An indoor environment flow field self-adaptive iterative sampling and reconstruction method, system and medium

Abstract

The application discloses an indoor complex flow field self-adaptive iterative sampling and high-precision reconstruction method and system and a medium, and belongs to the cross field of building environment detection and spatial data processing. The method controls an autonomous mobile sampling device to perform initial sparse sampling; according to the spatial distribution variance or scale characteristics of the sampling data, a Gaussian kernel radial basis or an ordinary Kriging interpolation reconstruction algorithm is adaptively called to generate an intermediate state flow field model; then, a flow field evaluation index is used for global analysis, and a high-uncertainty information enrichment area is automatically identified; a path is automatically planned in the area and multi-dimensional space (such as horizontal two-dimensional or three-dimensional) supplementary sampling is performed. The application iterates in a closed loop of'sampling-evaluation-adaptive matching-supplementary sampling' until the model converges, effectively overcomes the defect that a traditional fixed path sampling is difficult to capture local complex disturbance, and significantly improves the mapping efficiency and high-precision reconstruction capability of the building indoor flow field.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of building environmental engineering, fluid measurement and data processing, and specifically to a method and system for obtaining high-precision indoor airflow distribution using an autonomous mobile device combined with an adaptive algorithm. Background Technology

[0002] Accurate acquisition of indoor environmental flow field information (such as wind speed, wind direction, and their spatial distribution) is crucial for HVAC system optimization, indoor air quality assessment, pollutant diffusion prediction, and thermal comfort analysis. By measuring and reconstructing indoor flow fields and the distribution of specific pollutants with high precision, core data support can be provided for building environment control, ventilation organization optimization, and indoor air quality management. Particularly concerning is the indoor particulate matter (PM) concentration field, which is highly susceptible to disturbances from thermal plumes, pedestrian movement, and complex ventilation structures, often exhibiting strong local enrichment and non-uniform distribution characteristics in space. This places extremely high demands on the local feature capture capabilities of reconstruction algorithms.

[0003] In existing technologies, for the acquisition and reconstruction of indoor environmental parameter fields, there are technical solutions that employ mobile robots equipped with sensors for spatial inspection and parameter collection. For example, publication number CN113295214A discloses a mobile robot for indoor three-dimensional multi-environmental parameter field reconstruction. This mobile robot includes a mobile chassis, a microcomputer, a laser ranging radar, a support mechanism, and wind information monitoring modules and environmental parameter monitoring modules set at different heights. It can autonomously inspect the indoor environment and collect wind speed, wind direction, and other environmental parameters at different heights to achieve dynamic updates of the indoor environmental parameter field. This type of technical solution improves the automation level and spatial coverage of indoor environmental parameter acquisition to a certain extent.

[0004] However, the aforementioned existing technologies still have the following technical shortcomings when applied to high-precision mapping of complex indoor flow fields: First, existing solutions typically use a preset path or fixed grid method for one-time spatial inspection sampling. This fixed sampling mechanism does not consider the local high gradient and eddy current phenomena around air vents, heat sources, or obstacles, resulting in redundant sampling in stable areas and insufficient sampling in key disturbed areas, thus reducing the accuracy of obtaining local flow field features. Second, existing technologies typically use a single, fixed interpolation algorithm to post-process the data after sampling. Due to the significant differences in physical characteristics between laminar and turbulent flow states, the fitting mechanism of a single algorithm cannot simultaneously take into account the overall smoothness of the flow field and the characteristics of local violent fluctuations, leading to distortion in the reconstructed model. Third, existing solutions adopt a unidirectional linear processing flow of "sampling-reconstruction" without establishing a feedback mechanism based on the spatial characteristics of the flow field. They cannot automatically identify high uncertainty areas and perform targeted supplementary sampling during the reconstruction process, thus limiting the overall accuracy and mapping efficiency of complex indoor environment flow field reconstruction. Summary of the Invention

[0005] The purpose of this invention is to solve the above-mentioned technical problems and provide an adaptive iterative sampling and reconstruction method for indoor environmental flow fields. This invention is achieved through the following technical solution:

[0006] An adaptive iterative sampling and reconstruction method for indoor environmental flow fields, the method comprising the following steps:

[0007] Step S1: Control the mobile sampling device to perform the first round of sampling in the indoor space to be measured according to the initial planned path to obtain the initial discrete measurement point data set;

[0008] Step S2: Obtain the flow field scene features of the indoor space to be tested, and adaptively call the matching flow field reconstruction algorithm according to the flow field scene features; construct a continuous flow field model of the indoor space based on the current discrete measurement point data set and the called flow field reconstruction algorithm, and generate intermediate flow field data;

[0009] Step S3: Perform a global analysis of the intermediate flow field data using flow field evaluation indices, calculate the evaluation index values ​​of each spatial grid node, and identify continuous regions whose evaluation index values ​​meet preset judgment conditions as information-rich regions that need to be supplemented with additional sampling; the preset judgment conditions include at least one of threshold judgment and percentile judgment.

[0010] Step S4: Based on the spatial distribution characteristics of the information-rich area, automatically plan the coordinates and path of the new sampling point, and drive the mobile sampling device to the new sampling point to collect supplementary data;

[0011] Step S5: Incorporate the supplementary data into the discrete measurement point data set, and repeat steps S2 to S4 until the preset convergence condition is met, and output the final indoor flow field reconstruction result.

[0012] Furthermore, the flow field reconstruction algorithm includes at least one of geostatistical interpolation algorithm or kernel function-based interpolation method; the geostatistical interpolation algorithm includes Ordinary Kriging interpolation algorithm; the kernel function-based interpolation method includes Radial Basis Function interpolation method.

[0013] Further, in step S2, acquiring the flow field scene features of the indoor space under test includes any of the following methods:

[0014] Method 1: Based on the initial discrete measurement point data set, calculate the spatial distribution variance or Reynolds number estimate of the data to determine the flow field type characteristics; wherein, the Reynolds number estimate Re est according to The calculations are performed, where ρ and μ are the indoor air density and dynamic viscosity constants determined based on the average ambient temperature or real-time sensor measurements, and v... ref L is the time average of the velocity vector modulus at the initial discrete measurement points, or the arithmetic mean of the time averages at all measurement points. ref The equivalent hydraulic diameter or minimum geometric side length of the space to be measured, or the average nearest neighbor distance of the current discrete measuring point;

[0015] Method 2: Obtain the boundary geometric information of the indoor space to be measured by an environmental perception sensor mounted on a mobile sampling device to determine the spatial scale parameters. The environmental perception sensor includes a lidar or a depth camera. The boundary geometric information is obtained through real-time localization and mapping and boundary extraction.

[0016] Furthermore, the specific strategy for adaptively invoking the matching flow field reconstruction algorithm based on the flow field scene characteristics includes:

[0017] When the flow field type characteristic is determined to be laminar-dominated, i.e., low variance or low Reynolds number, or when the spatial scale parameter is less than a preset spatial scale threshold L, th At that time, the Gaussian kernel radial basis (RBF) interpolation algorithm with higher smoothness is invoked;

[0018] When the flow field type characteristic is determined to be turbulence-dominated, i.e., high variance or high Reynolds number, or when the spatial scale parameter is greater than or equal to a preset spatial scale threshold L, th At that time, the Ordinary Kriging interpolation algorithm with the ability to capture local fluctuations is invoked;

[0019] Wherein, the L th Press Lth =βL min Confirmed, L min β is the minimum geometric feature length of the indoor space to be measured, and β is a scaling factor of 0.05 to 0.20.

[0020] Furthermore, in step S3, the flow field evaluation index includes one or any combination of velocity gradient modulus, local velocity variance, flow field information entropy, and vorticity intensity.

[0021] The specific steps for identifying information-rich regions are as follows: calculate the evaluation index values ​​of all grid nodes, construct an evaluation index distribution map, extract continuous grid regions whose index values ​​are at the top of the statistical distribution by a predetermined percentage as candidate information-rich regions, and determine the candidate regions with an area of ​​not less than N grid units as information-rich regions.

[0022] Wherein, the preset percentage P is 5% to 20%, the continuous grid region is a region formed by connecting four or eight neighboring regions, and N≥4.

[0023] Furthermore, the convergence condition in step S5 is:

[0024] Calculate the root mean square error (RMSE) of the velocity vector modulus difference between intermediate flow field data generated in adjacent iterations at corresponding spatial grid points in the global domain, wherein the adjacent iterations include the k-th iteration and the (k-1)-th iteration;

[0025] When the RMSE is less than a preset accuracy threshold ∈, it is determined that the convergence condition is met, wherein the preset accuracy threshold ∈ is an absolute error threshold or a relative error percentage threshold.

[0026] Further, in step S4, the driving of the mobile sampling device to the new sampling point to collect supplementary data includes:

[0027] Plan the coordinates of the newly added sampling points, the coordinates including horizontal coordinates (x, y) and vertical height coordinates z;

[0028] The autonomous mobile chassis of the mobile sampling device is controlled to move horizontally to the horizontal coordinate (x, y), and the vertical lifting mechanism mounted on the chassis is controlled to adjust the sampling height to the vertical height coordinate z, so as to obtain supplementary flow field data in three-dimensional space.

[0029] Furthermore, in step S2, when the flow field reconstruction algorithm uses radial basis function interpolation, the shape parameter c of the kernel function is dynamically adjusted based on the average pairwise distance or the average nearest neighbor distance (average spatial distance) of the current discrete measurement point data set, where c = α·d avg d avgα is the average pairwise distance or the average nearest neighbor distance of the current discrete measurement points, and α is a scaling factor of 0.5 to 2.0.

[0030] This invention also provides an adaptive iterative sampling and reconstruction system for indoor environmental flow fields, characterized in that it includes:

[0031] A mobile sampling device and a sensor assembly mounted on the mobile sampling device; the sensor assembly includes at least a wind speed sensor and a lidar or depth camera for acquiring boundary geometry information;

[0032] Memory, used to store computer programs;

[0033] A processor is communicatively connected to the mobile sampling device and the sensor assembly; when executing the computer program, the processor implements the steps of the method as described in any one of claims 1 to 8.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] 1. Introducing flow field information entropy as an active sampling feedback indicator can identify high uncertainty regions in complex indoor flow fields. It is especially suitable for accurately capturing the high gradient enrichment phenomenon of small suspended particles such as particulate matter (PM) in airflow blind zones or local eddies, thereby improving the pertinence of supplementary sampling and the reconstruction accuracy of complex concentration fields / flow fields.

[0036] 2. Adaptively selecting the flow field reconstruction algorithm based on the characteristics of the indoor flow field scene can improve the accuracy of the flow field model in representing the real physical field distribution.

[0037] 3. By linking the mobile chassis with the lifting mechanism, automated mapping of three-dimensional space can be achieved, which helps to improve sampling efficiency and reduce human interference. Attached Figure Description

[0038] To more clearly illustrate the technical solution of the present invention, the accompanying drawings are briefly described below.

[0039] Figure 1 This is an overall flowchart of the method of the present invention.

[0040] Figure 2 This is a schematic diagram of the module structure of the system of the present invention.

[0041] Figure 3 This is a schematic diagram of the adaptive flow field sampling principle provided in an embodiment of the present invention; wherein, Figure 3 (a) is a schematic diagram of the initial measurement point distribution. Figure 3 (b) is a schematic diagram of the adaptive supplementary sampling point distribution. Figure 3In (b), the area marked with a dashed line is the information-rich area identified after flow field reconstruction and analysis.

[0042] Figure 4 This is a schematic diagram of the spatial distribution of flow field entropy in a complex workshop provided by an embodiment of the present invention. In the figure, 1 is a half-height partition, 2 is a high-entropy eddy current region, 3 is a heat source device, and 4 is a high-entropy thermal plume region. Detailed Implementation

[0043] The invention will be further described below with reference to the accompanying drawings. It should be noted that the "indoor space to be measured" and "continuous flow field model" described in this invention are not limited to a specific spatial dimension. Depending on the actual measurement requirements and hardware configuration, the flow field reconstruction can be either a two-dimensional planar flow field sampling and reconstruction performed on a horizontal plane or a specific cross-section at a specific height, or a full-domain three-dimensional flow field sampling and reconstruction combining information from different vertical heights. The spatial coordinates, grid nodes, and distance formulas involved in this invention are all adaptively compatible with both two-dimensional and three-dimensional vector calculations. The following embodiments mainly use typical three-dimensional and two-dimensional planar scenarios as examples for illustration, but should not be used to limit the scope of protection of this invention.

[0044] Example 1

[0045] This embodiment is applied to a constant temperature and humidity laboratory with dimensions of 10m × 10m × 3m. System hardware configuration: An omnidirectional mobile vehicle (AGV) with Mecanum wheels is used, with a 2m travel electric scissor lift mounted on the chassis. A 3D ultrasonic anemometer is installed on the top of the lift, and a LiDAR or depth camera is installed on the top of the AGV for SLAM mapping, extracting boundary geometric information of walls, doors, windows, and obstacles to obtain spatial scale parameters.

[0046] During the actual deployment or system accuracy calibration phase, this method uses standard computational fluid dynamics (CFD) software such as Airpak or Fluent to perform three-dimensional numerical simulations of the indoor space under test, extracting high-resolution benchmark flow field data for the verification and optimization of the reconstruction algorithm. It should be noted that, for the high-precision continuous flow field and particulate matter concentration field reconstruction that this invention focuses on, this system excludes FDS software, which is based on fire large eddy simulation, from sampling simulation verification to ensure the rigor and reliability of the heat and mass transfer model calculations under low-speed ventilation conditions.

[0047] Specific steps:

[0048] Step S1: Initial sampling.

[0049] The system divides the indoor plane into a sparse grid of 2m×2m, controls the AGV to stay at the center of each grid at a height of 1.5m for 30 seconds, collects the average wind speed vector, and obtains the initial dataset D0.

[0050] Step S2: Algorithm adaptation and flow field reconstruction.

[0051] The system calculates the variance σ of the spatial velocity modulus based on the initial discrete measurement point dataset D0 (containing N sampling points). 2 The variance is defined as: in Let V i v is the average velocity vector at the i-th sampling point within the time window; i Let v be the magnitude of the velocity vector. i =||V i ||; σ is the arithmetic mean of the velocity moduli at all sampling points; N is the total number of initial discrete sampling points. 2 If the airflow is less than the preset threshold (indicating smooth airflow and laminar flow dominance), the system adaptively selects the Gaussian kernel radial basis interpolation algorithm (RBF) for full-field reconstruction.

[0052] Algorithm model: For any point x in space, the formula for reconstructing its velocity V(x) is:

[0053]

[0054] Where x is the spatial location vector of the grid node to be reconstructed; x i w is the spatial location vector of the i-th known sampling point; i The weight to be determined; Let r be the Gaussian kernel function, where r = ||x - x i The kernel function parameter c is adaptively set according to the measurement point spacing as follows:

[0055] c = α·d avg

[0056] in The nearest neighbor average distance is given by α, which is a dimensionless proportionality coefficient that satisfies 0.5≤α≤2.0. The distance unit is consistent with the coordinate unit of the measuring point (m).

[0057] Step S3: Flow field analysis.

[0058] Selecting the velocity gradient modulus As an evaluation indicator, the determination of information-rich regions can employ a fixed threshold method or a percentile method. Both the fixed threshold method and the percentile method are specific implementations of the preset determination condition described in claim 1, with the percentile method corresponding to the determination method described in claim 5. In this embodiment, the fixed threshold method is used, setting the threshold T to 1.5 times the average gradient value across the entire field. Regions with gradient values ​​greater than T are identified as "information-rich regions" (mainly concentrated below the air outlet). In other embodiments, the percentile determination method described in claim 5 can also be used.

[0059] Step S4: Supplementary sampling.

[0060] Within the information-rich area, the K-means algorithm is used to plan several (e.g., 10 to 15) new sampling points, and the driving device is moved to the area and the elevator height is adjusted for fine data collection.

[0061] The specific process for planning new sampling points in information-rich areas is as follows:

[0062] (1) Constructing the input feature vector: Extract the spatial coordinates (x, y, z) and corresponding velocity gradient modulus values ​​of all grid nodes within the information enrichment region. To eliminate the influence of dimensions, normalize the spatial coordinates and gradient values, and construct a multidimensional feature vector v for each grid node. i =[x′ i y′ i , z′ i ,w·G′ i ], where w is the weight coefficient of the gradient index, G′ i This is the normalized velocity gradient modulus value;

[0063] (2) Perform K-means clustering: According to the system's limit on the number of supplementary sampling points (e.g., K = 10 to 15), Euclidean distance is used as the similarity metric. The above multidimensional feature vectors are input into the K-means algorithm for iterative clustering until the cluster centers no longer shift significantly, and the information-rich region is divided into K feature clusters.

[0064] (3) Generate physical coordinates of sampling points: Extract the cluster centers of the K feature clusters that have finally converged. Since these cluster centers are calculated in a multidimensional space containing the velocity gradient, the system inversely maps them back to three-dimensional physical space, discarding the velocity gradient dimension and retaining only the spatial coordinate dimension (X). k ,Y k Z k The calculated target location is used as the new sampling point target location for the mobile sampling device. If there is a physical obstacle at the calculated target location, the nearest barrier-free safety grid point is found in the neighborhood of its preset radius and replaced.

[0065] like Figure 3 The diagram shown illustrates the adaptive flow field sampling principle provided in an embodiment of the present invention. Figure 3 (a) shows the distribution of the initial measurement point data obtained in step S1. The dots (“o”) in the figure represent the initial uniformly distributed sampling measurement points. Figure 3 (b) illustrates the gradient-based adaptive supplementary sampling distribution state in step S4, with the cross ("x") in the figure representing newly added sampling points during the adaptive iteration process. Figure 3 In (b), the area marked with a dashed line is the information-rich area (i.e., the high gradient region) identified after flow field reconstruction and analysis. It can be seen that a large number of newly added measurement points are concentrated in this area.

[0066] Step S5: Iterative convergence.

[0067] After three rounds of iteration, the RMSE was reduced to 0.03 m / s, and the final flow field was output.

[0068] Example 2

[0069] This embodiment is applied to an industrial electronics workshop with dimensions of 20m×15m×5m.

[0070] Scene characteristics: There are multiple heating devices and half-height partitions, and the airflow exhibits significant turbulent characteristics.

[0071] Specific steps:

[0072] Step S1: Initial sampling.

[0073] Perform coarse sampling using a 3m×3m grid, and extend the single-point sampling time to 60s to obtain the time-averaged wind speed.

[0074] Step S2: Algorithm adaptation.

[0075] The system detected a significant sill value in the spatial semivariogram of the data and determined it to be a "strong turbulence / random field".

[0076] The system automatically switches to Ordinary Kriging interpolation. This algorithm uses the semi-variogram γ(h) to express spatial correlation and can effectively handle random fluctuations in the physical field. Estimation formula:

[0077]

[0078] Where the weight λ i This is obtained by solving the Kriging equations, which require unbiased constraints. And it minimizes the estimated variance. The empirical estimate of the semivariogram is obtained by statistical analysis grouping by distance: Where h = ||x i-x j || is the Euclidean distance between points, Ω(h) is the set of point pairs whose distances fall within the interval , and N(h) is the number of point pairs in this set. The semi-variogram model selects the spherical model: when 0 < h ≤ a, when h > a, y(h) = c0 + c; when h = 0, y(h) = 0. Where c0 is the nugget, c0 + c is the sill value, and a is the range. The parameters c0, c, and a are obtained by least squares fitting of the empirical semi-variogram .

[0079] Step S3: Flow field analysis.

[0080] In this embodiment, the flow field information entropy is selected as the evaluation index to capture the uncertainty of the flow field. The space is divided into several local windows, and each window is a three-dimensional voxel window of L x ×L y ×L z (such as 2m × 2m × 1m), and it slides to cover the entire domain with a fixed step size. Within each window, the probability distribution of the velocity modulus v = ||V|| is statistically analyzed. The value range of v is divided into B bins (such as B = 10 - 30), and the probability corresponding to the j-th bin is defined as where n j is the sample count falling into the j-th bin within this window, and the sample is the reconstructed velocity modulus value at each grid node within this window; the bin boundaries are equally wide divided according to the minimum and maximum values of the velocity modulus v within this window. Furthermore, the entropy value H of this window is calculated: According to the information theory specification, when p j = 0, it is stipulated that p j j log2p j = 0.

[0081] The system defines the window area where the entropy value is at the top preset percentage (such as the top 5% - 20%) of the global statistical distribution and satisfies the connectivity constraint as the "high-entropy information enrichment area", where the connectivity constraint is connected by three-dimensional 6-neighborhood or 26-neighborhood, and the number of consecutive windows is not less than N w ≥ 4.

[0082] Such as Figure 4The figure shows a schematic diagram of the spatial distribution of information entropy in a complex workshop provided by an embodiment of the present invention. In the figure, reference numeral 1 represents a half-height partition, and reference numeral 3 represents a heat source device (e.g., a reflow oven). In the complex disturbance scenario of this embodiment, the system calculates the global information entropy through a model; the denser the contour lines in the figure, the higher the information entropy. As shown in the figure, due to the thermal plume fluctuations generated by the heat source device 3, a high-entropy region is formed above it, as shown in reference numeral 4; simultaneously, when the airflow passes through the half-height partition 1, a vortex wake is generated on its leeward side, forming another high-entropy region, as shown in reference numeral 2. By identifying and extracting the high-entropy information-rich regions where reference numerals 2 and 4 are located, the system can accurately pinpoint the airflow blind spots and locations of severe disturbances in the environment.

[0083] Step S4: Supplementary sampling.

[0084] For high-entropy regions, the system no longer simply adds planar points, but instructs the AGV to perform multi-layered encrypted sampling in the vertical direction (such as sampling at heights of 1m, 2m, and 3m respectively) to capture the three-dimensional structure of the thermal plume.

[0085] Step S5: Iterative convergence.

[0086] After incorporating the new data, the Kriging interpolation was rerun to update the flow field. Due to the inherent volatility of the turbulent environment, this embodiment sets the convergence criterion (RMSE threshold) to 0.08 m / s (slightly wider than that for laminar flow) to prevent the system from entering an infinite loop. After four iterations, the model converged.

Claims

1. An adaptive iterative sampling and reconstruction method for indoor environmental flow fields, characterized in that, Includes the following steps: Step S1: Control the mobile sampling device to perform the first round of sampling in the indoor space to be measured according to the initial planned path to obtain the initial discrete measurement point data set; Step S2: Obtain the flow field scene features of the indoor space to be tested, and adaptively call the matching flow field reconstruction algorithm according to the flow field scene features; construct a continuous flow field model of the indoor space based on the current discrete measurement point data set and the called flow field reconstruction algorithm, and generate intermediate flow field data; Step S3: Perform a global analysis of the intermediate flow field data using flow field evaluation indices, calculate the evaluation index values ​​of each spatial grid node, and identify continuous regions whose evaluation index values ​​meet preset judgment conditions as information-rich regions that need to be supplemented with additional sampling; the preset judgment conditions include at least one of threshold judgment and percentile judgment. Step S4: Based on the spatial distribution characteristics of the information-rich area, automatically plan the coordinates and path of the new sampling point, and drive the mobile sampling device to the new sampling point to collect supplementary data; Step S5: Incorporate the supplementary data into the discrete measurement point data set, and repeat steps S2 to S4 until the preset convergence condition is met, and output the final indoor flow field reconstruction result.

2. The method according to claim 1, characterized in that, The flow field reconstruction algorithm includes at least one of geostatistical interpolation algorithm or kernel function-based interpolation method; the geostatistical interpolation algorithm includes ordinary kriging interpolation algorithm; the kernel function-based interpolation method includes radial basis function interpolation method.

3. The method according to claim 2, characterized in that, In step S2, acquiring the flow field scene features of the indoor space under test includes any of the following methods: Method 1: Based on the initial discrete measurement point data set, calculate the spatial distribution variance or Reynolds number estimate of the data to determine the flow field type characteristics; wherein, the Reynolds number estimate Re est according to The calculations are performed, where ρ and μ are the indoor air density and dynamic viscosity constants determined based on the average ambient temperature or real-time sensor measurements, and v... ref L is the time average of the velocity vector modulus at the initial discrete measurement points, or the arithmetic mean of the time averages at all measurement points. ref The equivalent hydraulic diameter or minimum geometric side length of the space to be measured, or the average nearest neighbor distance of the current discrete measuring point; Method 2: Obtain the boundary geometric information of the indoor space to be measured by an environmental perception sensor mounted on a mobile sampling device to determine the spatial scale parameters; wherein, the environmental perception sensor includes a lidar or a depth camera, and the boundary geometric information is obtained through real-time localization and mapping and boundary extraction.

4. The method according to claim 3, characterized in that, The specific strategy for adaptively invoking the matching flow field reconstruction algorithm based on the flow field scene characteristics includes: When the flow field type characteristic is determined to be laminar flow dominant, or the spatial scale parameter is less than the preset spatial scale threshold L th At that time, the Gaussian kernel radial basis interpolation algorithm is invoked; When the flow field type characteristic is determined to be turbulent, or the spatial scale parameter is greater than or equal to the preset spatial scale threshold L th At that time, the ordinary Kriging interpolation algorithm is invoked; Wherein, the L th Press L th =βL min Confirmed, L min β is the minimum geometric feature length of the indoor space to be measured, and β is a scaling factor of 0.05 to 0.

20.

5. The method according to claim 1, characterized in that, In step S3, the flow field evaluation index includes one or any combination of velocity gradient modulus, local velocity variance, flow field information entropy, and vorticity intensity. The step of identifying continuous regions whose evaluation index values ​​meet preset judgment conditions as information-rich regions requiring supplementary sampling includes: calculating the evaluation index values ​​of all grid nodes and constructing an evaluation index distribution map; extracting continuous grid regions whose index values ​​are at the top of the statistical distribution by a preset percentage as candidate information-rich regions; and determining candidate regions with an area of ​​not less than N grid units as information-rich regions. Wherein, the preset percentage P is 5% to 20%, the continuous grid region is a region formed by connecting four or eight neighboring regions, and N≥4.

6. The method according to claim 1, characterized in that, The convergence condition in step S5 is: Calculate the root mean square error of the velocity vector modulus difference between intermediate flow field data generated in adjacent iterations at corresponding spatial grid points in the global domain, wherein the adjacent iterations include the kth iteration and the (k-1)th iteration. When the root mean square error is less than a preset accuracy threshold, it is determined that the convergence condition is met; wherein, the preset accuracy threshold is an absolute error threshold or a relative error percentage threshold.

7. The method according to claim 1, characterized in that, In step S4, the driving of the mobile sampling device to the new sampling point to collect supplementary data includes: Plan the coordinates of the newly added sampling points, the coordinates including horizontal coordinates (x, y) and vertical height coordinates z; The autonomous mobile chassis of the mobile sampling device is controlled to move horizontally to the horizontal coordinate (x, y), and the vertical lifting mechanism mounted on the chassis is controlled to adjust the sampling height to the vertical height coordinate z, so as to obtain supplementary flow field data in three-dimensional space.

8. The method according to claim 2, characterized in that, In step S2, when the flow field reconstruction algorithm uses the radial basis function interpolation method, the shape parameter c of the kernel function is dynamically adjusted according to the average pairwise distance or the average nearest neighbor distance of the current discrete measurement point data set; where c = α·d avg d avg α is the average pairwise distance or the average nearest neighbor distance of the current discrete measurement points, and α is a scaling factor of 0.5 to 2.

0.

9. An adaptive iterative sampling and reconstruction system for indoor environmental flow fields, characterized in that, include: A mobile sampling device and a sensor assembly mounted on the mobile sampling device; the sensor assembly includes at least a wind speed sensor and a lidar or depth camera for acquiring boundary geometry information; Memory, used to store computer programs; A processor is communicatively connected to the mobile sampling device and the sensor assembly; when executing the computer program, the processor implements the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 8.

Images