a winter special monitoring plan. All test results form a structure health report, quantitatively displaying 9 indicators such as relationship path optimization rate and node coverage completeness.

[0071] Example 4: Refer to Figure 4 The structural conflict area identification process takes the waterfront park trail area as the analysis object. The system first collects the node weight change data of this area in the near future, with a time span from May to August 2023. Through the time series analysis module to process these data, using the sliding window method to calculate the daily weight fluctuation rate. Analysis found that there was an abnormal peak on July 12, the weight value increased by 0.28 compared with the previous day, which was three times the standard deviation of the normal fluctuation range. The system automatically marks this abnormal point as EP-0712 event, extracts its characteristic parameters: mutation amplitude 0.28, duration 6 hours, recovery rate 0.04 / h. The spatial mapping engine associates this event to the geographic coding unit in the knowledge graph, and determines the physical location as the third viewing platform in the D area of the waterfront trail within a radius of 15 meters.

[0072] The system loads the knowledge sub-graph of this area, including 17 road infrastructure nodes, 9 environmental parameter nodes and 5 maintenance event nodes. The logical consistency check is carried out in three dimensions: the spatial dimension verifies the adjacency relationship of the nodes, it is found that the correlation strength between the viewing platform drainage ditch node and the adjacent tile node is only 0.15, while the actual hydrological monitoring during the rainstorm shows that the water depth in this place reaches 3 cm; the time dimension tests the event sequence, the maintenance record shows that the tiles were replaced on June 20, but the crack sensor detected a crack density of 0.42 / cm² at the same location on July 5; the causal dimension analyzes the environmental association, the correlation coefficient between the ultraviolet intensity node and the crack development node is recorded as 0.18, but the material laboratory data shows that the light aging sensitivity coefficient of this stone variety should be 0.35±0.05. Refer to Table 1.

[0073] Table 1: Structural conflict analysis table of waterfront park trail

[0074]

[0075] The spatial conflict is marked as SP-07 (infrastructure association missing), the time conflict is marked as TM-09 (abnormal decay of maintenance effect), and the causal conflict is marked as CA-12 (environmental action mechanism deviation). The mixed conflict area is marked as MX-03 (multi-factor coupling anomaly), and the priority label P1 (need to be handled within 72 hours) is added. All marked information is stored in the knowledge graph meta-database, and the characteristic association with historical conflict cases is established.

[0076] The implementation of the knowledge graph construction process in this area is as follows: the road infrastructure entity node creation includes paving number PT-2021-087, attribute set records material type (granite), paving date (2021-03-15), design load (3kN / m²), etc.; the environmental parameter entity node creation includes microclimate unit MC-D03, attribute set records near-ground wind speed (annual average 2.3m / s), sunshine duration (daily average 5.7h), tidal influence coefficient (0.38), etc.; the maintenance event entity node creation includes maintenance record MR-20230620, attribute set records operation type (local replacement), use material (sesame gray granite), construction duration (3.5h), etc.

[0077] The spatial adjacency relationship is established by using the Delaunay triangulation algorithm, and the walkway area is divided into 258 triangular facets, and each facet center point is taken as a node position. The viewing platform node (N15) and the adjacent paving nodes (N16-N18) are topologically connected, and the relationship attributes are recorded: distance vector [1.2m, 0.8m, 1.5m], elevation difference [+0.3m, -0.2m, +0.1m], material joint type [straight joint, staggered joint, tongue-and-groove joint]. The time causal relationship is constructed through the event chain: 2023-06-10 rainstorm event node→2023-06-15 water accumulation monitoring node→2023-06-20 maintenance event node, and the relationship strength parameters are set: rainstorm to water accumulation strength 0.85, water accumulation to maintenance strength 0.92.

[0078] The reference parameters in the historical maintenance records of this area are extracted: the average maintenance period of granite paving is 24 months. The typical climate data in the environmental parameters are associated: annual precipitation 1800mm, extreme high temperature days 45 days, tidal immersion frequency 18 times / month. The climate correction model calculates the humidity acceleration factor 1.32, the temperature fluctuation factor 1.15, and the salt spray corrosion factor 1.28. The comprehensive correction coefficient is calculated as (1.32×0.4+1.15×0.3+1.28×0.3)=1.25, and the reference maintenance period is adjusted to 19.2 months. Through the period mapping function, the initial node weight is converted to 0.42 (0.5 is the ideal value in the interval [0,1]).

[0079] For SP-07 conflict: add drainage efficiency node, establish its bidirectional influence relationship with pavement node, and set the strength parameter to 0.78; reconstruct the spatial association of N15 node and add a 0.65 strength connection with the drainage ditch node. For TM-09 conflict: add a material compatibility field to the maintenance event node, recording the stone water absorption rate of 4.8% detected in the laboratory (standard value ≤ 3.5%); insert an intermediate node "material defect" and establish a new causal chain "maintenance operation → material defect → crack development". For CA-12 conflict: adjust the environmental action parameters, the ultraviolet influence coefficient is corrected from 0.18 to 0.31, and a new humidity retention influence coefficient of 0.27 is added. The reorganized subgraph contains 6 new nodes, 12 new relationships, and 28 adjusted attribute fields.

[0080] The system performs verification when reorganizing the simulation of continuous high humidity scenarios: input environmental parameters are relative humidity 95% for 5 days, and the updated graph outputs risk prediction: 2 cracks ≥ 3mm will appear on the east side of the hydrophilic platform (consistent with actual monitoring); the viewing platform area is expected to have a flatness degradation rate of 0.8mm / d (actual measured value 0.76mm / d). The reorganization log records the complete operation sequence: delete 3 outdated associations, add a causal path "tidal penetration → salt crystallization → surface spalling", and modify the temperature index term in the material aging formula. The final generated structure health report shows that the spatial coverage completeness has been improved to 98%, and the causal chain rationality score has reached 92 points.

[0081] Example 5: see Figure 5The node weight initialization process was implemented in the Bin Hai Landscape Avenue project. The system retrieved the historical maintenance records of the road section from the municipal maintenance management platform, and extracted the key benchmark parameters: the average maintenance interval of the sesame white granite paving stones was recorded as 26 months, and the main damage mode was edge spalling caused by salt spray corrosion, with a typical damage period of 14 months. The environmental parameter entity associated with the regional meteorological station observed data for ten years, and the typical climate characteristics included an average annual salt spray concentration of 0.28 mg / m³, a tidal immersion frequency of 22 times per month, and an average relative humidity of 92% in summer. The climate correction module initiated a multiple regression analysis, with salt spray concentration, tidal frequency, and humidity level as independent variables, and maintenance period as the dependent variable. The calculation obtained a salt spray corrosion acceleration factor of 1.42, a tidal erosion factor of 1.31, and a humidity aging factor of 1.18. The comprehensive correction coefficient was calculated using a weighted calculation mode, with a salt spray factor weight of 0.5, a tidal factor of 0.3, and a humidity factor of 0.2, and the final correction value was 1.37. The application of this coefficient adjusted the benchmark maintenance period, and the granite maintenance interval was corrected from 26 months to 19 months. The period conversion module performed a non-linear mapping, and converted the time period into node weight values through a piecewise linear function. The corrected 19-month period corresponded to an initial weight of 0.43, which was within the preset benchmark state interval of 0.4-0.6. For special climate conditions, the system added compensation parameters: an additional weight offset of 0.05 when the salt spray concentration exceeded 0.3 mg / m³, and an additional offset of 0.03 when the tidal immersion exceeded 25 times per month. The initial weight of the Landscape Avenue water-friendly section was finally set to 0.51, as it met both conditions.

[0082] The crack detection unit consists of 48 high-resolution line array cameras, each installed on a street lamp pole every 20 meters. The cameras collect surface images of the paving stones with a ground resolution of 5 mm, and automatically take pictures at 6:00-18:00 every day. The image processing pipeline includes four stages: the preprocessing stage performs light equalization and perspective correction; the feature extraction stage uses a multi-scale LoG operator to detect crack edges; the geometric analysis stage calculates crack width (accuracy 0.1 mm) and length; the density calculation stage calculates the proportion of crack pixels per unit area and generates a crack distribution heat map. The flatness detection system consists of 32 laser triangulation instrument arrays, each containing 12 laser emitters arranged in a ring. The instrument scans the surface of the paving stones with a point spacing of 2 mm, and obtains 1536 elevation points each time. The data processing process includes: point cloud registration to align multiple scanning data; surface reconstruction to generate a digital elevation model; relief analysis to calculate the surface curvature change rate; the joint detection module calculates the misalignment value by the height difference between the edges of adjacent paving stones, with an accuracy of 0.3 mm.

[0083] The backbone node accesses the municipal private network through optical fiber, and the edge node uses LoRa wireless ad hoc network. The data acquisition strategy sets a double trigger mechanism: in the regular mode, a complete data set is collected every 30 minutes; in the event trigger mode, it is reported immediately when the crack width mutation is ≥0.5mm or the flatness change is ≥2mm. A typical case occurred on the west section of Binhai Avenue: at 14:23 on July 15, the D07 sensor detected that the crack width increased from 1.2mm to 3.8mm, triggering real-time alarm. The system automatically retrieves the data of the adjacent 5 nodes to verify and confirm that it is not a false alarm, and generates an abnormal event log.

[0084] The spatial calibration module converts the coordinates of each sensor to the local coordinate system, and the registration error is controlled within ±3cm. The time synchronization service ensures that all data stamp errors are less than 50ms through the NTP protocol. The abnormal value filtering adopts the dynamic threshold method: the Hampel filter is applied to the crack width data, with a window size of 10 samples and a threshold multiple of 3.0; the moving median filter is used for flatness data, with a window length of 15 minutes. A sensor shielding case occurred on the east section of Binhai Avenue: on August 2, the C12 sensor produced abnormal fluctuation data due to leaf coverage, and the system repaired it through adjacent node data interpolation and generated a device maintenance work order.

[0085] The system selects 20 typical areas for manual review: crack detection accuracy comparison shows that the average error of sensor measured crack width is 0.12mm, and the density calculation error rate is 4.7%; in flatness detection, the surface roughness error is 0.28mm, and the joint misalignment detection error is 0.15mm. Environmental adaptability test records: the laser range finder thermal drift error is controlled within 0.4mm during high temperature period (35℃), and the camera lens fogging automatically triggers the cleaning program after tidal immersion. All verification data are stored in the knowledge graph quality file as the basis for subsequent adjustment.

[0086] When the weather station issues a typhoon warning, the system automatically activates the salt fog monitoring enhanced mode: the crack detection frequency is increased to every 10 minutes, and the flatness scanning increases the anti-wind vibration compensation algorithm. During the passage of "Haikui" typhoon, the system successfully captured the key data that the crack expansion rate of the hydrophilic pier area increased from 0.1mm / h to 0.8mm / h. When the system returns to the regular mode after the typhoon, it adjusts the climate factor weight according to the actual damage data, and the salt fog corrosion factor is adjusted from 1.42 to 1.57.

[0087] Each node has built-in self-checking program, which performs comprehensive diagnosis at 0 o'clock every day: camera module checks lens cleanliness and focusing accuracy; laser range finder calibrates reference distance; environmental sensor checks temperature and humidity readings. Diagnosis results generate device health index, and when below threshold, automatically dispatch maintenance resources. After six months of system operation, statistics show: camera module average availability rate 98.2%, laser range finder availability rate 97.6%, network communication interruption rate 0.7 times / month. Maintenance work order response time is controlled within 4 hours, and average time spent on fault repair is 1.5 hours.

[0088] It should be noted that the relational terms herein, such as first and second, and the like, are used solely to distinguish one from another entity or action without necessarily requiring or implying any actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0089] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are only by way of example and that changes can be made to the embodiments without departing from the spirit and scope of the application, which is defined in the appended claims and their equivalents.

Claims

1.A knowledge graph based analysis method for pedestrian tile road conditions, characterized in that, The method comprises the following steps: acquiring spatial topology data and historical maintenance records of road infrastructure; fusing real-time road condition monitoring data and environmental parameters collected by multiple sensors to construct a human walking brick state knowledge graph structure; analyzing the correlation between the spatial topology data and the historical maintenance records by a graph entity alignment algorithm to generate road condition evaluation features; combining the real-time road condition monitoring data and the environmental parameters to execute a dynamic path optimization algorithm to update the node weights of the knowledge graph; based on the updated node weights of the knowledge graph and the road condition evaluation features, running an agent decision engine to output the risk assessment results of the brick area; collecting the evaluation results and dynamic path optimization process data of the whole process to execute an incremental graph update algorithm to optimize the knowledge graph structure; the incremental graph update algorithm comprises: comparing the node weight change trend in the dynamic path optimization process data with the risk assessment results to identify the structure conflict area in the knowledge graph that needs to be optimized; based on the structure conflict area positioning results, performing subgraph reorganization operation to update the entity relationship. 2.The knowledge graph-based pedestrian tile road condition analysis method of claim 1, wherein, The specific implementation of the graph entity alignment algorithm comprises: extracting the geographic coordinate features in the spatial topology data and the time series features in the historical maintenance records; matching the geographic coordinate features and the time series features by a semantic similarity calculation model; establishing the correlation between the road infrastructure entities and the maintenance event entities according to the matching results; fusing the geographic coordinate features and the time series features to generate road condition evaluation features based on the correlation. 3.The knowledge graph-based pedestrian tile road condition analysis method of claim 2, wherein, The dynamic path optimization algorithm comprises: analyzing the crack distribution features in the real-time road condition monitoring data and the temperature and humidity gradient in the environmental parameters; positioning the nodes to be updated in the knowledge graph according to the crack distribution features; calculating the node weight adjustment amount based on the temperature and humidity gradient; selecting the adjacent node propagation path by using a heuristic rule set; iteratively updating the node weights along the selected propagation path until the convergence condition is met. 4.The knowledge graph-based pedestrian tile road condition analysis method of claim 1, wherein, The running of the agent decision engine comprises: acquiring the node weights and road condition evaluation features of the target brick area in the knowledge graph; calculating the deviation amount of the current brick state from the preset safety threshold by a state transition model; generating a risk assessment parameter adjustment instruction according to the deviation amount; optimizing the risk assessment parameter adjustment instruction by using a policy gradient algorithm; outputting the optimized risk assessment result. 5.The knowledge graph-based pedestrian tile road condition analysis method of claim 4, wherein, The calculation of the deviation amount of the current brick state from the preset safety threshold by the state transition model comprises: extracting the crack density index and the flatness index in the road condition evaluation features; fusing the environmental influence factors in the node weights to calculate a comprehensive degradation coefficient; taking the difference between the comprehensive degradation coefficient and the preset safety threshold as the deviation amount. 6.The knowledge graph-based pedestrian tile road condition analysis method of claim 1, wherein, The identification of the structure conflict area in the knowledge graph that needs to be optimized comprises: extracting the abnormal fluctuation features in the node weight change trend, mapping the abnormal fluctuation features to the spatial area of the knowledge graph; detecting the logical consistency of the entity relationship in the spatial area, and marking the entity relationship with logical inconsistency as a structure conflict area. 7.The knowledge graph-based pedestrian tile road condition analysis method of claim 1, wherein, The construction of the human walking brick state knowledge graph structure comprises: defining three types of nodes, i.e., road infrastructure entities, environmental parameter entities and maintenance event entities, establishing the spatial adjacency relationship and the time causal relationship between the nodes, and initializing the node weights as the baseline state values. 8.The knowledge graph-based pedestrian tile road condition analysis method of claim 7, wherein, The initialization of the node weights comprises: The method comprises the following steps: extracting a reference maintenance cycle parameter in a historical maintenance record; associating typical climate data in an environmental parameter entity; correcting the reference maintenance cycle parameter according to the typical climate data; and converting the corrected reference maintenance cycle parameter into a node weight reference state value. 9.The knowledge graph-based pedestrian tile road condition analysis method of claim 1, wherein, The multi-source sensor comprises a crack detection sensor and a flatness detection sensor. The crack detection sensor collects crack width features and crack density features. The flatness detection sensor collects tile surface undulation features and joint misalignment features.

Images