Multi-source cross-domain behavior data fusion analysis system and method based on knowledge graph

Abstract

The invention discloses a multi-source cross-domain behavior data fusion analysis system and method based on a knowledge graph, and belongs to the technical field of data processing, and the method specifically comprises the steps: deploying a gas sensor in a public vehicle in a target region, and periodically detecting the concentration value and position coordinates of pollution gas on a driving path; constructing a space-time knowledge graph structure which comprises a static building layer, a dynamic path layer and a fusion data layer; calculating the spatial relative relationship between the public vehicle and the building facades on the two sides, and mapping the gas concentration values to virtual monitoring points of the building facades in combination with the real-time wind direction and wind speed parameters to generate a concentration distribution diagram of the building facades; building gas permeation flux is calculated according to the window state, and the indoor pollution gas concentration change process of the target building is deduced; integrating indoor pollution gas concentration deduction results according to a time axis to form an air quality spatio-temporal evolution graph of the target area; according to the invention, accurate control and effective treatment of the air quality of the crowded city scene are realized.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a multi-source cross-domain behavior data fusion analysis system and method based on a knowledge graph. Background Art

[0002] Traditional air quality monitoring systems rely primarily on fixed-site deployments, exposing significant limitations when addressing complex scenarios like these. Fixed monitoring approaches not only face high equipment installation and maintenance costs, but also struggle to effectively cover hidden spaces like narrow lanes and back alleys within urban villages. This results in numerous blind spots within the monitoring range, making it difficult to ensure the integrity and timeliness of data collection. Furthermore, this type of monitoring model lacks the ability to accurately analyze the diffusion paths and penetration mechanisms of pollutants in complex building environments, making it difficult to meet the demands of refined air quality management in high-density communities.

[0003] The above situation poses many challenges to air quality monitoring and pollution source tracing in complex areas such as urban villages. There is an urgent need to explore more targeted monitoring and analysis methods to achieve accurate control and effective management of air quality in such special scenarios. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-source cross-domain behavior data fusion analysis system and method based on knowledge graph to solve the following technical problems: Traditional air quality monitoring systems mainly rely on fixed-site deployment, which exposes obvious limitations when dealing with crowded urban areas and lacks spatial adaptability and data validity.

[0005] The purpose of the present invention can be achieved through the following technical solutions: The multi-source cross-domain behavior data fusion analysis method based on knowledge graph includes the following steps: Gas sensors are deployed on public vehicles in the target area to periodically detect the concentration and location of pollutants along the driving path, generating a time-series gas concentration dataset. Construct a spatiotemporal knowledge graph structure, which includes a static building layer, a dynamic path layer, and a fused data layer. The static building layer loads the 3D building outline data and identifies the window locations. The dynamic path layer maps the driving trajectories of public vehicles. The fused data layer stores the collected time-series gas concentration datasets. According to the real-time position of the public vehicle, the current road section in the dynamic path layer is matched, and the spatial relative relationship between the public vehicle and the facades of the buildings on both sides is calculated. The gas concentration value is mapped to the virtual monitoring point of the building facade in combination with the real-time wind direction and speed parameters to generate the building facade concentration distribution map; Obtain the exterior concentration distribution map of the target building, calculate the gas permeation flux based on the window type, opening size, and opening status, and deduce the change process of the indoor pollutant gas concentration of the target building through the dynamic pressure balance equation; The deduction results of indoor pollutant gas concentrations are integrated along the time axis to form a spatiotemporal evolution map of air quality in the target area with a single building as a unit.

[0006] As a further solution of the present invention: the specific process of the gas sensor detecting polluted gas is: Parallel independent detection channels are set up to capture carbon dioxide concentration values, total volatile organic compound concentration values ​​and fine particulate matter concentration values ​​respectively. Each independent detection channel shares the air inlet and uses grating wave splitting technology to isolate spectral interference; When the vehicle is in a narrow alley, the low-speed detection mode is activated, and the sampling period of each gas component is extended to a multiple of the standard value. When the vehicle enters the main road, it switches to the high-speed detection mode and prioritizes the collection of carbon dioxide concentration values. All collected data are attached with millisecond-level timestamps and compressed by Huffman coding, and transmitted to the regional gateway via the narrowband Internet of Things.

[0007] As a further solution of the present invention: the specific process of constructing the spatiotemporal knowledge graph structure is: The digital surface model of urban village buildings is obtained from the urban planning database, the geometric topological structure of the building facades is extracted, and the spatial coordinates of the windows are annotated. The dynamic path layer integrates the fixed routes, stops, and real-time trajectories of public vehicles, discretizing the continuous trajectories into a sequence of path nodes in seconds. The fusion data layer establishes a four-dimensional spatiotemporal index matrix, in which the row vectors correspond to the path node numbers, and the column vectors are associated with the pollutant gas concentration, temperature, humidity, wind speed, and wind direction angle values. A unique spatial hash code is generated for each building, and the spatial hash code is associated with all path nodes within a set range around the building.

[0008] As a further solution of the present invention, the data collection frequency of each path node in the dynamic path layer within a set period of time in the past is counted, and building facade grids with a coverage frequency lower than the minimum daily requirement are marked as target grids. The real-time location and the route to be traveled of the adjacent public vehicles are obtained, and a supplementary monitoring path covering the target grid is generated, with the total mileage increment of the supplementary monitoring path not exceeding the upper limit of the percentage of the original route to be traveled. Path update instructions are sent to nearby public vehicles. The path update instructions are embedded in the vehicle navigation system and disguised as regular traffic congestion avoidance prompts. When the target public vehicle enters the newly added path node, the high-precision detection mode is automatically triggered.

[0009] As a further solution of the present invention: the specific process of generating the building facade concentration distribution map is: When a public vehicle enters a target road section, a collection of building facade grids within a set range around the current path node is obtained. A gas diffusion ellipse model is constructed based on the real-time wind speed and direction, with the major axis of the ellipse consistent with the wind direction and the minor axis length inversely proportional to the wind speed. The vertical distance from the center point of each facade grid to the edge of the diffusion ellipse is calculated, and the concentration distribution weight coefficient is determined based on the distance value. The measured pollutant gas concentration values ​​of vehicles are corrected for atmospheric stability. The correction parameters include temperature gradient value and solar radiation intensity value. The corrected pollutant gas concentration values ​​are distributed to each facade grid according to the weight coefficient. The grid concentration median aggregation is performed once every set time, and the building facade concentration contour line distribution map is generated by cubic spline interpolation.

[0010] As a further solution of the present invention: the specific process of calculating the gas permeation flux based on the window opening size and opening state is: Obtain the physical dimensions of the building's facade window units, including the width and height of the window openings; identify the structural type of the window units. For sliding windows, measure the ratio of the window sash travel distance to the total track length and convert this into an equivalent ventilation area reduction factor. For casement windows, a real-time image of the window's opening status is acquired, and the window's opening angle is calculated using a geometric projection transformation algorithm. When the opening angle is less than 30 degrees, a ventilation efficiency reduction factor dominated by the turbulence effect is used; when the opening angle is greater than 30 degrees, a ventilation efficiency reduction factor dominated by the laminar flow effect is used. For windows with structural damage, the effective ventilation area compensation value is calculated based on the image edge damage characteristics. The standard window area value is multiplied by the reduction coefficient and then superimposed with the compensation value to obtain the final effective ventilation area value; based on the polluted gas concentration gradient value at the center point of the window in the building facade concentration distribution map, combined with the component data of the real-time wind speed value in the normal direction of the window, the gas permeation flux value per unit time is calculated according to the principle of gas molecular diffusion dynamics.

[0011] As a further solution of the present invention: the specific process of deducing the change process of the indoor pollutant gas concentration of the target building is: Using the gas permeation flux value as the boundary input condition, a three-dimensional Cartesian coordinate grid space is established in the interior space of the building; the indoor pollutant gas background concentration field is initialized, and a deduction equation including the mass conservation equation, turbulent mixing equation, and adsorption and deposition equation is constructed; The operator splitting method is used to decompose the deduction equation into three calculation units: convection term, diffusion term, and source term. The convection term uses the upwind method to calculate the concentration transfer process in the gas flow direction. The diffusion term uses the finite difference method to solve the concentration gradient propagation process. The source term adjusts the pollution release intensity based on the abnormal gas concentration fluctuations collected at the virtual monitoring points on the facade. The concentration field synchronous update calculation of the entire grid space is performed every thirty seconds. When the Euclidean distance change rate of the concentration field between adjacent time steps is lower than the preset convergence threshold, the iterative calculation is terminated, and the pollutant gas concentration distribution matrix of each grid cell under steady-state conditions is output as the inversion result.

[0012] As a further solution of the present invention: for each building's indoor polluted gas concentration deduction results, the percentage coefficient of the total window effective ventilation area to the building's exterior area is calculated, and when the coefficient is lower than a set threshold, a boundary condition insufficient flag is generated; the directional distribution characteristics of the concentration gradient vectors of adjacent building facade monitoring points are analyzed, and the maximum angle between the vectors is calculated. If the maximum angle between the vectors exceeds ninety degrees, a data conflict flag is generated; the volume ratio of the indoor space covered by the polluted gas is calculated using gas movement path simulation technology, and when the volume ratio does not reach a preset standard value, a deduction blind spot flag is generated; When any marker generation is detected, a topological anomaly warning map is dynamically rendered on the atlas interface of the corresponding building, and the abnormal spatial coordinates and type identifier are marked using thermal coding.

[0013] As a further solution of the present invention: the repair operation after generating the mark is: When a boundary condition deficiency marker is generated, the average infiltration parameters of historical windows of similar buildings are automatically called for calculation; when a data conflict marker is generated, multiple public vehicles are dispatched to perform coordinated monitoring in the target area, and a spatial interpolation algorithm is used to reconstruct the concentration gradient field; when a deduction blind spot marker is generated, indoor partition structure constraints are added to the computational grid to create equivalent infiltration channels at non-standard wall locations; All repair operations generate versioned correction records. The corrected deduction results are re-verified for gas motion path coverage. If the verification fails, the grid resolution adaptive improvement mechanism is triggered until all marks are removed and the coverage meets the requirements.

[0014] The present invention also includes a multi-source cross-domain behavior data fusion analysis system based on a knowledge graph, which is used to implement the above-mentioned multi-source cross-domain behavior data fusion analysis method based on a knowledge graph, including: Gas sensors are used to periodically detect the concentration and location coordinates of pollutants along the driving path and generate a time-series gas concentration dataset. The knowledge graph construction module is used to construct a spatiotemporal knowledge graph structure, which includes a static building layer, a dynamic path layer, and a fused data layer. The static building layer loads the 3D outline data of the building and identifies the window locations. The dynamic path layer maps the driving trajectories of public vehicles. The fused data layer stores the collected time-series gas concentration datasets. The concentration distribution generation module is used to match the current road section in the dynamic path layer according to the real-time position of the public vehicle, calculate the spatial relative relationship between the public vehicle and the facades of the buildings on both sides, and map the gas concentration value to the virtual monitoring point on the outer surface of the building in combination with the real-time wind direction and speed parameters to generate the concentration distribution map of the building facade; The polluted gas deduction module is used to obtain the concentration distribution map of the target building's facade, calculate the gas permeation flux based on the window type, opening size and opening status, and deduce the change process of the indoor polluted gas concentration of the target building through the dynamic pressure balance equation; The spatiotemporal evolution integration module is used to integrate the deduction results of indoor pollutant gas concentrations along the time axis to form a spatiotemporal evolution map of air quality in the target area with a single building as a unit.

[0015] Beneficial effects of the present invention: The present invention deploys gas sensors on public vehicles and constructs a spatiotemporal knowledge graph to form a multi-source cross-domain data fusion and analysis solution, effectively overcoming the shortcomings of traditional fixed monitoring systems. The present invention utilizes the dynamic sampling characteristics of public vehicles to significantly reduce equipment installation and operation and maintenance costs, avoiding the high investment problem of traditional fixed sites. It matches the real-time position of vehicles through a dynamic path layer, and combines wind direction and speed parameters to accurately map gas concentration values ​​to building facades to generate concentration distribution maps, achieving effective coverage of hidden spaces such as narrow alleys and back streets in urban villages, and solving the problem of blind spots in the monitoring range. It calculates gas permeation flux based on window type, opening size, and opening state, and deduces changes in indoor pollutant gas concentration through dynamic pressure balance equations, improving the analysis of pollutant diffusion paths and penetration mechanisms in complex building environments and meeting the needs of refined governance in high-density communities. By counting the frequency of data collection at path nodes, supplementary monitoring paths are automatically generated and embedded in the on-board navigation system to ensure the integrity and timeliness of data collection. A credibility assessment and self-repair mechanism is constructed. When insufficient boundary conditions, data conflicts, or deduction blind spots are detected, historical parameters of similar buildings are automatically called, multiple vehicles are dispatched for collaborative monitoring, or the calculation grid is adjusted to ensure the accuracy of the deduction results of indoor pollutant gas concentrations, thereby achieving precise control and effective governance of air quality in crowded urban scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 It is a flow chart of the multi-source cross-domain behavior data fusion analysis method based on knowledge graph of the present invention; Figure 2 It is a module diagram of the multi-source cross-domain behavior data fusion analysis system based on knowledge graph of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] See also Figure 1 As shown, the present invention is a multi-source cross-domain behavior data fusion analysis method based on knowledge graph, which includes the following steps: First, gas sensors are deployed differently based on the structural characteristics and driving scenarios of different types of public vehicles. Sanitation vehicles integrate sensor modules on their sides or front bumpers, capturing road dust and gases from garbage stations close to the ground. Sprinkler trucks install waterproof detection equipment on their roofs to simultaneously collect volatile gases from the road surface during operation. Buses deploy sensors under their windows or on the roofs to cover areas with high passenger flow. Delivery trucks are equipped with lightweight equipment on the outside of their cargo boxes to facilitate detection in narrow alleyways. Various sensors periodically collect pollutant concentrations, including carbon dioxide, total volatile organic compounds, and fine particulate matter. The location coordinates are calculated by integrating the global satellite navigation system with base station positioning, and millisecond-level timestamps are added to generate time-series data sets. For example, when delivery trucks travel through urban villages, their sensors can collect gas data in back streets and alleys that are difficult to cover with traditional fixed stations.

[0020] Secondly, a spatiotemporal knowledge graph is constructed, consisting of a static building layer, a dynamic path layer, and a fused data layer. The static building layer obtains the three-dimensional outlines of buildings from the urban planning database, accurately annotating parameters such as facade geometry, window location, and opening dimensions to form a digital twin model of the building space. Features such as lane spacing and window orientation of urban village buildings are extracted. The dynamic path layer integrates fixed routes for public vehicles, such as sanitation vehicles, with real-time trajectories, discretizing driving trajectories into a sequence of path nodes with second-level accuracy. The fused data layer establishes a four-dimensional spatiotemporal index matrix, associating path nodes with data such as pollution concentration, temperature, humidity, and wind speed and direction. A unique spatial hash code is generated for each building, which is then associated with surrounding path nodes, achieving semantic fusion of static building and dynamic monitoring data.

[0021] The third step is to spatially map pollution data based on the real-time location of public vehicles. When a vehicle enters a target road section, the system matches the current node in the dynamic path layer, calculates the spatial relationship between the vehicle and the facades of buildings on both sides, and uses a gas diffusion model to map the concentration values ​​detected by the vehicle to virtual monitoring points on the building facades, combined with real-time wind direction and speed parameters. For example, when a sprinkler truck is operating on a main road, the exhaust pollutant concentration is distributed to the exterior windows of buildings on both sides of the road based on wind direction. When a sanitation truck passes a garbage station, garbage odor data is mapped to the facades of surrounding buildings, generating a refined concentration distribution map.

[0022] The fourth step is to deduce changes in indoor pollutant gas concentrations based on the exterior facade concentration distribution. After acquiring the target building's exterior facade concentration data, the system analyzes the window type and status. For sliding windows, the equivalent ventilation area is determined by measuring the ratio of blade travel distance to the total track length. For casement windows, image recognition technology is used to calculate the opening angle, incorporating aerodynamic principles to distinguish ventilation efficiency under turbulent and laminar flow conditions. For structurally damaged windows, the effective ventilation area is dynamically compensated based on the extent of damage. Based on these parameters, the dynamic pressure balance equation is used to deduce changes in pollutant gas concentrations within the target building, integrating factors such as external pollution concentration, indoor and outdoor pressure differential, and gas diffusion rate.

[0023] Finally, the indoor pollution simulation results for each building are integrated along a timeline to create a spatiotemporal evolution map of air quality in the target area, with each building as the unit. This map displays the distribution and changing trends of pollutants in different buildings at different times and times in real time. It can visually illustrate abnormal TVOC concentrations caused by the accumulation of renovation waste in urban villages or the accumulation of pollutants caused by ventilation system failures in commercial complexes, providing data support and decision-making tools for precise urban air quality management.

[0024] In a preferred embodiment of the present invention, the specific process of the gas sensor detecting polluted gas is: The sensor utilizes a parallel, independent detection channel design, with separate detection modules for carbon dioxide, total volatile organic compounds (TVOC), and fine particulate matter (PM2.5 / PM10). Each channel shares a common air inlet, ensuring synchronized sampling and preventing timing discrepancies from affecting detection results. A grating-wavelength splitter integrated at the front end of each channel utilizes the principle of light diffraction to separate mixed optical signals by wavelength, allowing the characteristic absorption peaks of carbon dioxide at 4.26μm and TVOC at 3.3μm to enter their respective detection units. This achieves physical isolation of spectral signals and ensures detection accuracy.

[0025] The detection mode switches dynamically according to the vehicle's driving scenario. When the vehicle enters areas with poor air mobility, such as narrow alleys or urban villages, the system activates low-speed detection mode and extends the sampling period of each gas component to a specific multiple of the standard value. This multiple is determined by the sensor model and energy consumption strategy to increase the reaction time of the gas in the detection module and ensure that the concentration fluctuation data is fully captured. For example, when a courier truck travels through the alleys of an urban village, the low-speed mode can effectively collect changes in TVOC concentrations caused by residents' cooking, renovation, or small workshop production emissions. When the vehicle enters open areas such as main roads, the system switches to high-speed detection mode and prioritizes the collection of carbon dioxide concentration values. This is because as a key indicator of traffic pollution, it is closely related to vehicle volume and driving speed. High-frequency collection can quickly identify regional pollution trends. In high-speed mode, the sensor reduces energy consumption in non-priority channels through dynamic power adjustment, balancing data collection efficiency and device endurance.

[0026] All collected data is timestamped with millisecond-level atomic clock calibration to ensure high-precision alignment in the temporal dimension. It is then compressed using Huffman coding, an algorithm that constructs an optimal prefix coding tree based on statistical symbol frequencies, reducing data redundancy and transmission overhead. The compressed data is then encrypted and transmitted via narrowband IoT to a regional gateway, which performs preliminary data cleaning and formatting to remove outliers, providing a standardized dataset for the subsequent construction of a spatiotemporal knowledge graph.

[0027] In another preferred embodiment of the present invention, the specific process of constructing the spatiotemporal knowledge graph structure is: The construction of the spatiotemporal knowledge graph begins with digital modeling of the static building layer. The system retrieves digital surface models of urban village buildings from the city planning database. This model, constructed using LiDAR scanning and Building Information Modeling (BIM) technology, contains detailed data such as the building's 3D geometric coordinates, facade materials, and window locations and dimensions. Using edge detection and feature extraction algorithms, the system automatically identifies the geometric topology of the building's facade, including the boundary contours of the wall and windows, the aspect ratio of the window openings, and the window type. To annotate the spatial coordinates of the windows, computer vision technology is used to locate the window openings in the digital surface model, generating a coordinate dataset containing window sill height, sash width, and opening direction, providing key parameters for subsequent gas infiltration analysis.

[0028] The dynamic path layer integrates fixed routes, stops, and real-time trajectory data for public vehicles such as sanitation trucks, water trucks, buses, and delivery trucks, forming a dynamic monitoring network covering the entire city. For fixed-route vehicles like buses and sanitation trucks, the system obtains the longitude and latitude sequences of the preset routes through onboard terminals and discretizes them into path nodes at second-level intervals. Each node contains information such as the stop name, estimated time of arrival, and geographic coordinates. For vehicles with real-time trajectories, such as delivery trucks, location data is transmitted back in real time via 4G / 5G communication modules. A Gaussian filter algorithm is used to remove noise to generate a continuous trajectory, which is then sampled at one-second intervals to generate dynamic path nodes. All nodes are uniformly geocoded using the UTM coordinate system to ensure spatial consistency across different vehicle trajectory data. The node sequence also includes vehicle type labels, such as "sanitation truck - cleaning route" and "delivery truck - delivery area," to facilitate the system's identification of different vehicle driving characteristics during data fusion.

[0029] The fused data layer establishes a four-dimensional spatiotemporal index matrix, whose row vectors correspond to path node numbers in the dynamic path layer, and whose column vectors integrate multi-dimensional environmental parameters such as pollutant concentration, temperature, humidity, wind speed, and wind direction. Each element in the matrix represents the multi-parameter monitoring value at a specific moment and spatial location. For example, for node number "W001" at 14:00:05 on May 10, 2024, the corresponding CO2 concentration is 850 ppm, the wind speed is 2.3 m / s, and the wind direction is 180°. To quickly associate buildings with monitoring data, the system generates a unique spatial hash code for each building. This code is based on the spatial relationship between the building's geometric center coordinates and the surrounding path nodes. A circular area with a radius of 50 meters is defined around the building, and all path node numbers within this area are mapped to a fixed-length string using a hash function, forming a bidirectional association between the building and the monitoring node. When querying historical monitoring data around a specific building, the spatial hash code can be used to quickly retrieve the associated path nodes and corresponding data, significantly improving data query efficiency.

[0030] In a preferred embodiment of this embodiment, the data collection frequency of each path node in the dynamic path layer over the past 24 hours is counted. Frequency statistics are calculated in hourly units, and the difference between the actual number of collections for each node and the preset minimum requirement is calculated. Building facade grids whose collection frequency falls below the minimum requirement for 12 consecutive hours are marked as target grids. These grids are often located in deep alleys in urban villages, on the shady sides of buildings, or between dense building clusters. Conventional vehicle routes are difficult to cover, easily forming monitoring blind spots.

[0031] Once the target grid is determined, the system uses a heuristic algorithm to generate a supplementary monitoring path based on the real-time location of nearby public vehicles and the route to be traveled. The algorithm uses "minimum mileage increment" as its optimization goal. Based on the vehicle's current route, it inserts the shortest path segment passing through the target grid to ensure that the total mileage increment does not exceed a reasonable proportion of the original route. During the path generation process, the system prioritizes vehicles traveling within 500 meters of the target grid and sends path update instructions through the API interface of the on-board navigation system. To avoid interfering with the driver's normal operation, the update instructions are disguised as regular traffic congestion avoidance prompts, such as "The road ahead is congested, it is recommended to detour to the road ahead." This prompt is displayed synchronously with the real-time traffic data of the map navigation system to ensure that the driver naturally accepts and executes the route adjustment.

[0032] When a target public vehicle enters a newly added route node, the onboard sensors automatically trigger high-precision detection mode. In this mode, the sensor's sampling frequency is increased to twice that of standard mode, all-channel synchronous detection is enabled, and an anti-interference filtering algorithm is activated to perform real-time noise reduction on the data, reducing the impact of environmental noise on the detection results. For example, when a courier vehicle receives a supplementary route instruction and detours through a deep alley in an urban village, the sensor collects data at 0.5-second intervals, accurately capturing characteristic pollutants such as formaldehyde emitted by small workshops in the alley. After this data is transmitted back in real time via the narrowband Internet of Things, the system automatically updates the concentration distribution map of the corresponding building facade, providing more accurate boundary conditions for subsequent indoor pollution simulations.

[0033] In another preferred embodiment of the present invention, the specific process of generating the building facade concentration distribution map is: When a public vehicle enters a target road section, the system demarcates a surrounding area based on the geographic coordinates of the current route node and extracts a grid of all building facades within that area. The gridding accuracy is determined by building size: high-rise buildings are divided into 5m x 5m blocks, and low-rise buildings are divided into 3m x 3m blocks, ensuring that the spatial resolution matches the monitoring requirements.

[0034] The system constructs a gas diffusion ellipse model based on real-time wind speed and direction. The ellipse is centered on the vehicle's position, with the major axis aligned with the wind direction and the minor axis inversely proportional to wind speed. This intuitively reflects the wind's impact on pollutant dispersion. To determine the concentration weights for each grid, the system calculates the vertical distance from the grid center to the edge of the ellipse. The smaller the distance, the greater the weight. Using inverse distance weighting, the system ensures that concentration decays with distance.

[0035] Before assigning measured concentration values, atmospheric stability corrections must be performed. Correction parameters include temperature gradient and solar radiation intensity. Under temperature inversion conditions, atmospheric stability is high, pollutant diffusion is weak, and concentration values ​​are corrected upward. Under strong solar radiation, turbulence is high, and diffusion capacity is strong, so concentration values ​​are corrected downward. This correction process is based on atmospheric boundary layer theory, using empirical formulas to establish a mapping between stability parameters and correction factors.

[0036] After the corrected concentration values ​​are weighted and distributed to each grid, the system performs median aggregation at set intervals to suppress outlier interference. Finally, a continuous and smooth concentration contour map is generated through cubic spline interpolation. This algorithm avoids the jagged effect of linear interpolation and clearly illustrates the gradient changes of pollutants on building facades, providing precise boundary conditions for indoor pollution penetration analysis.

[0037] In another preferred embodiment of the present invention, the specific process of calculating the gas permeation flux based on the window opening size and opening state is: The calculation of gas permeation flux closely integrates the physical characteristics and real-time status of windows, achieving accurate ventilation efficiency evaluation through multi-dimensional parameter analysis. First, the width and height data of the window unit are obtained from the static building layer to determine the standard window area, that is, the theoretical maximum ventilation area. Different methods are used to analyze the actual ventilation capacity for different window types: For sliding windows, a micro-displacement sensor installed in the window frame track monitors the fan blade travel distance in real time. This is then normalized based on the total track length to generate an equivalent ventilation area reduction factor. For example, if the fan blade travel distance accounts for 40% of the total track length, the current ventilation area is 40% of the standard area. This approach avoids treating a partially opened window as fully open, ensuring that the ventilation area calculation accurately reflects the actual degree of opening.

[0038] The casement window's open state is imaged using a corner vision sensor, and the opening angle is analyzed using a geometric projection transformation algorithm. This algorithm first identifies the edge features of the window frame and sash, then uses a Hough transform to detect straight lines and calculate the angle between them. When the opening angle is less than 30 degrees, the airflow around the window is predominantly turbulent. The system uses a reduction factor based on aerodynamic experimental fitting to account for the impact of turbulent airflow on ventilation efficiency at small opening angles. When the angle exceeds 30 degrees, the airflow becomes predominantly laminar, and a different set of reduction factors is used to reflect the smooth ventilation characteristics of large opening angles, ensuring that the ventilation efficiency assessment is more consistent with actual flow field characteristics. For windows with structural damage, such as cracked glass or deformed frames, the system uses an image edge detection algorithm to identify damage features and calculates effective ventilation area compensation based on the location and area of ​​the damaged area. For example, if there is an uncovered 10 cm x 15 cm damaged area in the lower left corner of the window, this area is directly used as compensation, representing an additional ventilation path, to avoid underestimation of infiltration flux due to structural defects. The standard area of ​​the window is multiplied by the corresponding reduction coefficient, and then the structural damage compensation value is added to obtain the effective ventilation area that reflects the current ventilation capacity. The system combines the polluted gas concentration gradient at the center of the window in the facade concentration distribution map, that is, the concentration change per unit distance, to characterize the driving force for the diffusion of pollutants into the room. And the component of the real-time wind speed in the normal direction of the window, that is, the wind speed component perpendicular to the window surface is obtained by decomposing the wind speed vector, reflecting the transport effect of external airflow on pollutants. Based on the principles of gas molecular diffusion dynamics, the ventilation capacity of the window, the external concentration gradient and the driving effect of the airflow are comprehensively considered to achieve accurate calculation of the gas penetration per unit time.

[0039] In a preferred embodiment of the present invention, the specific process of deducing the change process of the indoor pollutant gas concentration of the target building is as follows: First, a three-dimensional Cartesian coordinate grid is established within the building's interior space. The gridding accuracy is adjusted based on the building's function and monitoring requirements. Residential buildings use a relatively loose grid, such as a basic grid of 1.5 m x 1.5 m x 1.5 m, to effectively capture concentration changes in key activity areas such as bedrooms and living rooms. Commercial office buildings use a high-density grid of 1 m x 1 m x 1 m to accommodate the complex airflow and pollutant diffusion characteristics within open spaces. After the gridding is complete, the indoor pollutant background concentration field is initialized. The background value is the statistical median of historical monitoring data sets for similar buildings. For example, the default background concentration of carbon dioxide in residential buildings is the median of daily household activity periods in urban areas. This value is derived by analyzing long-term monitoring data from a large number of similar buildings in the area, ensuring that the initial conditions are close to real-life scenarios.

[0040] A deductive system consisting of three core equations was constructed. The mass conservation equation, using the grid cell as the basic calculation unit, describes the mass balance of pollutants in indoor spaces, ensuring the dynamic conservation of inflows, outflows, and source and sink terms. The turbulent mixing equation uses the k-ε two-equation model to characterize indoor air flow, capturing the enhanced effect of turbulence caused by human activity in the ventilation system on pollutant diffusion. The adsorption and deposition equation quantifies the adsorption and release of pollutants on indoor surfaces such as walls, floors, and furniture. Parameter settings are based on material properties, such as the TVOC adsorption coefficient of latex paint walls and the formaldehyde release rate of wooden furniture.

[0041] The system also loads basic pollution emission parameters from a building type feature library, which correlates building usage with typical activity scenarios. For residential buildings, the carbon dioxide emission rate is mapped to the human basal metabolic rate, which affects the resting adult male's rate. The system dynamically adjusts the source intensity based on the number of people in the room. For scenarios involving cooking, the system matches the total volatile organic compound (TVOC) emission curve to the kitchen volume. The curve's shape is fitted by analyzing measured data from household cooking fume emissions, reflecting the TVOC emission patterns of kitchens of varying sizes during cooking.

[0042] The operator splitting method is used to decompose the complex derivation equation into three independent computational units. The convection term uses the upwind scheme to calculate the concentration transfer in the gas flow direction. This scheme identifies the direction of the airflow velocity and preferentially interpolates the concentration values ​​from the upstream grid. This effectively suppresses numerical oscillations under strong convection conditions and ensures a clear depiction of the concentration front. The diffusion term uses the central difference method to solve the molecular diffusion process driven by the concentration gradient. Incorporating the physical properties of different gases, this method accurately simulates the microscopic diffusion process dominated by Brownian motion. The dynamic source term uses the collaborative sensing of virtual monitoring points on the facade and the indoor mobile sensing network to capture abnormal concentration fluctuations in real time. When a monitoring point detects a sudden increase in concentration amplitude within a short period, the system automatically matches the preset pollution source release model. If the pulse characteristics match the instantaneous peak characteristics of nicotine concentration, that is, the steep rise lasts for approximately 5 to 10 minutes, the system activates the smoking behavior release model and superimposes an additional source intensity on the corresponding room grid cell. If the pulse characteristics match the volatility of detergent, that is, the concentration rises gently for more than 30 minutes, the chemical cleaning model is invoked to dynamically calculate the release rate based on the amount of detergent used.

[0043] The full-grid concentration field is updated with a 30-second time step, and the discretized system of equations is solved using the Gauss-Seidel iteration method. When the rate of change of the Euclidean distance of the concentration field between adjacent time steps falls below a preset convergence threshold (this threshold is determined by the building's usage; residential and commercial buildings have different thresholds based on their respective accuracy requirements), the indoor pollution distribution is considered to have reached a steady state, and the calculation is terminated. The final output, a three-dimensional concentration distribution matrix, contains the pollutant concentration values ​​corresponding to each grid cell and can intuitively present a variety of pollution distribution phenomena. Examples include pollutant accumulation in poorly ventilated areas, such as near the air conditioning return vent at the corner of a wardrobe; indoor concentration gradient changes caused by external pollution penetrating through windows when the windows are open; and abnormal local concentration increases caused by sudden activities such as cooking and smoking.

[0044] In another preferred embodiment of the present invention, for each building's simulation results, a coefficient is first calculated as the percentage of the total effective window ventilation area to the building's exterior surface area. This coefficient reflects the building's natural ventilation potential. If the coefficient falls below a preset threshold, it indicates that the total window area is insufficient or the window opening state is limited, potentially resulting in insufficient boundary conditions to support accurate indoor airflow simulation. In this case, an insufficient boundary condition flag is generated.

[0045] We further analyze the directional distribution characteristics of concentration gradient vectors at adjacent building facade monitoring points and assess the consistency of the monitoring data by calculating the maximum angle between these vectors. If the maximum angle between these vectors exceeds 90 degrees, it indicates a significant conflict in the concentration trends at adjacent monitoring points. This conflict may be caused by monitoring equipment errors, sudden changes in meteorological conditions, or interference from local pollution sources. In this case, a data conflict flag will be generated.

[0046] Using gas path simulation technology, the volume ratio of indoor spaces covered by polluted gases is calculated. This metric reflects the model's ability to capture the pollutant's diffusion range. If the volume ratio falls below the preset standard, it indicates that the model may have insufficiently simulated areas, such as narrow passages or the interiors of complex structures. In these cases, a blind spot indicator is generated.

[0047] Once any marker is detected, the corresponding building's map interface dynamically renders a topological anomaly warning map. This warning map uses thermal coding to mark the spatial coordinates of the anomaly, with specific identifiers used to distinguish different types of anomalies: insufficient boundary conditions are marked with a blue circle, data conflicts are marked with a red triangle, and deduction blind spots are marked with a yellow rectangle. This visual approach allows operators to intuitively identify the location and type of anomaly, providing clear guidance for subsequent remediation.

[0048] In a preferred embodiment of the present invention, the repair operation after generating the mark is: When a boundary condition deficiency flag is generated, the average historical window infiltration parameters of similar buildings are automatically incorporated into the calculation. This parameter library, built based on long-term monitoring data from similar buildings in the region, contains average infiltration coefficients for windows of varying orientations and structures. By incorporating historical parameters, we can compensate for deficiencies in the current boundary conditions and ensure that the model accurately simulates the impact of natural ventilation on indoor air quality.

[0049] When data conflict flags are generated, multiple public vehicles are dispatched to conduct coordinated monitoring in the target area. These vehicles follow a pre-set grid route, increasing the density of monitoring points and acquiring more comprehensive pollution distribution data. A spatial interpolation algorithm is then used to reconstruct the concentration gradient field. This algorithm combines kriging interpolation with inverse distance weighted methods to generate a smooth, continuous concentration distribution surface based on the monitoring point data, effectively eliminating data conflicts and improving the accuracy of boundary conditions.

[0050] When blind spot markers are generated, interior partition structural constraints are added to the computational grid. By analyzing building plans and on-site photos, non-standard wall locations, such as temporary partitions and furniture placement, are identified and equivalent permeable channels are created within the model. These channels are assigned different resistance coefficients based on the actual structural permeability, ensuring that pollutants can diffuse effectively into previously under-simulated areas and expanding coverage of gas movement paths.

[0051] All repair operations will generate versioned correction records, which include information such as the repair time, repair type, parameter adjustment details, etc., forming a complete operation audit trail. The corrected deduction results need to be re-verified for gas movement path coverage. If the verification fails, the grid resolution adaptive improvement mechanism will be triggered. This mechanism automatically adjusts the calculation grid accuracy of the corresponding area based on the characteristics of the abnormal area, such as refining the basic grid into a higher density grid to improve the model's ability to capture complex airflow. This process is continuously iterated until all marks are eliminated and the coverage meets the requirements, ensuring that the deduction results can accurately reflect the actual distribution of indoor pollutants.

[0052] Please refer to Figure 2 As shown, the present invention also includes a multi-source cross-domain behavior data fusion analysis system based on a knowledge graph, which is used to implement the above-mentioned multi-source cross-domain behavior data fusion analysis method based on a knowledge graph, including: Gas sensors are used to periodically detect the concentration and location coordinates of pollutants along the driving path and generate a time-series gas concentration dataset. The knowledge graph construction module is used to construct a spatiotemporal knowledge graph structure, which includes a static building layer, a dynamic path layer, and a fused data layer. The static building layer loads the 3D outline data of the building and identifies the window locations. The dynamic path layer maps the driving trajectories of public vehicles. The fused data layer stores the collected time-series gas concentration datasets. The concentration distribution generation module is used to match the current road section in the dynamic path layer according to the real-time position of the public vehicle, calculate the spatial relative relationship between the public vehicle and the facades of the buildings on both sides, and map the gas concentration value to the virtual monitoring point on the outer surface of the building in combination with the real-time wind direction and speed parameters to generate the concentration distribution map of the building facade; The polluted gas deduction module is used to obtain the concentration distribution map of the target building's facade, calculate the gas permeation flux based on the window type, opening size and opening status, and deduce the change process of the indoor polluted gas concentration of the target building through the dynamic pressure balance equation; The spatiotemporal evolution integration module is used to integrate the deduction results of indoor pollutant gas concentrations along the time axis to form a spatiotemporal evolution map of air quality in the target area with a single building as a unit.

[0053] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A multi-source cross-domain behavior data fusion analysis method based on knowledge graph, characterized by: The following steps are involved: Gas sensors are deployed on public vehicles in the target area to periodically detect the concentration and location of pollutants along the driving path, generating a time-series gas concentration dataset. Construct a spatiotemporal knowledge graph structure, which includes a static building layer, a dynamic path layer, and a fused data layer. The static building layer loads the 3D building outline data and identifies the window locations. The dynamic path layer maps the driving trajectories of public vehicles. The fused data layer stores the collected time-series gas concentration datasets. According to the real-time position of the public vehicle, the current road section in the dynamic path layer is matched, and the spatial relative relationship between the public vehicle and the facades of the buildings on both sides is calculated. The gas concentration value is mapped to the virtual monitoring point of the building facade in combination with the real-time wind direction and speed parameters to generate the building facade concentration distribution map; Obtain the exterior concentration distribution map of the target building, calculate the gas permeation flux based on the window type, opening size, and opening status, and deduce the change process of the indoor pollutant gas concentration of the target building through the dynamic pressure balance equation; The deduction results of indoor pollutant gas concentrations are integrated along the time axis to form a spatiotemporal evolution map of air quality in the target area with a single building as a unit.

2. The multi-source cross-domain behavior data fusion analysis method based on knowledge graph according to claim 1 is characterized in that: The specific process of the gas sensor detecting polluted gas is as follows: Parallel independent detection channels are set up to capture carbon dioxide concentration values, total volatile organic compound concentration values ​​and fine particulate matter concentration values ​​respectively. Each independent detection channel shares the air inlet and uses grating wave splitting technology to isolate spectral interference; When the vehicle is in a narrow lane, the low-speed detection mode is activated, and the sampling period of each gas component is extended to a multiple of the standard value. When the vehicle enters the main road, it switches to the high-speed detection mode, giving priority to collecting carbon dioxide concentration values; All collected data are attached with millisecond timestamps, compressed by Huffman coding, and transmitted to the regional gateway via narrowband Internet of Things.

3. The multi-source cross-domain behavior data fusion analysis method based on knowledge graph according to claim 1 is characterized in that: The specific process of constructing the spatiotemporal knowledge graph structure is as follows: The digital surface model of urban village buildings is obtained from the urban planning database, the geometric topological structure of the building facades is extracted, and the spatial coordinates of the windows are annotated. The dynamic path layer integrates the fixed routes, stops, and real-time trajectories of public vehicles, discretizing the continuous trajectories into a sequence of path nodes in seconds. The fusion data layer establishes a four-dimensional spatiotemporal index matrix, in which the row vectors correspond to the path node numbers, and the column vectors are associated with the pollutant gas concentration, temperature, humidity, wind speed, and wind direction angle values. A unique spatial hash code is generated for each building, and the spatial hash code is associated with all path nodes within a set range around the building.

4. The multi-source cross-domain behavior data fusion analysis method based on knowledge graph according to claim 3 is characterized in that: Count the data collection frequency of each path node in the dynamic path layer within the past set time period, mark the building facade grids whose coverage frequency is lower than the minimum daily requirement, mark them as target grids, obtain the real-time location and route to be traveled of nearby public vehicles, and generate supplementary monitoring paths covering the target grids. The total mileage increment of the supplementary monitoring paths does not exceed the percentage upper limit of the original route to be traveled; Path update instructions are sent to nearby public vehicles. The path update instructions are embedded in the vehicle navigation system and disguised as regular traffic congestion avoidance prompts. When the target public vehicle enters the newly added path node, the high-precision detection mode is automatically triggered.

5. The multi-source cross-domain behavior data fusion analysis method based on knowledge graph according to claim 1 is characterized in that: The specific process of generating the building facade concentration distribution map is as follows: When a public vehicle enters the target road section, a set of building facade grids within a set range around the current path node is obtained; a gas diffusion ellipse model is constructed based on the real-time wind speed and direction, with the major axis of the ellipse consistent with the wind direction, and the minor axis length inversely proportional to the wind speed; Calculate the vertical distance from the center point of each facade grid to the edge of the diffusion ellipse, and determine the concentration distribution weight coefficient based on the distance value; The measured pollutant gas concentration values ​​of vehicles are corrected for atmospheric stability. The correction parameters include temperature gradient value and solar radiation intensity value. The corrected pollutant gas concentration values ​​are distributed to each facade grid according to the weight coefficient. The grid concentration median aggregation is performed once every set time, and the building facade concentration contour line distribution map is generated by cubic spline interpolation.

6. The multi-source cross-domain behavior data fusion analysis method based on knowledge graph according to claim 1 is characterized in that: The specific process of calculating the gas permeation flux based on the window opening size and opening state is as follows: Obtain the physical dimensions of the building's facade window units, including the width and height of the window openings; identify the structural type of the window units. For sliding windows, measure the ratio of the window sash travel distance to the total track length and convert this into an equivalent ventilation area reduction factor. For casement windows, a real-time image of the window's opening status is acquired, and the window's opening angle is calculated using a geometric projection transformation algorithm. When the opening angle is less than 30 degrees, a ventilation efficiency reduction factor dominated by the turbulence effect is used; when the opening angle is greater than 30 degrees, a ventilation efficiency reduction factor dominated by the laminar flow effect is used. For windows with structural damage, the effective ventilation area compensation value is calculated based on the image edge damage characteristics. The standard window area value is multiplied by the reduction coefficient and then superimposed with the compensation value to obtain the final effective ventilation area value; based on the polluted gas concentration gradient value at the center point of the window in the building facade concentration distribution map, combined with the component data of the real-time wind speed value in the normal direction of the window, the gas permeation flux value per unit time is calculated according to the principle of gas molecular diffusion dynamics.

7. The multi-source cross-domain behavior data fusion analysis method based on knowledge graph according to claim 6 is characterized in that: The specific process of deducing the change process of indoor pollutant gas concentration of the target building is as follows: Using the gas permeation flux value as the boundary input condition, a three-dimensional Cartesian coordinate grid space is established in the interior space of the building; the indoor pollutant gas background concentration field is initialized, and a deduction equation including the mass conservation equation, turbulent mixing equation, and adsorption and deposition equation is constructed; The operator splitting method is used to decompose the deduction equation into three calculation units: convection term, diffusion term, and source term. The convection term uses the upwind method to calculate the concentration transfer process in the gas flow direction. The diffusion term uses the finite difference method to solve the concentration gradient propagation process. The source term adjusts the pollution release intensity based on the abnormal gas concentration fluctuations collected at the virtual monitoring points on the facade. The concentration field synchronous update calculation of the entire grid space is performed every thirty seconds. When the Euclidean distance change rate of the concentration field between adjacent time steps is lower than the preset convergence threshold, the iterative calculation is terminated, and the pollutant gas concentration distribution matrix of each grid cell under steady-state conditions is output as the inversion result.

8. The multi-source cross-domain behavior data fusion analysis method based on knowledge graph according to claim 1 is characterized in that: For each building's indoor pollutant gas concentration deduction results, the percentage coefficient of the total window effective ventilation area to the building's exterior area is calculated. When the coefficient is lower than the set threshold, a boundary condition deficiency flag is generated. The directional distribution characteristics of the concentration gradient vectors at adjacent building facade monitoring points are analyzed, and the maximum angle between the vectors is calculated. If the maximum angle between the vectors exceeds 90 degrees, a data conflict flag is generated. The volume ratio of the indoor space covered by the polluted gas is calculated using gas movement path simulation technology. When the volume ratio does not reach the preset standard value, a deduction blind spot flag is generated. When any marker generation is detected, a topological anomaly warning map is dynamically rendered on the atlas interface of the corresponding building, and the abnormal spatial coordinates and type identifier are marked using thermal coding.

9. The multi-source cross-domain behavior data fusion analysis method based on knowledge graph according to claim 8 is characterized in that: The repair operations after generating the markup are: When a boundary condition deficiency flag is generated, the average infiltration parameters of historical windows of similar buildings are automatically called for calculation; when a data conflict flag is generated, multiple public vehicles are dispatched to perform coordinated monitoring in the target area, and a spatial interpolation algorithm is used to reconstruct the concentration gradient field; when a deduction blind spot flag is generated, indoor partition structure constraints are added to the computational grid to create equivalent infiltration channels at non-standard wall locations; All repair operations generate versioned correction records. The corrected deduction results are re-verified for gas motion path coverage. If the verification fails, the grid resolution adaptive improvement mechanism is triggered until all marks are removed and the coverage meets the requirements.

10. A multi-source cross-domain behavior data fusion analysis system based on a knowledge graph, used to implement the multi-source cross-domain behavior data fusion analysis method based on a knowledge graph according to any one of claims 1 to 9, characterized in that: include: Gas sensors are used to periodically detect the concentration and location coordinates of pollutants along the driving path and generate a time-series gas concentration dataset. The knowledge graph construction module is used to construct a spatiotemporal knowledge graph structure, which includes a static building layer, a dynamic path layer, and a fused data layer. The static building layer loads the 3D outline data of the building and identifies the window locations. The dynamic path layer maps the driving trajectories of public vehicles. The fused data layer stores the collected time-series gas concentration datasets. The concentration distribution generation module is used to match the current road section in the dynamic path layer according to the real-time position of the public vehicle, calculate the spatial relative relationship between the public vehicle and the facades of the buildings on both sides, and map the gas concentration value to the virtual monitoring point on the outer surface of the building in combination with the real-time wind direction and speed parameters to generate the concentration distribution map of the building facade; The polluted gas deduction module is used to obtain the concentration distribution map of the target building's facade, calculate the gas permeation flux based on the window type, opening size and opening status, and deduce the change process of the indoor polluted gas concentration of the target building through the dynamic pressure balance equation; The spatiotemporal evolution integration module is used to integrate the deduction results of indoor pollutant gas concentrations along the time axis to form a spatiotemporal evolution map of air quality in the target area with a single building as a unit.

Images