A method for three-dimensional modeling of air pollutant concentrations

Abstract

The embodiment of the present disclosure provides a kind of air pollutant concentration three-dimensional modeling method, belong to data processing technical field, specifically include: set cruise area and divide initial grid to cruise area;Set global cruise path strategy and local cruise path strategy;Traversal all initial grid, collect pollutant concentration at the endpoint of each initial grid;The real-time gradient module of two endpoints is calculated, and the average gradient module of each initial grid is calculated accordingly;According to average gradient module, the next initial grid is divided;According to stratified cruise path strategy, traverse all global grid, evenly collect the pollutant concentration of multiple points at each edge of each global grid and carry out PCHIP interpolation fitting;The point cloud data corresponding to global grid is obtained, and the transition layer between different density grids is established accordingly;According to point cloud data and concentration interpolation curve, construct three-dimensional space isosurface concentration field model and output. Through the scheme of the present disclosure, the modeling efficiency and accuracy are improved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a method for three-dimensional modeling of air pollutant concentration. Background Technology

[0002] Currently, with social development and industrialization, environmental pollution and damage are becoming increasingly serious, necessitating the monitoring and management of air pollution. In current pollutant concentration field modeling techniques, the traditional Marching Cubes (MC) algorithm is often combined with linear interpolation methods. However, it has significant shortcomings in dynamic adaptability and interpolation accuracy. In high-gradient regions (such as pollution source boundaries), linear interpolation exhibits severe errors. Linear interpolation assumes that pollutant concentrations change linearly within the grid, failing to accurately fit nonlinear scenarios such as air mass diffusion and turbulent mixing. This leads to smoothed interpolation results at fronts, resulting in significant deviations from actual abrupt changes in data. Therefore, existing methods have the following shortcomings:

[0003] (1) Insufficient sampling in high gradient regions: When enough sampling points are not obtained in regions where pollution concentration changes drastically, the model accuracy will decrease and it will be unable to accurately characterize the boundary of concentration change.

[0004] (2) Waste of resources in low gradient areas: In areas with gentle pollution changes, excessively dense sampling not only wastes the drone's endurance resources, but also reduces the overall mission efficiency, which is time-consuming and labor-intensive.

[0005] (3) Inaccurate nonlinear distribution interpolation: Traditional linear interpolation or simple polynomial fitting is only a rough estimate of concentration. The fitting error is large for complex air mass distribution and it is difficult to reflect the true three-dimensional morphology.

[0006] It is evident that there is an urgent need for a three-dimensional modeling method for air pollutant concentrations that offers both high modeling efficiency and accuracy. Summary of the Invention

[0007] In view of this, the present disclosure provides a three-dimensional modeling method for air pollutant concentration, which at least partially solves the problems of poor modeling efficiency and accuracy in the prior art.

[0008] This disclosure provides a method for three-dimensional modeling of air pollutant concentration, including:

[0009] Step 1: Set the cruise area and divide the cruise area into an initial grid;

[0010] Step 2: Set the global cruise path strategy and the local cruise path strategy based on the initial grid.

[0011] Step 3: Traverse all initial grids according to the global cruise path strategy and the local cruise path strategy, and collect pollutant concentrations at the endpoints of each initial grid.

[0012] Step 4: Calculate the real-time gradient modulus of the two endpoints based on the pollutant concentration, and calculate the average gradient modulus of each initial grid accordingly.

[0013] Step 5: Divide the current initial mesh into its next adjacent initial mesh based on the average gradient magnitude, until all initial meshes are divided to obtain the global mesh;

[0014] Step 6: Traverse the global grid according to the hierarchical cruise path strategy, collect pollutant concentrations at multiple points on each edge of the global grid evenly, and perform PCHIP interpolation fitting to obtain the concentration interpolation curve.

[0015] Step 7: Obtain the point cloud data corresponding to the global grid and establish a transition layer between grids of different densities accordingly;

[0016] Step 8: Construct and output a three-dimensional spatial isosurface concentration field model based on point cloud data and concentration interpolation curves.

[0017] According to a specific implementation of an embodiment of this disclosure, step 1 specifically includes:

[0018] Step 1.1: Obtain the location coordinates and altitude input by the user to establish the cruise area;

[0019] Step 1.2: Divide the cruise area into multiple initial grids according to the preset area density.

[0020] According to a specific implementation of an embodiment of this disclosure, the global training path strategy includes: setting any initial grid as the origin, controlling the UAV to start from the origin and first cruise and scan each initial grid sequentially along the y-axis until the y-axis boundary of the cruise area is reached, then shifting one initial grid along the x-axis, and then cruise and scan each initial grid sequentially along the opposite y-axis until the y-axis boundary of the cruise area is reached, then shifting one initial grid along the x-axis and repeating the cruise and scan process, and when the x-axis boundary of the cruise area is reached, shifting one initial grid along the z-axis and repeating the aforementioned repeated cruise and scan process, until all initial grids are scanned;

[0021] The local cruise path strategy includes: controlling the UAV to cruise in a spiral pattern, collecting pollutant concentrations sequentially at the endpoints of each initial grid, saving the uniform and inaccurate data of the UAV sliding along the edges between each endpoint, and continuing to cruise in a spiral pattern to the adjacent grid after all endpoints of the current grid have been cruised, until all initial grids have been scanned.

[0022] According to a specific implementation of an embodiment of this disclosure, step 4 specifically includes:

[0023] Step 4.1: Based on the spatial distance and concentration difference between two adjacent endpoints in each initial grid, the real-time gradient modulus corresponding to the edge is estimated using the first-order central difference method.

[0024] Step 4.2: Calculate the average gradient magnitude of all endpoints based on the real-time gradient magnitude of all edges.

[0025] According to a specific implementation of an embodiment of this disclosure, step 5 specifically includes:

[0026] The average gradient magnitude of the current initial grid is matched with multiple preset segmentation schemes, and a cruise step size is assigned according to the matching results. The next initial grid adjacent to it is segmented according to the cruise step size until all initial grids are segmented to obtain the global grid.

[0027] According to a specific implementation of an embodiment of this disclosure, step 6 specifically includes:

[0028] Step 6.1: Layer the cells in the global grid according to their size, and cruise according to the layered cruise path strategy;

[0029] Step 6.2: For each cell edge with sliding sampling, obtain the cell edge sampling point sequence and the concentration change information of the two endpoints;

[0030] Step 6.3: Analyze the concentration change trend and gradient information at the two ends of the cell edge. Based on the slope of the secant line at the two ends, filter and retain the points in the sampling point sequence that are consistent with the trend, remove the points with contradictory trends, and remove the points whose concentration values ​​are higher or lower than the endpoint concentration values.

[0031] Step 6.4: For each endpoint on the edge, obtain the position and value of the neighboring grid vertices or sampling points on the grid edge in the direction of this edge;

[0032] Step 6.5: Use the PCHIP interpolation method to perform curve fitting on the boundary concentration values ​​of the monotonic point set and remove invalid points to obtain the concentration interpolation function corresponding to each edge.

[0033] Step 6.6: For each cell edge that does not have sliding sampling, obtain the concentration values ​​of the two endpoints of the cell and the concentration values ​​of the nearest neighboring grid endpoints or sliding sampling points on the same edge as the two endpoints, and substitute them into the PCHIP interpolation method to obtain the concentration interpolation function corresponding to each edge.

[0034] According to a specific implementation of an embodiment of this disclosure, step 7 specifically includes:

[0035] Step 7.1: Obtain the point cloud data corresponding to the global grid based on the size of each cell in the global grid;

[0036] Step 7.2: Connect the endpoints in a cell that are not connected to adjacent cells to form a transition layer.

[0037] According to a specific implementation of an embodiment of this disclosure, step 8 specifically includes:

[0038] Step 8.1: Match all point cloud data with interpolation functions and store all endpoint information and interpolation function information;

[0039] Step 8.2: Obtain the isosurface threshold input by the user, use the MC algorithm to traverse all cell voxels one by one, and use the isosurface threshold and interpolation function to interpolate on the corresponding edges. Establish a closed three-dimensional spatial isosurface concentration field model through the points generated by interpolation.

[0040] Step 8.3: The three-dimensional spatial isosurface concentration field model is processed by a three-dimensional image library and then output for display.

[0041] The three-dimensional modeling scheme for air pollutant concentration in this embodiment includes: Step 1, setting a cruise area and dividing the cruise area into initial grids; Step 2, setting a global cruise path strategy and a local cruise path strategy based on the initial grids; Step 3, traversing all initial grids according to the global cruise path strategy and the local cruise path strategy, and collecting pollutant concentrations at the endpoints of each initial grid; Step 4, calculating the real-time gradient magnitude of the two endpoints based on the pollutant concentrations and calculating the average gradient magnitude of each initial grid accordingly; Step 5, dividing the current initial grid into its next adjacent initial grid based on the average gradient magnitude of the current initial grid, until all initial grids are divided to obtain a global grid; Step 6, traversing the global grid according to the layered cruise path strategy, uniformly collecting pollutant concentrations at multiple points along each edge of the global grid and performing PCHIP interpolation fitting to obtain a concentration interpolation curve; Step 7, acquiring point cloud data corresponding to the global grid and establishing a transition layer between grids of different densities accordingly; Step 8, constructing and outputting a three-dimensional spatial isosurface concentration field model based on the point cloud data and the concentration interpolation curve.

[0042] The beneficial effects of the embodiments of this disclosure are as follows: By combining the nonlinear interpolation technology of polynomial curve fitting and improved gradient algorithm with the solution of this disclosure, the sampling density of high gradient region is significantly improved while maintaining flight path coverage, thereby improving the accuracy of pollutant three-dimensional model. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart illustrating a three-dimensional modeling method for air pollutant concentration provided in this embodiment of the disclosure;

[0045] Figure 2 This is a schematic diagram illustrating the specific implementation process of a three-dimensional modeling method for air pollutant concentration provided in this embodiment of the disclosure;

[0046] Figure 3 This is a schematic diagram of a global cruise path strategy provided in an embodiment of the present disclosure;

[0047] Figure 4 This is a schematic diagram of a local cruise path strategy provided in an embodiment of the present disclosure;

[0048] Figure 5 This is a schematic diagram of a step size control algorithm provided in an embodiment of the present disclosure;

[0049] Figure 6 This is a schematic diagram of a segmentation effect provided in an embodiment of the present disclosure;

[0050] Figure 7 This is a schematic diagram of a layered cruise path strategy provided in an embodiment of the present disclosure;

[0051] Figure 8 A sliding edge global monotonicity PCHIP interpolation curve and interpolation position effect diagram are provided for embodiments of this disclosure;

[0052] Figure 9 A sliding edge global monotonicity PCHIP interpolation curve and interpolation position effect diagram are provided for embodiments of this disclosure;

[0053] Figure 10 A schematic diagram of a PCHIP concentration interpolation curve without sliding sampling points provided in an embodiment of this disclosure;

[0054] Figure 11 This is an overall flowchart of calculating the interpolation curves of the sliding sampling edge and the non-sliding sampling edge, provided by an embodiment of the present disclosure;

[0055] Figure 12 This is a schematic diagram of a planar transition layer connecting grids of different densities provided in an embodiment of the present disclosure;

[0056] Figure 13 This is a schematic diagram illustrating a modeling effect provided in an embodiment of this disclosure. Detailed Implementation

[0057] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0058] The following specific examples illustrate the implementation of this disclosure. Those skilled in the art can easily understand other advantages and effects of this disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this disclosure, and not all of them. This disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0059] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0060] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this disclosure. The illustrations only show the components related to this disclosure and are not drawn according to the number, shape and size of the components in actual implementation. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0061] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.

[0062] This disclosure provides a three-dimensional modeling method for air pollutant concentration, which can be applied to air monitoring processes in environmental management scenarios.

[0063] See Figure 1 This is a flowchart illustrating a three-dimensional modeling method for air pollutant concentration provided in an embodiment of this disclosure. Figure 1 and Figure 2 As shown, the method mainly includes the following steps:

[0064] Step 1: Set the cruise area and divide the cruise area into an initial grid;

[0065] In practice, a three-dimensional square cruising area can be established based on the location coordinates and altitude of two points input by the user. Let the coordinates of the two points input by the user be (x1, y1, z1) and (x2, y2, z2), where z represents the altitude.

[0066] The three-dimensional dimensions of the entire cruise area are:

[0067] X-direction length: L x =∣x2 x1∣

[0068] Y-direction length: L y =∣y2 y1∣

[0069] Z-axis height: L z =∣z2 z1∣

[0070] The user can then set the density of the patrol area (e.g., 30x30x20). The algorithm pre-divides an initial, equidistant MC grid in the area to be monitored based on the density. The drone patrols along an S-shaped or spiral path, traversing the cubic region in the air to ensure that all vertices of the global grid are covered. The user-defined grid density is... This indicates that the area is divided into three parts in the X, Y, and Z directions, respectively. , , There are several equal-sided grid cells. The side lengths of each grid cell are as follows:

[0071]

[0072] Step 2: Set the global cruise path strategy and the local cruise path strategy based on the initial grid.

[0073] In practice, the initial global cruise path strategy is as follows: Figure 3 As shown,

[0074] A. S-shaped global cruise path strategy

[0075] The drone cruise begins. The drone scans the three-dimensional airspace following a locally spiral and globally S-shaped cruise path. Each initial cube's subdivided mesh needs to be scanned and traversed using the S-shaped cruise method mentioned in step three. The global scan always begins along the y-coordinate, generating a columnar mesh of tightly packed cubes parallel to the y-axis.

[0076] B. After the scanning distance reaches the upper limit set by the user, it shifts one grid to the x-coordinate and continues to scan in the opposite direction along the y-axis to generate a columnar grid. The overall cruising process path is similar to an S-shape.

[0077] C. If the x-axis direction also reaches the upper limit, that is, after all the vertices on the current (y, x) direction extension plane have been alternately scanned, shift one grid in the z-coordinate direction and continue to generate another (y, x) extension plane adjacent to the previous plane in the previously mentioned way, until the z-axis direction also reaches the upper limit of the alternating scan.

[0078] Initial local cruise path strategy as follows Figure 4 As shown:

[0079] A. Pre-divide an initial equidistant MC mesh in the area to be monitored. The UAV cruises along a spiral path, traversing the cubic region in the air to ensure that all vertices of the global mesh are covered.

[0080] B. Whenever a grid endpoint is reached, hover for 0.5 seconds to obtain stable endpoint concentration data, which is used for endpoint constraints in subsequent nonlinear interpolation algorithms.

[0081] C. Save the uniform, imprecise data of the UAV sliding edges between each endpoint of the grid, which will be used for subsequent calculations to calculate more accurate interpolation points based on PCHIP interpolation compared to linear interpolation.

[0082] D. The local sampling cruise path for each initial size of the grid voxel is shown in the initial cruise path diagram below. The UAV flies sequentially from point 1 (starting point) to point 12 (ending point).

[0083] E. If the next grid endpoint or edge to be sampled has already been sampled, then quickly glide along the path without hovering, saving time.

[0084] F. If the cruise length in the current direction has reached the upper limit of the cruise area, the cruise direction will be adjusted according to the global cruise path strategy, and then the spiral local grid cruise will continue along the adjusted direction.

[0085] Step 3: Traverse all initial grids according to the global cruise path strategy and the local cruise path strategy, and collect pollutant concentrations at the endpoints of each initial grid.

[0086] Step 4: Calculate the real-time gradient modulus of the two endpoints based on the pollutant concentration, and calculate the average gradient modulus of each initial grid accordingly.

[0087] In practice, when the UAV slides along the grid edge to sample, it records a series of concentration values ​​of neighboring measurement points at uniform sampling intervals. By using the spatial distance and concentration difference between the previous and current recording points, the magnitude of the concentration gradient on the edge is estimated using the first-order central difference method to obtain the intensity or rate of concentration change between the two points.

[0088] First-order central difference gradient calculation formula:

[0089] Let the two points be:

[0090] Point A: Location The concentration is

[0091] Point B: Location The concentration is

[0092] Therefore, the real-time gradient magnitude (concentration change rate) between the two points is:

[0093]

[0094] in The gradient modulus is derived from the two coordinate points and the concentration.

[0095] After traversing all points of a cube mesh of initial grid size, the average gradient magnitude of these points is calculated based on their values, which guides the step size of the control drone. The formula for calculating the average gradient magnitude is shown below.

[0096]

[0097] in:

[0098] Let be the i-th real-time gradient modulus.

[0099] N is the total number of gradient magnitudes calculated from the sampled points.

[0100] This represents the average gradient magnitude corresponding to all sampling points.

[0101] Step 5: Divide the current initial mesh into its next adjacent initial mesh based on the average gradient magnitude, until all initial meshes are divided to obtain the global mesh;

[0102] In practice, the adjustment of the cruise step size and strategy is triggered when the cruise completes the last point of an initial block or a subdivided initial block.

[0103] Obtain the average gradient calculated in step four. Based on the magnitude of the average gradient, divide the initial block according to the geometric sequence 8, 64, 512, 4096… using the formula below to dynamically adjust the drone's step size. The formula for calculating the drone's cruise step size based on the average gradient magnitude is as follows:

[0104]

[0105] in:

[0106] L is the side length of the initial grid square, which can be set by the user.

[0107] , , This is the gradient threshold, which can be set by the user according to their needs. For example... Set it to 0.1. Set it to 0.5. Setting it to 1 means that the original cruise step size L is used when the gradient is less than 0.1; the cruise step size is L / 2 when the gradient is between 0.1 and 0.5; the cruise step size is L / 4 when the gradient is between 0.5 and 1; and the cruise step size is L / 8 when the gradient is greater than or equal to 1. The step size control algorithm flow is as follows: Figure 5 As shown:

[0108] The formula for calculating the number of blocks is:

[0109]

[0110] Where an is the number of squares after segmentation, and n is the segmentation number. The drone cruises at different calculated step lengths, generating grid squares of varying densities, where the side length of each small square is the calculated step length. A planar schematic diagram of the segmentation effect is shown below. Figure 6 As shown.

[0111] Step 6: Traverse the entire global grid according to the hierarchical cruise path strategy, collect pollutant concentrations at multiple points on each edge of the global grid, and perform PCHIP interpolation fitting to obtain the concentration interpolation curve.

[0112] In practice, within each initially subdivided grid, the grid is subdivided and cruised in a step size similar to an S-shaped cruise. The cruise, by default, divides the high-density grid into multiple horizontal layers from bottom to top or top to bottom. Users can also set the sampling order to be from front to back or from left to right. The specific cruise path is as follows: Figure 7 Taking the initial grid S-shaped cruise path slice map divided into 8 parts as an example, other levels of grid division are traversed in the same way. Whether the cruise traversal is from bottom to top or from top to bottom depends on whether the starting point is at the top or bottom.

[0113] For each grid edge, during the sliding process of the UAV from one hovering point located at the grid vertex to another, 10 imprecise values ​​and their positions can be continuously measured at the two endpoints on the sliding edge, and the sampling interval for each point is equal, ensuring that the positions and spacing of each sampling point are roughly uniform.

[0114] By analyzing the concentration change trend at the two ends of the cell edge, and using the slope of the secant line at the two ends as a benchmark, points in the sampling point sequence that are consistent with the trend are selected and retained, while points with contradictory trends are removed.

[0115] Trend determination: Calculate the two endpoints of the edge and secant slope This defines the global trend direction (increasing / decreasing) of that edge.

[0116] Point selection rule: For any sampling point (s) on the edge i , c i If it is related to the previous point (s) i-1 ,c i-1 Local slope of ) If the sign of the curve is consistent with the global slope k (i.e., both increasing or decreasing), it is determined to be a trend-consistent point and retained; otherwise, it is considered a trend-contradictory point and discarded, thereby ensuring that the generated PCHIP interpolation curve is globally monotonic.

[0117] For each endpoint on the edge, obtain the position and value of the neighboring grid vertices or sampling points on the grid edge in the direction of this edge.

[0118] Based on the monotonicized set of sliding edge points, the boundary concentration values ​​are curve-fitted using the PCHIP interpolation method. Taking 10 spatial points of the UAV on a straight line, plus the two endpoints and their two nearest neighbors, as an example, denoted as:

[0119]

[0120] Where (x) i ,y i ,z i Let be the spatial coordinates of the i-th sampling point, and c be the spatial coordinates of the i-th sampling point. i This represents the pollutant concentration value measured at this point.

[0121] The PCHIP interpolation function is defined as follows: It projects these 14 points in three-dimensional space onto an ordered one-dimensional parameter s along the boundary direction, constructing a monotonic sequence.

[0122]

[0123] Then in each segment [s i ,s i On [+1], the concentration interpolation function is:

[0124]

[0125] Among them, H i (s) is the cubic interpolation function for the i-th segment of PCHIP.

[0126]

[0127]

[0128] The basis functions for Hermite interpolation are:

[0129]

[0130]

[0131]

[0132]

[0133] In PCHIP interpolation, the derivative mi is automatically calculated using the local difference quotient combined with a monotonicity constraint algorithm. In PCHIP, the gradient vector between two sampling points can constrain the interpolation curve between adjacent sampling points to always remain monotonic.

[0134] Since interpolation can only be performed on edges (including edge endpoints), the curves between adjacent sampling points that do not belong to this sampling edge and the endpoints are removed from the 14 points respectively. This results in a globally monotonic PCHIP interpolation curve that starts from the endpoint of the edge and ends at the endpoint of the edge. The domain is always controlled between the two endpoints and on the two endpoints.

[0135] The following is a diagram showing the effect of globally monotonicizing the PCHIP interpolation curve and interpolation position after truncating the first and last curves: Figure 8 and Figure 9 As shown:

[0136] For grid edges that the drone does not directly slide over, sufficiently continuous discrete points cannot be obtained. Using the position and concentration data of the two points before and after the calculated point, the PCHIP interpolation curve is obtained. For each endpoint on the edge, the two nearest grid vertices or sampling points on the sampling edge in the direction of this edge are obtained. Simply substitute the coordinates, concentration values, and two adjacent points (a total of four points) on the same straight line as these two endpoints into the PCHIP interpolation method, and then extract the PCHIP concentration interpolation curve between the two endpoints (including the two endpoints) to generate the concentration interpolation curve between these two endpoints. Figure 10 This is a schematic diagram of the PCHIP concentration interpolation curve without sliding sampling points. The dashed line represents the removed curve, and the solid line represents the interpolation curve. Figure 11 The overall flowchart for calculating the interpolation curves for sliding and non-sliding sampling edges.

[0137] Step 7: Obtain the point cloud data corresponding to the global grid and establish a transition layer between grids of different densities accordingly;

[0138] In practice, the process of establishing a transition layer between grids of different densities is as follows:

[0139] A. Obtaining point cloud data

[0140] Obtain the point cloud of the global grid described above.

[0141] B. Set transition layers between meshes of different densities

[0142] For vertices in a high-density mesh with fewer than 6 connecting edges, simply connect them to the nearest unconnected point in the lower-density mesh. If there are multiple nearest points of equal distance, connect all the nearest equidistant points. A planar diagram illustrating the mesh connection method for the transition layer is shown below. Figure 12 As shown.

[0143] Step 8: Construct and output a three-dimensional spatial isosurface concentration field model based on point cloud data and concentration interpolation curves.

[0144] In practice, the process of constructing the 3D MC mesh model is as follows:

[0145] A. Integrate global variable-density grid point cloud data and interpolation functions for each grid edge.

[0146] All grid point cloud data and interpolation functions from the above steps are correlated, and all point information and edge interpolation function information are stored.

[0147] B. Calculate the positions of each isosurface point based on the interpolation function.

[0148] The user inputs the isosurface threshold, and following the steps of the MC algorithm, all square grid voxels are traversed one by one. The threshold and interpolation function are used to interpolate on the corresponding edges, and a closed three-dimensional spatial isosurface concentration field model grid is established through the points generated by interpolation.

[0149] C. Calculate and store three-dimensional isosurface models under different concentrations in advance.

[0150] The locations of the 3D model mesh data for each isosurface under commonly used thresholds are calculated in advance and stored in memory.

[0151] C. Model Output and Visualization

[0152] The generated 3D isosurface model mesh data is processed by a computer 3D image library and then displayed on the screen. The threshold can be adjusted in real-time using a slider to generate 3D isosurface models at different concentration thresholds. If the isosurface model corresponding to the threshold has been pre-calculated and stored, the stored isosurface model mesh data can be directly retrieved, and the corresponding 3D isosurface model can be quickly displayed. Figure 13 As shown.

[0153] The three-dimensional modeling method for air pollutant concentration provided in this embodiment is based on a real-time concentration gradient model. A dynamic attention mechanism adjusts the mesh density, refining the mesh in high-gradient regions and expanding it in low-gradient regions. A geometric progression partitioning strategy (8...) is employed. n (Level of encryption). Using a step size that varies linearly based on the gradient magnitude will cause sampling points to "deviate" from the grid nodes, making it impossible to directly use MC. Additional interpolation or resampling is required, increasing computational complexity and error. Therefore, this technique uses an initial grid volume and establishes a transition layer at the boundary of grids with different densities. This is because, according to the characteristics of the MC algorithm, each point in the grid needs to be connected to at least 6 edges. Therefore, a transition layer is needed between different grid densities to achieve seamless connection between grids of different densities, making the model effect presented by the isosurface continuous and avoiding the topological breakage problem caused by local encryption in traditional methods. It also makes full use of 10 sampling points with uniform sampling time intervals but inaccurate position and concentration values ​​on the UAV sliding path and the endpoints at the grid vertices to perform PCHIP interpolation calculations for each grid edge traversed by the UAV. Even if the grid edge has not been traversed, the concentration fitting curve can still be estimated using PCHIP based on the endpoints of the edge, their concentration, and gradient information, resulting in a significant reduction in interpolation error in abrupt frontal changes. There is no need to store neural network weights (traditional models require storing 10^6 levels of parameters); all interpolation coefficients are generated in real-time. Memory usage is determined solely by the currently processed grid block, with peak memory requirements reduced from 16GB to 512MB. Users can deploy multiple sampling drones to perform parallel sampling of the airspace. The grid formed by the cruise sampling points is separable and consistent with the sampling point distribution. Given a consistent sampling volume size, independently parallel sampled grids can be precisely aligned, ensuring accurate data stitching from multiple drones and integrating small sampling grids into a large sampling grid. Furthermore, the interpolation calculation for each edge is completely independent, constituting a local computation, supporting GPU / CUDA core-level parallelism, significantly reducing time consumption. Through parallel sampling by multiple drones, sliding sampling data is injected into the modeling process in real-time, shortening the assimilation cycle to 20 minutes and improving the capture rate of sudden pollution events. Furthermore, by combining data measured at hovering grid points with data measured dynamically on the sliding path, the error between the MC model and the actual environmental concentration can be effectively reduced, thus improving the overall prediction accuracy.

[0154] The method of this disclosure will be further described below with reference to a specific embodiment:

[0155] Example 1:

[0156] 3D dynamic modeling of ammonia leaks in a chemical plant. The system and algorithm have been tested in an industrial park, verifying the feasibility of the entire process, from data acquisition using a gradient-driven attention mechanism, calculation of nonlinear interpolation formulas for each edge, transmission to 3D heatmap generation and pollution response. The test included...

[0157] The experimental scenario is as follows:

[0158] The ammonia storage area in the chemical plant has a diffusion range of approximately 200m × 200m × 100m (length × width × height). The drone used was a DJI Mavic 3 Pro platform equipped with an electrochemical ammonia sensor with a range of 0-100ppm and an accuracy of ±2%. The initial grid size was set to 10m × 10m × 10m (length × width × height), and gradient thresholds G1, G2, and G3 were set to 0.1ppm / m, 0.6ppm / m, and 2ppm / m, respectively. The hovering point sampling time was 0.5 seconds to ensure the stability of the endpoint concentration data. The experimental scenario and configuration are shown in Table 1.

[0159]

[0160] The specific process is as follows:

[0161] Gradient-driven cruise: Ten drones flew along S-shaped and spiral paths, hovering for 0.5 seconds at each grid vertex to acquire endpoint concentration values. The concentration gradient magnitude between points traversed in historical records was calculated in real time. When the gradient magnitude exceeded 0.1 ppm / m, a grid densification attention mechanism was triggered, dividing the local grid into 5m × 5m sub-blocks; when the gradient magnitude exceeded 0.6 ppm / m, a grid densification attention mechanism was triggered, dividing the local grid into 1.25m × 1.25m sub-blocks. In actual monitoring, 18.7% of the area was densified due to high gradients, covering the core area of ​​the leakage front. The gradient-driven grid densification rules are shown in Table 2.

[0162]

[0163] (2) Accurate nonlinear interpolation calculation: At the boundary of the leakage front, PCHIP interpolation was performed using 10 discrete points (approximately 1 meter or 0.5 meters apart, with sampling time intervals uniformly collected at different positions on the sliding edge) uniformly collected along the sliding path of the UAV. The maximum error of the interpolation curve was 1.2 ppm, which is significantly better than the 12.5 ppm error of traditional linear interpolation. For the unsampled edge, the gradient was calculated using the central difference method of the nearest 6 points, and PCHIP interpolation was performed in combination with the endpoint gradient, which made up for the inaccuracy of traditional MC linear interpolation.

[0164] (3) Parallel acceleration optimization: The PCHIP interpolation calculation is distributed to multi-core CPUs for parallel computing, and the MC mesh modeling task is distributed to the GPU for calculation, reducing the modeling time from 30 minutes on the CPU to 2 minutes. In addition, models composed of isosurfaces with different thresholds can also be generated and cached in memory at the same time. In this way, when users view models under different isosurfaces, they do not need to recalculate the positions of each isosurface in real time. They can directly display the 3D MC model by using the interpolation positions calculated in the cache.

[0165] (4) Modeling and observation: The three-dimensional MC model after modeling can be observed. The threshold can be adjusted in real time by the slider to generate isosurfaces under different concentration thresholds, which is convenient for users to observe and make decisions.

[0166] This solution successfully identified a dumbbell-shaped concentration distribution structure at the 7-meter level at the leak front, while traditional MarchingCubes combined with linear interpolation methods could only distinguish clustered contamination areas larger than 10 meters. This structure consists of two high-concentration core areas, each approximately 3 meters in diameter, connected by a concentration transition zone about 2 meters wide, resembling a dumbbell. Hardware power consumption was kept below 25W, less than one-tenth of the 280W power consumption of a traditional Xeon E5 server cluster. The entire process from data acquisition to model output took approximately 30 minutes.

[0167] It should be understood that the various parts of this disclosure can be implemented in hardware, software, firmware, or a combination thereof.

[0168] The above description is merely a specific embodiment of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.

Claims

1. A three-dimensional modeling method for air pollutant concentration, characterized in that, include: Step 1: Set the cruise area and divide the cruise area into an initial grid; Step 2: Set the global cruise path strategy and the local cruise path strategy based on the initial grid. Step 3: Traverse all initial grids according to the global cruise path strategy and the local cruise path strategy, and collect pollutant concentrations at the endpoints of each initial grid. Step 4: Calculate the real-time gradient modulus of the two endpoints based on the pollutant concentration, and calculate the average gradient modulus of each initial grid accordingly. Step 5: Divide the current initial mesh into its next adjacent initial mesh based on the average gradient magnitude, until all initial meshes are divided to obtain the global mesh; Step 6: Traverse the global grid according to the hierarchical cruise path strategy, collect pollutant concentrations at multiple points on each edge of the global grid evenly, and perform PCHIP interpolation fitting to obtain the concentration interpolation curve. Step 6 specifically includes: Step 6.1: Layer the cells in the global grid according to their size, and cruise according to the layered cruise path strategy; Step 6.2: For each cell edge with sliding sampling, obtain the cell edge sampling point sequence and the concentration change information of the two endpoints; Step 6.3: Analyze the concentration change trend and gradient information at the two ends of the cell edge. Based on the slope of the secant line at the two ends, filter and retain the points in the sampling point sequence that are consistent with the trend, remove the points with contradictory trends, and remove the points whose concentration values ​​are higher or lower than the endpoint concentration values. Step 6.4: For each endpoint on the edge, obtain the position and value of the neighboring grid vertices or sampling points on the grid edge in the direction of this edge; Step 6.5: Use the PCHIP interpolation method to perform curve fitting on the boundary concentration values ​​of the monotonic point set and remove invalid points to obtain the concentration interpolation function corresponding to each edge. Step 6.6: For each cell edge that does not have sliding sampling, obtain the concentration values ​​of the two endpoints of the cell and the concentration values ​​of the nearest neighboring grid endpoints or sliding sampling points on the same edge as the two endpoints, and substitute them into the PCHIP interpolation method to obtain the concentration interpolation function corresponding to each edge. Step 7: Obtain the point cloud data corresponding to the global grid and establish a transition layer between grids of different densities accordingly; Step 8: Construct and output a three-dimensional spatial isosurface concentration field model based on point cloud data and concentration interpolation curves.

2. The method according to claim 1, characterized in that, Step 1 specifically includes: Step 1.1: Obtain the location coordinates and altitude input by the user to establish the cruise area; Step 1.2: Divide the cruise area into multiple initial grids according to the preset area density.

3. The method according to claim 2, characterized in that, The global cruise path strategy includes: setting any initial grid as the origin, controlling the UAV to start from the origin and first cruise and scan each initial grid sequentially along the y-axis until the y-axis boundary of the cruise area is reached, then shifting one initial grid along the x-axis, and then cruise and scan each initial grid sequentially along the opposite y-axis until the y-axis boundary of the cruise area is reached, then shifting one initial grid along the x-axis and repeating the cruise scanning process, and when the x-axis boundary of the cruise area is reached, shifting one initial grid along the z-axis and repeating the aforementioned repeated cruise scanning process, until all initial grids have been scanned; The local cruise path strategy includes: controlling the UAV to cruise in a spiral pattern, collecting pollutant concentrations sequentially at the endpoints of each initial grid, saving the uniform and inaccurate data of the UAV sliding along the edges between each endpoint, and continuing to cruise in a spiral pattern to the adjacent grid after all endpoints of the current grid have been cruised, until all initial grids have been scanned.

4. The method according to claim 3, characterized in that, Step 4 specifically includes: Step 4.1: Based on the spatial distance and concentration difference between two adjacent endpoints in each initial grid, the real-time gradient modulus corresponding to the edge is estimated using the first-order central difference method. Step 4.2: Calculate the average gradient magnitude of each initial grid based on the real-time gradient magnitudes of all edges.

5. The method according to claim 4, characterized in that, Step 5 specifically includes: The average gradient magnitude of the current initial grid is matched with multiple preset segmentation schemes, and a cruise step size is assigned according to the matching results. The next initial grid adjacent to it is segmented according to the cruise step size until all initial grids are segmented to obtain the global grid.

6. The method according to claim 5, characterized in that, Step 7 specifically includes: Step 7.1: Obtain the point cloud data corresponding to the global grid based on the size of each cell in the global grid; Step 7.2: Connect the endpoints in a cell that are not connected to adjacent cells to form a transition layer.

7. The method according to claim 6, characterized in that, Step 8 specifically includes: Step 8.1: Match all point cloud data with interpolation functions and store all endpoint information and interpolation function information; Step 8.2: Obtain the isosurface threshold input by the user, use the MC algorithm to traverse all cell voxels one by one, and use the isosurface threshold and interpolation function to interpolate on the corresponding edges. Establish a closed three-dimensional spatial isosurface concentration field model through the points generated by interpolation. Step 8.3: The three-dimensional spatial isosurface concentration field model is processed by a three-dimensional image library and then output for display.

Images