A method and system for reducing the order of flow field data for air pollution prediction

Abstract

The present invention discloses a method and system for reducing the order of flow field data for air pollution prediction, and belongs to the technical field of air pollution prediction. Existing air pollution data processing schemes need to process a large amount of data, and the calculation time is generally long, the memory consumption is large, and the simulation cost is very high. A method for reducing the order of flow field data for air pollution prediction of the present invention decomposes large-scale original flow field data into discrete blocks, that is, different sub-matrices, and each sub-matrix can be decomposed and calculated separately, so that large-scale simulation data can be processed under limited hardware resources, and the speed of decomposition calculation is improved. At the same time, the present invention solves the singular value of each sub-matrix to obtain the flow characteristic basis function, so that a small number of basis functions can be used to accurately describe the original flow field data, achieve the purpose of data reduction, further reduce the requirements for computing resources and equipment, and reduce the simulation cost.

Description

Technical Field

[0001] The invention relates to a flow field data order reduction processing method and system for air pollution prediction, belonging to the technical field of air pollution prediction. Background Art

[0002] In the normal operation and emergencies of some factories or industrial parks, a small amount of radioactive substances and other atmospheric pollutants may be released. The diffusion and deposition of these pollutants may pose a potential threat to the surrounding environment and public health. Therefore, accurately monitoring and predicting the propagation path and concentration distribution of these pollutants in the atmosphere has become an important task for environmental protection and public safety.

[0003] For example, for nuclear power plants or chemical plants, the geographical environment around them is generally complex, including a variety of terrain and buildings, which can have a significant impact on the diffusion process of atmospheric pollutants. In addition, pollutants emitted by factories may be affected by variable meteorological conditions, such as wind speed, wind direction, temperature gradient, etc., making their diffusion behavior more difficult to predict.

[0004] In this context, numerical simulation technology based on computational fluid dynamics (CFD) has become a key tool for studying the diffusion process of atmospheric pollutants.

[0005] Furthermore, a Chinese patent (publication number: CN106650158B) discloses a method for estimating the global atmospheric environment of a city in real time based on CFD and multiple data sources, including combining historical data of urban environmental monitoring stations, global mesoscale meteorological forecast results, national meteorological data, urban key pollution source data, urban geographic three-dimensional model and real-time monitoring data of the road surface of the motor vehicle exhaust detection system, using CFD as a calculation engine, adaptively switching the environmental quality mode according to meteorological information, using a multi-scale grid discretized urban model, introducing a multi-component pollution model, and considering the meteorological dry and wet deposition process, to establish a real-time estimation method for the global air environment quality of the city.

[0006] Although the above scheme can simulate the diffusion process of atmospheric pollutants in complex terrain and changeable meteorological conditions through computational fluid dynamics simulation, such numerical simulation requires processing a large amount of data, the calculation time is generally long, the memory consumption is large, and a high hardware configuration is required, resulting in a high simulation cost.

[0007] Furthermore, in actual application scenarios, the prediction of air pollution often faces the challenges of tight time and fast iteration, and requires repeated large-scale computational fluid dynamics numerical simulations. Therefore, the above-mentioned large-scale data processing scheme cannot achieve rapid prediction of air pollution, which is not conducive to the actual promotion and application of air pollution prediction schemes.

[0008] The information disclosed in this Background Art is only for understanding the background of the inventive concept and therefore it may include information that does not constitute the prior art. Summary of the invention

[0009] In view of the above problems or one of the above problems, the first object of the present invention is to provide a flow field data reduction processing method and system for air pollution prediction, decomposing large-scale original flow field data into discrete blocks, that is, different sub-matrices, each sub-matrix can be decomposed and calculated separately, so that the sub-matrix allocated by each processor is very small in size, and can also be decomposed and calculated for a computer with a general configuration, and through the distribution and parallelization of data, the speed of decomposition calculation can also be improved, so that large-scale computational fluid dynamics simulation data can be processed under limited hardware resources. At the same time, the present invention solves the singular value of each sub-matrix to obtain the flow characteristic basis function, so that a small number of basis functions can be used to accurately describe the original flow field data, achieve the purpose of data reduction, further reduce the requirements for computing resources and equipment, and reduce the cost of simulation.

[0010] In response to the above problem or one of the above problems, the second purpose of the present invention is to provide a flow field data reduction processing method and system for atmospheric pollution prediction, which can effectively improve the efficiency of decomposition calculation and shorten the calculation time, so as to reconstruct the flow field data faster and realize the rapid prediction of atmospheric pollution. It can also significantly reduce the calculation time and memory consumption, is suitable for various scenarios, is conducive to the practical promotion and application of atmospheric pollution prediction schemes, and can provide stronger technical support for the monitoring and risk assessment of the surrounding environment of factories or parks.

[0011] To achieve one of the above purposes, the first technical solution of the present invention is:

[0012] A flow field data reduction processing method for air pollution prediction includes the following contents:

[0013] Through the scene twin units created in advance, the corresponding geometric grid model is constructed according to the terrain and building data of a certain area;

[0014] Using the air sampling unit created in advance, sample working conditions are sampled according to the historical meteorological and gas emission data of a certain area to obtain air sample data;

[0015] The flow field generation unit created in advance is used to couple the geometric grid model and the air sample data, and computational fluid dynamics simulation is performed to obtain a snapshot data matrix containing time evolution as the original flow field data;

[0016] Use the flow field decomposition unit created in advance to decompose the snapshot data matrix into several sub-matrices, and solve the singular value of each sub-matrix to obtain the flow characteristic basis function;

[0017] Based on the flow field reduction reconstruction unit created in advance, the reduced-order flow field data is constructed according to the flow characteristic basis function, and the flow field data reduction processing for atmospheric pollution prediction is realized.

[0018] The present invention decomposes large-scale original flow field data into discrete blocks, i.e., different sub-matrices, by creating scene twin units, air sampling units, flow field generation units, flow field decomposition units, and flow field reduction reconstruction units. Each sub-matrix can be decomposed and calculated separately, so that the sub-matrix assigned by each processor is very small in scale, and can also be decomposed and calculated for computers with general configurations, and through the distribution and parallelization of data, the speed of decomposition calculation can also be improved, so that large-scale computational fluid dynamics simulation data can be processed under limited hardware resources. At the same time, the present invention solves the singular value of each sub-matrix to obtain the flow characteristic basis function, so that a small amount of basis functions can be used to accurately describe the original flow field data, achieve the purpose of data reduction, further reduce the requirements for computing resources and equipment, and reduce the cost of simulation.

[0019] Therefore, the present invention can effectively improve the efficiency of decomposition calculation and shorten the calculation time, thereby reconstructing flow field data more quickly and realizing rapid prediction of air pollution. It can also significantly reduce the calculation time and memory consumption, and is suitable for various scenarios. It is conducive to the practical promotion and application of air pollution prediction schemes, and can provide more powerful technical support for the monitoring and risk assessment of the surrounding environment of factories or parks.

[0020] As the preferred technical measures:

[0021] The method of constructing the corresponding geometric mesh model based on the terrain and building data of a certain area through the scene twin unit created in advance is as follows:

[0022] The certain area is a certain factory or a certain park;

[0023] Obtain terrain and building data of a factory or park, including surrounding terrain information and building data;

[0024] The scene twin unit is used to process the surrounding terrain information and building data to obtain a geometric grid model, which is used to characterize the spatial characteristics of a certain area and serve as the basis for computational fluid dynamics simulation.

[0025] As the preferred technical measures:

[0026] Using the previously created air sampling unit, based on the historical meteorological and gas emission data of a certain area, the method for obtaining air sample data is as follows:

[0027] Obtain historical meteorological and gas emission data for a certain area, including wind speed information, wind direction information, temperature information, and pollutant emission data;

[0028] The historical meteorological and gas emission data are sampled through the air sampling unit to determine the reasonable operating parameter values ​​and obtain the air sample data as the boundary input condition for computational fluid dynamics simulation.

[0029] As the preferred technical measures:

[0030] The method for sampling historical meteorological and gas emission data through the air sampling unit is as follows:

[0031] Step 1, processing wind speed information, wind direction information, temperature information and pollutant emission data to obtain four dimensions;

[0032] Step 2, divide each dimension into n non-overlapping intervals;

[0033] Step 3: For each dimension, randomly select a data point from its n intervals and ensure that the data point in each interval is sampled only once, thus obtaining n groups of data points.

[0034] Step 4: Randomly arrange the n groups of data points in each dimension to obtain air sample data.

[0035] As the preferred technical measures:

[0036] The method of using the previously created flow field generation unit to couple the geometric grid model and the air sample data and perform computational fluid dynamics simulation to obtain the snapshot data matrix containing the time evolution is as follows:

[0037] Based on the geometric grid model and the flow characteristics of the atmospheric flow field, a potential temperature calculation equation is established to measure the heat contained in air blocks at different heights.

[0038] Based on the air sample data, component transport equations are established to simulate the diffusion process of atmospheric pollutants;

[0039] According to the potential temperature calculation equation and the component transport equation, computational fluid dynamics simulation calculation is performed to obtain a three-dimensional discrete data set including time evolution;

[0040] The three-dimensional discrete data set is processed to obtain a data snapshot matrix.

[0041] As the preferred technical measures:

[0042] Using the flow field decomposition unit created earlier, the snapshot data matrix is ​​decomposed into several sub-matrices, and the singular value of each sub-matrix is ​​solved to obtain the flow characteristic basis function as follows:

[0043] Decompose the snapshot data matrix into several sub-matrices by row;

[0044] Calculate the covariance matrix of each submatrix and perform eigenvalue decomposition on the covariance matrix to obtain a reduced-order covariance matrix;

[0045] The reduced-order covariance matrix is ​​subjected to singular value decomposition to obtain eigenvalues; and based on the eigenvalues, the flow characteristic basis functions are established.

[0046] As the preferred technical measures:

[0047] The method of performing singular value decomposition on the reduced-order covariance matrix and establishing the flow characteristic basis function is as follows:

[0048] Perform singular value decomposition on the reduced-order covariance matrix to obtain several singular values;

[0049] Sort a number of singular values ​​and select the cutoff rank of the reduced covariance matrix according to a preset accuracy tolerance;

[0050] According to the truncation rank, the singular value decomposition result of the reduced-order covariance matrix is ​​truncated to obtain a new right singular matrix and a singular value matrix;

[0051] Since the reduced-order covariance matrix is ​​a symmetric square matrix, the new right singular matrix and the transposed singular value matrix are multiplied to obtain a new combined matrix;

[0052] Perform singular value decomposition on the combined matrix, and truncate the decomposed matrix to obtain the eigenvalues ​​of the reduced-order covariance matrix;

[0053] The eigenvalue is multiplied by the column vector on the submatrix to obtain the flow characteristic basis function.

[0054] As the preferred technical measures:

[0055] Based on the previously created flow field reduction reconstruction unit and the flow characteristic basis function, the method for constructing the reduced-order flow field data is as follows:

[0056] Perform singular value decomposition on the reduced-order covariance matrix to obtain a left singular value matrix, a singular value matrix, and a right singular value matrix;

[0057] According to the left singular value matrix, the singular value matrix, the characteristic coefficients are calculated;

[0058] Multiply the flow characteristic basis function and the characteristic coefficient to obtain the flow field characteristic value;

[0059] Arrange multiple flow field eigenvalues ​​to construct reduced-order flow field data.

[0060] To achieve one of the above purposes, the second technical solution of the present invention is:

[0061] A flow field data reduction processing method for air pollution prediction comprises the following steps:

[0062] Step 1: According to the historical meteorological and gas emission data of a certain area, sample working conditions are sampled to obtain air sample data;

[0063] Step 2: Perform computational fluid dynamics simulation calculation on the air sample data to obtain a snapshot data matrix containing time evolution as the original flow field data;

[0064] Step 3: decompose the snapshot data matrix into several sub-matrices, and perform parallel singular value calculation on each sub-matrix to obtain the flow characteristic basis function;

[0065] Step 4: According to the flow characteristic basis function, the reduced-order flow field data is constructed to realize the flow field data reduction processing for air pollution prediction.

[0066] The present invention solves the singular values ​​of the original flow field data to obtain the flow characteristic basis functions, so that the original flow field data can be accurately described using a small number of basis functions, thereby achieving the purpose of data reduction, reducing the requirements for computing resources and equipment, and improving computing efficiency.

[0067] In order to further reduce the memory consumption of singular value decomposition and improve the calculation speed, the present invention decomposes large-scale original flow field data into discrete blocks, that is, different sub-matrices, and each sub-matrix can be decomposed and calculated separately. In this way, the sub-matrix allocated to each processor is very small in size, so it can also be decomposed and calculated for a computer with a general configuration. In addition, through the distribution and parallelization of data, the speed of decomposition calculation can also be improved, so that large-scale computational fluid dynamics simulation data can be processed under limited hardware resources, thereby further improving the efficiency of decomposition calculation and shortening the calculation time.

[0068] To achieve one of the above purposes, the third technical solution of the present invention is:

[0069] A flow field data reduction processing system for air pollution prediction, comprising:

[0070] one or more processors;

[0071] A storage device for storing one or more programs;

[0072] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned flow field data reduction processing method for air pollution prediction.

[0073] Compared with the prior art solutions, the present invention has the following beneficial effects:

[0074] The present invention decomposes large-scale original flow field data into discrete blocks, i.e., different sub-matrices, and each sub-matrix can be decomposed and calculated separately. In this way, the sub-matrix assigned to each processor is small in size, and can be decomposed and calculated for a computer with a general configuration. In addition, the speed of decomposition calculation can be improved through data distribution and parallelism, so that large-scale computational fluid dynamics simulation data can be processed under limited hardware resources. At the same time, the present invention solves the singular value of each sub-matrix to obtain the flow characteristic basis function, so that the original flow field data can be accurately described by using a small number of basis functions, the purpose of data reduction is achieved, and the requirements for computing resources and equipment are further reduced.

[0075] Therefore, the present invention can effectively improve the efficiency of decomposition calculations, shorten the calculation time, thereby reconstructing flow field data more quickly, and can significantly reduce the calculation time and memory consumption. It is suitable for various scenarios and can provide more powerful technical support for the monitoring and risk assessment of the surrounding environment of factories or parks. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a schematic diagram of a first flow chart of the flow field data order reduction processing method of the present invention;

[0077] Figure 2 A second flow chart of the flow field data order reduction processing method of the present invention;

[0078] Figure 3 A geometric model of the flow field data order reduction processing method of the present invention;

[0079] Figure 4 A grid model of the flow field data order reduction processing method of the present invention;

[0080] Figure 5 It is a data partitioning result of the flow field data order reduction processing method of the present invention. DETAILED DESCRIPTION

[0081] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0082] On the contrary, the present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention as defined by the claims. Further, in order to make the public have a better understanding of the present invention, some specific details are described in detail in the following detailed description of the present invention. Those skilled in the art can fully understand the present invention without the description of these details.

[0083] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention pertains. The terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" as used herein includes any and all combinations of one or more of the associated listed items.

[0084] like Figure 1 As shown, the first specific embodiment of the flow field data reduction processing method for air pollution prediction of the present invention is:

[0085] A flow field data reduction processing method for air pollution prediction includes the following contents:

[0086] Through the scene twin units created in advance, the corresponding geometric grid model is constructed according to the terrain and building data of a certain area;

[0087] Using the air sampling unit created in advance, sample working conditions are sampled according to the historical meteorological and gas emission data of a certain area to obtain air sample data;

[0088] The flow field generation unit created in advance is used to couple the geometric grid model and the air sample data, and computational fluid dynamics simulation is performed to obtain a snapshot data matrix containing time evolution as the original flow field data;

[0089] Use the flow field decomposition unit created in advance to decompose the snapshot data matrix into several sub-matrices, and solve the singular value of each sub-matrix to obtain the flow characteristic basis function;

[0090] Based on the flow field reduction reconstruction unit created in advance, the reduced-order flow field data is constructed according to the flow characteristic basis function, and the flow field data reduction processing for atmospheric pollution prediction is realized.

[0091] A second specific embodiment of the flow field data reduction processing method for air pollution prediction of the present invention:

[0092] A flow field data reduction processing method for air pollution prediction includes a scene twin unit, an air sampling unit, a flow field generation unit, a flow field decomposition unit, and a flow field reduction reconstruction unit.

[0093] The scene twin unit is used to construct a computational fluid dynamics simulation model; the air sampling unit is used to determine the sample operating parameters for computational fluid dynamics simulation calculations; the flow field generation unit obtains a spatial three-dimensional discrete data set containing time evolution and constructs it into a snapshot data matrix; the flow field decomposition unit is used to decompose the snapshot data matrix into flow characteristic basis functions and corresponding characteristic coefficients; the flow field reduced-order reconstruction unit is used to reconstruct the original flow field data based on the flow characteristic basis functions and the corresponding characteristic coefficients.

[0094] In this embodiment: the scene twin unit includes the following contents:

[0095] A certain area is a certain factory, and the terrain data of the factory building and its surroundings are obtained; through the terrain data of the factory building and its surroundings, the geometric and grid models of the atmospheric environment of the factory are constructed as the basis of the computational fluid dynamics simulation model.

[0096] In this embodiment: the air sampling unit includes the following contents:

[0097] Based on a factory's historical meteorological data such as wind speed, wind direction, temperature, etc., as well as records of different types of pollutant emissions, air sampling is carried out and reasonable operating parameter values ​​are determined as boundary input conditions for computational fluid dynamics simulation calculations.

[0098] In this embodiment: the air sampling unit is constructed based on the hierarchical Monte Carlo sampling method, and its sampling mechanism is as follows:

[0099] Assume that the system has m factors and each factor has n levels. First, the design space of each factor will be divided into n sub-design spaces, and a value will be randomly selected in each sub-design space. In this way, a corresponding sampling matrix with n samples will be generated for each design space. The sampling process must comply with two principles: first, the sample points in each sub-design space must be randomly selected, and second, there must be only one value selected in each sub-design space.

[0100] Suppose there is a dimensional parameter space, the dimension indicates the type of parameter, and each parameter The value range is , in this parameter space The steps for sampling are as follows:

[0101] Step 1: Split each dimension into Interval: For the Parameters , its range Divide into disjoint intervals, each of which has a length for:

[0102]

[0103] Step 2: Perform random sampling in each dimension: For each dimension , randomly select a point from each interval and ensure that the points in each interval are sampled only once.

[0104] Step 3: Randomly arrange the sampling points of each dimension: In each dimension, the sampling points of each dimension are randomly arranged independently to form Group sampling points.

[0105] In this embodiment: the flow field generation unit includes the following contents:

[0106] The integral or partial differential equations in the control equations are replaced by discrete algebraic forms, and then the specific numerical solutions are solved for data comparison and explanation of experimental phenomena. The physical basis behind it and the control equations based on different physical laws are listed below. The control equations derived from the three physical laws, that is, all engineering fluid problems in reality follow the laws of conservation of matter, conservation of momentum, and conservation of energy.

[0107] The continuity equation is given by:

[0108]

[0109] The momentum equation is shown below:

[0110]

[0111] The energy conservation equation is shown below:

[0112]

[0113] In the formula, is the density, It's time. For speed, For pressure, is the shear force, is the Reynolds number, is the temperature, is the thermal conductivity, is the isobaric specific heat capacity, is the heating / cooling source term, is the temperature change, is the Hamiltonian operator.

[0114] There are some differences between the flow behavior of atmospheric flow field and ordinary flow field. Therefore, in order to better simulate the atmospheric flow field, additional formulas are needed to describe the flow field under the atmospheric environment. The atmospheric flow field is reflected in the unevenness of temperature, pressure, and air density in the vertical direction. As the altitude increases, the temperature, pressure, and air density decrease accordingly.

[0115] In atmospheric science, the concept of potential temperature is used to measure the heat contained in air parcels at different altitudes. In the dry adiabatic process, although there is no heat balance or latent heat conversion, there will still be obvious temperature changes due to the expansion or compression work done by the air parcel during the rise and fall. For different air parcels, air parcels with high potential temperature represent warmer air parcels, and air parcels with low potential temperature represent colder air parcels.

[0116] Temperature The expression is as follows:

[0117]

[0118] In the formula, Normal pressure, usually 1000mbar, For pressure, is the dry air constant, is the isobaric specific heat capacity.

[0119] The dry air constant is calculated as follows:

[0120] Furthermore, the expression of the atmospheric flow field is as follows:

[0121]

[0122] In the formula, is the heating / cooling source term, is the thermal conductivity, is the density, Indicates the speed in different directions. Indicates the distance in different directions. The value is 0, 1 or 2, which is used to indicate Three directions, is the isobaric specific heat capacity, By modifying the relevant formulas, a more realistic simulation calculation of the atmospheric flow field can be achieved.

[0123] For atmospheric pollutants, in addition to solving the control equation, it is also necessary to solve the component transport equation that the pollutants satisfy during the diffusion process. The calculation formula is as follows:

[0124]

[0125]

[0126] In the formula, is the mass average density, is the subscript of the component, For the The density of the component, For the The mass content of the components, is the turbulent viscosity, For the The diffusion coefficient of the component, is the Schmidt number of the fluid, It is The source term for the creation or absorption of a component.

[0127] The kinematic viscosity of the gas is calculated using the Sutherland formula, which is as follows:

[0128]

[0129] In the formula, is the kinematic viscosity, Under atmospheric pressure The viscosity coefficient; is a constant related to the type of gas; ; Can represent a variety of different air pollutants.

[0130] According to the determined sample operating condition parameters, computational fluid dynamics simulation calculations are performed to obtain simulation calculation data, namely a three-dimensional discrete data set.

[0131] In this embodiment: According to the three-dimensional discrete data set, the method of constructing a data snapshot matrix is ​​as follows:

[0132] Read a three-dimensional discrete data set, select several physical quantity data as samples in the three-dimensional discrete data set, and form a data snapshot matrix , which is expressed as follows:

[0133]

[0134] In the formula, is the number of spatial grid points, To calculate the number of sample conditions, Indicated in Among the sample conditions, Physical quantity data at grid points, matrix The size is OK List.

[0135] In this embodiment: the flow field decomposition unit includes the following contents:

[0136] Data snapshot matrix Perform singular value decomposition to obtain the left singular value matrix , singular value matrix and the right singular value matrix , which is expressed as follows:

[0137]

[0138] Since the size of the singular value represents the contribution of the corresponding flow mode to the flow field energy, the singular value matrix Included In order to reduce the number of equations, some modes with smaller contributions can be deleted, the first r modes can be truncated, and only larger singular values ​​can be retained. The expression is as follows:

[0139]

[0140] in, is the left singular value matrix of the first r modes of interception, is the singular value matrix of the first r modes of interception, is the right singular value matrix of the first r modes of truncation.

[0141] Furthermore, in computational fluid dynamics calculations, the number of grids is often much larger than the number of time steps, i.e. , so the singular value method is not directly used to analyze the data snapshot matrix Instead of decomposing, singular value decomposition is performed through the following steps, as follows:

[0142] First, the data snapshot matrix The covariance matrix of Perform eigenvalue decomposition and only take the first r-order eigenvalues. The expression is as follows:

[0143]

[0144] in, is the covariance matrix No. eigenvalues, is the covariance matrix No. feature vectors.

[0145] Covariance matrix The size is OK List, .

[0146] The covariance matrix is ​​decomposed into singular values, and its expression is as follows:

[0147]

[0148] Since the covariance matrix is ​​a symmetric square matrix, the left and right singular value matrices obtained by decomposing the covariance matrix are transposed of each other.

[0149] For the data snapshot matrix and its covariance matrix , their right singular value matrices are the same, and the data snapshot matrix The singular values ​​of the covariance matrix The positive quadratic root of , so the expression of the flow characteristic basis function is as follows:

[0150]

[0151] in, , is the covariance matrix No. The eigenvalue of values, is the data snapshot matrix No. List.

[0152] In this embodiment, since the data snapshot matrix The number of rows is M, that is, the number of grid points in the calculation model is very large, and a OK The left singular value matrix of the column requires a lot of CPU resources and memory space, so the data snapshot matrix Distribute data to different CPUs by rows or columns and perform calculations in parallel to speed up data calculations and reduce computing resource consumption.

[0153] The data snapshot matrix Decompose it into p sub-matrices by row and distribute them to p CPU cores. The expression is as follows:

[0154]

[0155] in, For the sub-matrices.

[0156] Furthermore, Perform singular value decomposition, the expression is as follows:

[0157]

[0158] in, For the The left singular value matrix of the submatrices, For the The singular value matrix of the submatrices, For the The right singular value matrix of the sub-matrices.

[0159] because The scale is , so the computational resource consumption of singular value decomposition is very small.

[0160] Therefore, for the data snapshot matrix , and distribute it to each CPU core, the expression is as follows:

[0161]

[0162] Furthermore, the method for parallel calculation of the flow field decomposition unit is as follows:

[0163] Step 1: Calculate the covariance matrix on each CPU core separately The singular value decomposition of is expressed as follows:

[0164]

[0165]

[0166] in, is the combined matrix.

[0167] Step 2: Snapshot the data matrix Decompose by row into Sub-matrices , and perform singular value decomposition on the submatrix on each CPU core. The singular value decomposition expression on the CPU core is as follows:

[0168]

[0169] Step 3: For each submatrix , according to the singular value of singular value decomposition, truncate, in the CPU cores, truncated rank Satisfies the following inequality:

[0170]

[0171] in, is the set accuracy tolerance, is a sub-matrix No. Due to the characteristics of singular value decomposition, the singular values ​​are arranged from high to low, with a precision tolerance of As the boundary, select each submatrix The truncated rank .

[0172] Step 4: Submatrix The singular value decomposition result of is truncated, and its expression is as follows:

[0173]

[0174]

[0175] in, is the order of truncation of the i-th submatrix, For the The front of the submatrix Order right singular value matrix; For the The front of the submatrix Order singular value matrix.

[0176] Sub-matrix The transpose point multiplication of the right singular matrix and the singular value matrix of , and the new combined matrix is ​​obtained , which is expressed as follows:

[0177]

[0178] Step 5: Combine the matrix Perform singular value decomposition and truncate the decomposed matrix. The expression is as follows:

[0179]

[0180]

[0181] in, is the left singular value matrix, is the singular value matrix, is the right singular value matrix.

[0182] Truncated rank Meet the following requirements:

[0183]

[0184] in, is the set accuracy tolerance.

[0185] Step 6: Calculate the submatrix allocated on each CPU The corresponding flow characteristic basis function is On each CPU core, the components of the flow characteristic basis function are divided by The vectors in and the local data on each submatrix Perform vector multiplication to calculate, the calculation formula is as follows:

[0186]

[0187] in, is the jth flow characteristic basis function of the ith submatrix, represents the first j columns of the left singular value matrix;

[0188] , .

[0189] Step 7, using the flow characteristic basis function to reconstruct the data snapshot matrix, which includes the following contents:

[0190] For each flow characteristic basis function , each has a corresponding set of characteristic coefficients , The calculation formula is as follows:

[0191]

[0192] Then for a certain time step or a certain working condition In this case, the flow characteristic basis function can be used by the following formula and characteristic coefficient To reconstruct the corresponding flow field , realize parallel flow field decomposition calculation, the specific formula is as follows:

[0193]

[0194] Furthermore, in order to compare the computational resource consumption of different decomposition methods, two criteria are used: (1) computational complexity expressed in floating-point operations; (2) communication overhead expressed in floating-point numbers that need to be transmitted. Then for ordinary singular value decomposition, the expression of its complexity O is as follows:

[0195]

[0196] The communication cost is .

[0197] For the flow field data order reduction processing method for air pollution prediction proposed by the present invention, the calculation formula of the complexity on each submatrix is ​​as follows:

[0198]

[0199] in, is the communication complexity for the i-th sub-matrix.

[0200] The overall communication overhead is:

[0201]

[0202] like Figure 2 As shown, the third specific embodiment of the flow field data reduction processing method for air pollution prediction of the present invention is as follows:

[0203] A flow field data reduction processing method for air pollution prediction uses a parallel method to reduce the order of CFD simulation data of air pollution diffusion in a factory. It has the advantages of low memory consumption and high computational efficiency. It includes a scene twin unit, an air sampling unit, a computational fluid dynamics numerical solution method, and a flow field decomposition unit.

[0204] The scene twin unit is used to construct a factory and its surrounding terrain and atmospheric environment into a geometric and mesh model as the basis for computational fluid dynamics simulation.

[0205] The air sampling unit is used to sample historical meteorological data of a factory, such as wind speed, wind direction, temperature, etc., and records of different types of pollutant emissions, to determine reasonable operating parameter values ​​as boundary input conditions for computational fluid dynamics simulation calculations.

[0206] The air sampling unit is a stratified Monte Carlo sampling method that can approximate random sampling from a multivariate parameter distribution. It belongs to a stratified sampling technique and is suitable for uniform sampling in multidimensional space. In the air sampling unit, each sample point in the sub-design space must be randomly selected, and secondly, only one value is selected in each sub-design space.

[0207] Then, according to the determined sample operating condition parameters, CFD simulation calculation is performed to obtain a simulation data set.

[0208] The flow field decomposition unit is a mathematical method used to decompose a non-square matrix into three matrices to characterize the characteristics of the data. Its principles and steps are as follows:

[0209] The physical quantity data obtained by simulating the diffusion of atmospheric pollution in a factory are combined into a data snapshot matrix, and the flow field data matrix is ​​defined as , read the computational fluid dynamics simulation calculation data, and in the computational fluid dynamics calculation results, select the physical quantity data in the calculation sample library as samples to form a data snapshot matrix.

[0210] The flow field decomposition unit converts the data snapshot matrix Distribute the data to different CPUs by row and run the flow field decomposition method in parallel to speed up the calculation of data and reduce the consumption of computing resources. The number of rows M is very large, which can avoid OK Matrix of left singular values ​​of columns.

[0211] In the flow field decomposition unit, the data snapshot matrix Decompose it into p sub-matrices by row and distribute them to p CPU cores. The expression is as follows:

[0212]

[0213] in, is the i-th sub-matrix, Perform singular value decomposition, the expression is as follows:

[0214]

[0215] because The scale is , so the computational resource consumption of singular value decomposition is very small.

[0216] The data snapshot matrix Decompose into Different sub-matrices, whose expressions are as follows:

[0217]

[0218] Perform singular value decomposition on the submatrix on each CPU core, then in the The decomposition expression on the CPU core is as follows:

[0219]

[0220] The singular value decomposition results on each CPU core are subject to the accuracy tolerance Due to the characteristics of singular value decomposition, the singular values ​​are arranged from high to low, and the first few singular values ​​can represent most of the matrix information. The original data matrix can be approximated by truncation. As the boundary, select each submatrix The truncated rank , through the previous The singular value decomposition process is approximated by the singular value of the order.

[0221] Truncated rank The following inequality is satisfied:

[0222]

[0223] in, is the set accuracy tolerance, is a sub-matrix No. The singular values ​​of order, is a sub-matrix No. Order singular value, when the singular value changes from Reduce to ,if Less than the set accuracy tolerance When The truncated singular value matrix is ​​obtained by truncating And the right singular value matrix , which is expressed as follows:

[0224]

[0225]

[0226] Will and The transpose of the matrix is ​​multiplied and the matrices calculated on each CPU are combined to obtain a new combined matrix , which is expressed as follows:

[0227]

[0228] For this combined matrix Perform singular value decomposition again, the expression is as follows:

[0229]

[0230]

[0231] Among them, according to the set accuracy tolerance , for the matrix The singular value decomposition result of is truncated, and its expression is as follows:

[0232]

[0233] so that is The approximation of , and the tolerance is set .

[0234] Finally, the submatrix allocated on each CPU is calculated The corresponding flow characteristic basis function is On a CPU core, the basis function components are divided by The vectors in and the submatrices on each processor Perform vector multiplication to calculate, the calculation formula is as follows:

[0235]

[0236] in, , .

[0237] For each flow characteristic basis function , each has a corresponding set of characteristic coefficients , The calculation formula is:

[0238]

[0239] Then for a certain time step or a certain working condition In this case, the flow characteristic basis function can be used by the following formula and characteristic coefficient To reconstruct the corresponding flow field , which is expressed as follows:

[0240]

[0241] In this way, the parallel flow field decomposition calculation is completed, and the original data snapshot matrix can be decomposed into different sub-matrices, and the singular value decomposition of these sub-matrices is performed separately. ) compared with the original data matrix ( ), so when performing singular value decomposition on the submatrix, the maximum size matrix generated is ( ), requiring less memory capacity and being easier to run on devices with different configurations. At the same time, since the singular value decomposition of the submatrix is ​​easier to implement, several submatrices can be calculated simultaneously through parallel programs, thereby improving the efficiency of the calculation.

[0242] This method is efficient and resource-friendly for large-scale simulation calculations of atmospheric pollution diffusion in a factory, making the use of reduced-order models for related research and applications more feasible and efficient. It is of great significance for improving the accuracy and efficiency of atmospheric pollution prediction in a factory, laying a solid foundation for the popularization and innovative application of computational fluid dynamics in multiple fields, and providing stronger technical support for monitoring and risk assessment of the environment around a factory.

[0243] A specific embodiment of applying the present invention to reduce the order of flow field data of a nuclear power plant:

[0244] Since the atmospheric flow field near a nuclear power plant is very complex, the number of degrees of freedom in the simulation may be very large. For example, the number of grids required for the atmospheric pollution diffusion simulation model of a nuclear power plant may have reached millions. For a system with 1,000,000 grid points, it is necessary to read this 106×106 matrix into the memory. Assuming that the data is stored in 16-bit floating point numbers, the memory required is as high as 1862.65GB. Such a large resource consumption is unrealistic.

[0245] In this example, the present invention is applied to reduce the order of flow field data of a nuclear power plant, which includes the following contents:

[0246] The atmospheric environment around a nuclear power plant is subjected to computational fluid dynamics under various meteorological and emission conditions. The spatial discrete data set obtained by CFD simulation is used as the source of the original snapshot data to perform flow field data reduction. The specific processing method is as follows:

[0247] Step 1: Based on the buildings and terrain elevation data around the nuclear power plant, construct the geometric and mesh models corresponding to the atmospheric environment around the nuclear power plant. The geometric and mesh models include geometric models and mesh models. Figure 3 As shown, Figure 3 The layout of nuclear power plant buildings is shown, and the grid model is as follows Figure 4 As shown, Figure 4 A more detailed mesh of the nuclear power plant buildings is shown.

[0248] Step 2: Obtain the meteorological data (wind speed, wind direction, temperature) of the geographical location of the nuclear power plant, as well as the pollution emission data of the nuclear power plant (pollutant xenon-133). Sample these four operating parameters through the air sampling unit, and obtain a total of 500 sets of data.

[0249] The four parameters of wind speed, wind direction, temperature and xenon-133 emission concentration are recorded as , each parameter has a different value range according to historical data records. , the interval length is .

[0250] The air sampling unit first divides the value range of each parameter into 500 intervals of equal length. The range of the intervals is:

[0251]

[0252] in, For the length of each interval, .

[0253] For each parameter , a point is randomly selected in each of the 500 intervals. The position of this point is determined by sampling from the distribution of the original data. For example, in In the interval, the parameter The value of can be expressed as:

[0254]

[0255] in, For the The first parameters, Is in accordance with A random variable with a distribution law of parameters.

[0256] Next, each parameter In order to avoid that the sample points of each parameter fall in the same order, the order of the samples is disrupted so that the first The samples are combined with samples of other parameters to form a unique sample point.

[0257] Finally, 500 groups of sample points were generated, and their expressions are as follows:

[0258]

[0259] in, , , , is a random arrangement of sample points.

[0260] Step 3: Using the 500 groups of sample points obtained by the air sampling unit as boundary conditions for computational fluid dynamics simulation calculations, a computational fluid dynamics simulation is performed.

[0261] The grid for computational fluid dynamics calculations is divided into several regions, resulting in several sub-matrices. The decomposition of each sub-matrix is ​​performed independently, see Figure 5 In this embodiment, the number of grids is about 2250000 and the number of sample conditions is 500, so the snapshot data matrix The size of is 2250000 rows and 500 columns. The concentration of pollutant xenon-133 is selected as the physical quantity data. Its expression is as follows:

[0262]

[0263] Step 4: Snapshot Data Matrix Decompose into 56 sub-matrices by row .

[0264] Step 5: For each sub-matrix Perform singular value decomposition to obtain the sub-matrix decomposition result.

[0265] Step 6: Apply the flow field decomposition unit to obtain the flow characteristic basis function on each sub-matrix. Reconstruct the original flow field data through the flow characteristic basis function to complete the snapshot data matrix Parallel singular value decomposition of .

[0266] In this embodiment, the number of sub-matrices is tested. The results of the parallel flow decomposition method are compared with those of the ordinary, serial singular value method, and different accuracy tolerances are tested. The error between the flow characteristic basis function obtained by decomposition and the original data matrix includes the following:

[0267] For the number of submatrices , decompose the original data matrix There are 56 sub-matrices, and their expressions are as follows:

[0268]

[0269] Setting the accuracy tolerance , compare the reconstructed data and snapshot data matrix The error of When the value is 0.001, the maximum error is only The order of magnitude is very high, so the flow field data order reduction processing method and the corresponding order reduction model proposed in the present invention have good accuracy.

[0270] For a common serial flow field decomposition unit, its complexity The calculation formula is as follows:

[0271]

[0272]

[0273]

[0274] The specific calculation results of communication overhead are as follows:

[0275]

[0276] The flow field data order reduction processing method proposed in this invention is applied to the sub-matrix number , accuracy tolerance Under the condition that each submatrix contains grids, the singular value decomposition truncated rank of each submatrix is about 20, and the truncation rank of the parallel flow field decomposition is , the calculation formula of the complexity of each submatrix is ​​as follows:

[0277]

[0278]

[0279]

[0280] The overall communication overhead is:

[0281]

[0282]

[0283] Therefore, it can be clearly concluded that the application of the present invention can reduce the computational complexity by two orders of magnitude. At the same time, since each sub-matrix is ​​decomposed independently, the requirements for computing resources and equipment can be further reduced, so that large-scale computational fluid dynamics simulation data can be processed even with limited hardware resources.

[0284] In addition, the present invention utilizes parallel computing to improve the efficiency of decomposition calculations and shorten the calculation time, thereby generating reduced-order flow field data more quickly, which can significantly reduce the calculation time and memory consumption. It can also promote the use of computational fluid dynamics reduced-order schemes in a wider range of application scenarios, and provide stronger technical support for monitoring and risk assessment of the environment around nuclear power plants.

[0285] An embodiment of a device applying the method of the present invention:

[0286] An electronic device comprising:

[0287] one or more processors;

[0288] A storage device for storing one or more programs;

[0289] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned flow field data reduction processing method for air pollution prediction.

[0290] A computer medium embodiment using the method of the present invention:

[0291] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned flow field data reduction processing method for air pollution prediction.

[0292] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, systems, and computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) containing computer-usable program codes.

[0293] The present application is described by flowcharts or / and block diagrams of the methods, devices (systems), and computer program products of the embodiments of the present application. It should be understood that each process or / and block in the flowchart or / and block diagram and the combination of the processes or / and blocks in the flowchart or / and block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart or / and block diagram. Figure 1 Process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0294] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 Process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0295] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 Process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0296] The unit in this application is an object that objectively describes the morphological structure with the help of physical or virtual expressions. The object is not equal to the object and is not limited to physical and virtual. It can be a data processing function, software program, processing mode, usage method, operation method, workflow, application process, electronic hardware, circuit module, processing system, system imitation or simulation object.

[0297] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field can still modify or replace the specific implementation methods of the present invention with equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A flow field data reduction processing method for air pollution prediction, characterized in that: Includes the following: Through the scene twin units created in advance, the corresponding geometric grid model is constructed according to the terrain and building data of a certain area; Using the air sampling unit created in advance, sample working conditions are sampled according to the historical meteorological and gas emission data of a certain area to obtain air sample data; The flow field generation unit created in advance is used to couple the geometric grid model and the air sample data, and computational fluid dynamics simulation is performed to obtain a snapshot data matrix containing time evolution as the original flow field data; Use the flow field decomposition unit created in advance to decompose the snapshot data matrix into several sub-matrices and distribute them to the corresponding number of CPU cores. Perform singular value calculation on each sub-matrix on each CPU core to obtain the flow characteristic basis function. The method to obtain the flow characteristic basis function is as follows: Decompose the snapshot data matrix into several sub-matrices by row; Calculate the covariance matrix of each submatrix and perform eigenvalue decomposition on the covariance matrix to obtain a reduced-order covariance matrix; Perform singular value decomposition on the reduced covariance matrix to obtain the eigenvalues; And based on the eigenvalues, the flow characteristic basis function is established; The method to establish the flow characteristic basis function is as follows: Perform singular value decomposition on the reduced-order covariance matrix to obtain several singular values; Sort a number of singular values ​​and select the cutoff rank of the reduced covariance matrix according to a preset accuracy tolerance; According to the truncation rank, the singular value decomposition result of the reduced-order covariance matrix is ​​truncated to obtain a new right singular matrix and a singular value matrix; Since the reduced-order covariance matrix is ​​a symmetric square matrix, the new right singular matrix and the transposed singular value matrix are multiplied to obtain a new combined matrix; Perform singular value decomposition on the combined matrix, and truncate the decomposed matrix to obtain the eigenvalues ​​of the reduced-order covariance matrix; Perform vector multiplication calculation on the eigenvalue and the column vector on the submatrix to obtain the flow characteristic basis function; Based on the flow field reduction reconstruction unit created in advance, the reduced-order flow field data is constructed according to the flow characteristic basis function, and the flow field data reduction processing for atmospheric pollution prediction is realized.

2. The method for reducing the order of flow field data for air pollution prediction according to claim 1, characterized in that: The method of constructing the corresponding geometric mesh model based on the terrain and building data of a certain area through the scene twin unit created in advance is as follows: The certain area is a certain factory or a certain park; Obtain terrain and building data of a factory or park, including surrounding terrain information and building data; The scene twin unit is used to process the surrounding terrain information and building data to obtain a geometric grid model, which is used to characterize the spatial characteristics of a certain area and serve as the basis for computational fluid dynamics simulation.

3. The method for reducing the order of flow field data for air pollution prediction according to claim 1, characterized in that: Using the previously created air sampling unit, based on the historical meteorological and gas emission data of a certain area, the method for obtaining air sample data is as follows: Obtain historical meteorological and gas emission data for a certain area, including wind speed information, wind direction information, temperature information, and pollutant emission data; The historical meteorological and gas emission data are sampled through the air sampling unit to determine the reasonable operating parameter values ​​and obtain the air sample data as the boundary input condition for computational fluid dynamics simulation.

4. The method for reducing the order of flow field data for air pollution prediction according to claim 3, characterized in that: The method for sampling historical meteorological and gas emission data through the air sampling unit is as follows: Step 1, processing wind speed information, wind direction information, temperature information and pollutant emission data to obtain four dimensions; Step 2, divide each dimension into n non-overlapping intervals; Step 3: For each dimension, randomly select a data point from its n intervals and ensure that the data point in each interval is sampled only once, thus obtaining n groups of data points. Step 4: Randomly arrange the n groups of data points in each dimension to obtain air sample data.

5. The method for reducing the order of flow field data for air pollution prediction according to claim 1, characterized in that: The method of using the previously created flow field generation unit to couple the geometric grid model and the air sample data and perform computational fluid dynamics simulation to obtain the snapshot data matrix containing the time evolution is as follows: Based on the geometric grid model and the flow characteristics of the atmospheric flow field, a potential temperature calculation equation is established to measure the heat contained in air blocks at different heights. Based on the air sample data, component transport equations are established to simulate the diffusion process of atmospheric pollutants; According to the potential temperature calculation equation and the component transport equation, computational fluid dynamics simulation calculation is performed to obtain a three-dimensional discrete data set including time evolution; The three-dimensional discrete data set is processed to obtain a data snapshot matrix.

6. The method for reducing the order of flow field data for air pollution prediction according to claim 1, characterized in that: Based on the previously created flow field reduction reconstruction unit and the flow characteristic basis function, the method for constructing the reduced-order flow field data is as follows: Perform singular value decomposition on the reduced-order covariance matrix to obtain a left singular value matrix, a singular value matrix, and a right singular value matrix; According to the left singular value matrix, the singular value matrix, the characteristic coefficients are calculated; Multiply the flow characteristic basis function and the characteristic coefficient to obtain the flow field characteristic value; Arrange multiple flow field eigenvalues ​​to construct reduced-order flow field data.

7. A flow field data reduction processing system for air pollution prediction, characterized by: It includes: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a flow field data reduction processing method for air pollution prediction as described in any one of claims 1-6.

Images