Grid processing method for environmental data based on graph convolutional network

Abstract

The present invention relates to the field of data processing, and to a grid point processing method for environmental data based on a graph convolutional network; the method comprises obtaining air quality monitoring data and meteorological monitoring data in a target area; performing missing processing on all monitoring data, and mapping the monitoring data of a station to a grid point matrix divided by the target area; using wind direction data and wind speed data to generate a dynamic wind field map, and using the Dijkstra algorithm to calculate a wind field adjacency matrix; constructing a mask matrix at each moment according to air quality concentration data, and constructing a feature vector set Z at each moment according to the wind field adjacency matrix, the mask matrix and meteorological monitoring data; generating a target matrix Y at each moment according to the mask matrix and the air quality concentration data; and inputting the feature vector set Z matrix into a trained graph convolutional neural network model to obtain an estimated matrix P of the target matrix. The present invention can improve the accuracy of environmental data gridding.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to a grid point processing method for environmental data based on a graph convolutional network. Background Art

[0002] In recent years, environmental pollution issues have attracted people's attention, and the density of automatic environmental monitoring sites has also greatly increased. In the context of vigorously developing intelligent grid forecasting technology and the requirements of human production activities for location-based air quality services, generating regular grid data from monitoring data with irregular spatial resolution and discrete distribution of monitoring sites has important social service and business application value.

[0003] Commonly used methods for gridding site data include objective analysis, remote sensing inversion, data assimilation, statistical interpolation, Kerry interpolation, tensor completion, etc. However, existing technologies only process single environmental data without considering the impact of other factors on the target data, resulting in unsatisfactory gridding effects. Summary of the invention

[0004] In view of the shortcomings of the above-mentioned technologies, the present invention provides a grid processing method for environmental data based on graph convolutional networks. The present invention uses meteorological monitoring data for gridding of environmental data, and utilizes the powerful fitting ability of neural networks to fit the impact of meteorological data on air quality data and the changing patterns of air quality data during the gridding process, thereby improving the accuracy of gridding of environmental data.

[0005] A grid point processing method for environmental data based on a graph convolutional network of the present invention comprises the following steps:

[0006] S1: Acquire air quality monitoring data of N air quality monitoring stations and meteorological monitoring data of M meteorological monitoring stations in the target area; the air quality monitoring data includes air quality concentration data, and the meteorological monitoring data includes humidity, temperature, wind direction and wind speed data;

[0007] S2: Preprocess the missing parts of all monitoring data, and map the monitoring data of the corresponding stations to the grid matrix divided by scale in the target area according to the coordinates of the monitoring stations;

[0008] S3: Generate a dynamic wind field map using the wind direction data and wind speed data in the meteorological monitoring data in the grid matrix, and calculate the wind field adjacency matrix using the Dijkstra algorithm;

[0009] S4: construct a mask matrix at each moment according to the air mass concentration data in the grid matrix, and construct a feature vector set Z at each moment according to the wind field adjacency matrix, the mask matrix and the meteorological monitoring data;

[0010] S5: Generate a target matrix Y at each moment according to the mask matrix and the air quality concentration data;

[0011] S6: Input the feature vector set Z matrix into the trained graph convolutional neural network model to obtain an estimated matrix P of the target matrix.

[0012] The present invention provides a grid point processing method for environmental data based on a graph convolutional network, which has the following beneficial effects:

[0013] 1. The present invention converts wind speed data and wind direction data into a dynamic wind field matrix, and then obtains a wind field adjacency matrix based on the dynamic wind field matrix. Finally, the adjacency matrix is ​​input into the model, and the mapping relationship of the wind field's influence on air quality can be learned by using a neural network model, thereby improving the gridding accuracy of air quality data;

[0014] 2. The present invention adds the spectrum-based GCN layer to the graph convolutional neural network model. During the training process, it can filter out the unfavorable factors that affect the gridding accuracy, making the grid processing effect more accurate;

[0015] 3. The present invention uses a mask matrix after random zeroing to randomly initialize the target matrix, which increases the robustness of the model and makes the gridding effect of the model on air quality more stable;

[0016] 4. When the present invention grids the environmental quality data, the F nodes before the current time node are used to provide information to the current node, so that the gridding effect distribution is more reasonable;

[0017] 5. The present invention also fully considers the influence of temperature and humidity on air quality, and the gridding effect is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0020] Figure 2 is a schematic diagram of generating mask and mask2 matrices according to an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of generating a feature vector set Z provided by an embodiment of the present invention;

[0022] Figure 4is a schematic diagram of generating a target matrix Y provided by an embodiment of the present invention;

[0023] Figure 5 It is a main data flow diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] Figure 1 It is a schematic diagram of the overall process of the present invention, such as Figure 1 As shown, the method includes:

[0026] S1: Acquire air quality monitoring data of N air quality monitoring stations and meteorological monitoring data of M meteorological monitoring stations in the target area; the air quality monitoring data includes air quality concentration data, and the meteorological monitoring data includes humidity, temperature, wind direction and wind speed data;

[0027] In the embodiment of the present invention, the step S1 may include:

[0028] S11: Obtain data of each time node with a monitoring frequency of one hour at N air quality monitoring stations in the target area;

[0029] S12: Obtain the data of each time node of humidity, temperature, wind direction, and wind speed monitored by M meteorological stations in the target area with a time frequency of one hour;

[0030] S13: Sort the air quality concentration data and meteorological monitoring data in chronological order, and align the data by time. For the missing parts, fill them with -1.

[0031] The air quality concentration data and the meteorological monitoring data are sorted according to time to form a time series, and the subsequent process will process the time series to achieve grid point processing of the time series.

[0032] It can be understood that in the embodiment of the present invention, the air quality monitoring data of the air quality monitoring station can select PM2.5 concentration data, and of course it can also be other air quality concentration data such as PM10, CO, SO2, NO2, O3, AQI, etc. For the convenience of expression, the PM2.5 concentration data is used as an example in the embodiment of the present invention.

[0033] S2: Preprocess the missing parts of all monitoring data, and map the monitoring data of the corresponding stations to the grid matrix divided by scale in the target area according to the coordinates of the monitoring stations;

[0034] In the embodiment of the present invention, step S2 includes the following steps:

[0035] S21: Obtain the boundary of the target area, divide the target area into fishing grid points (grids) according to a certain scale, and obtain the coordinates of the center point of each grid. ij , forming the center point coordinate matrix L, i.e., the grid matrix, where i and j are the row and column coordinates of the grid matrix respectively;

[0036] Among them, in this embodiment, the scale can be divided according to 1km×1km. Of course, in actual situations, technical personnel in this field can choose different scales for division according to actual needs, so as to divide the target area into different grids; the shape of the target area can be a rectangle, a circle or other irregular shapes. If it is a non-rectangular shape, it is necessary to fill the non-rectangular target area outward into a rectangle, so as to facilitate the division of the target area into grids of the same size, which is convenient for subsequent grid processing of environmental data.

[0037] Among them, after the target area is divided into multiple grids in the present invention, the coordinates of the center point of each grid are used as the coordinates of the grid, and a center point coordinate matrix is ​​formed; the row and column coordinates of the coordinate matrix can reflect the position of the grid in the target area.

[0038] S22: According to the coordinates and longitude and latitude distance of the air quality monitoring station, traverse and calculate the first distance from the air quality monitoring station to each grid point in the grid matrix, determine the grid point position when the first distance is the shortest, and record it in the first mapping table D 1 middle;

[0039] Among them, the first mapping table D 1 Including N air quality monitoring sites mapped to the shortest distance mapping coordinates in the grid matrix, Represents the row and column labels when the i-th air quality monitoring station is mapped into the data matrix, i∈[1,N].

[0040] S23: Read the monitoring data of all air quality monitoring stations, and for the same grid point, fill in the default monitoring data at the current moment with the average value of the monitoring data of the monitoring station before and after the monitoring station or the average value of the monitoring data of all moments of the monitoring station;

[0041] In the embodiment of the present invention, the monitoring data of all N air quality monitoring stations are read. If it is found that the monitoring data at the current moment is missing, it is assumed that the monitoring data Default is Then If multiple consecutive default values ​​appear, they will be filled directly with the average value of the site, where It represents the air quality concentration value at the nth station at time t, t-1 represents the moment before time t, and t+1 represents the moment after time t, where t∈(0, L), L is the length of the monitoring data time series.

[0042] S24: According to the coordinates and longitude and latitude distance of the meteorological monitoring station, traverse and calculate the second distance from the meteorological monitoring station to each grid point in the grid matrix, determine the grid point position when the second distance is the shortest, and record it in the second mapping table D 2 middle;

[0043] Among them, the first mapping table D 2 The M meteorological monitoring stations are mapped to the shortest distance mapping coordinates in the grid matrix. Represents the row and column labels when the i-th meteorological monitoring station is mapped into a data matrix, i∈[1,M].

[0044] S25: Reading meteorological monitoring data of all meteorological monitoring stations, for the same grid point, supplementing the default monitoring data at the current moment with the average value of the monitoring data of the monitoring station before and after the moment or the average value of the monitoring data of the monitoring station at all moments;

[0045] In the embodiment of the present invention, the meteorological monitoring data of all M meteorological monitoring stations are read, and for each acquired meteorological data, Default is Then If multiple consecutive default values ​​appear, they will be filled directly with the average value of the site, where represents the meteorological data value of the mth meteorological monitoring station at time t, t-1 represents the moment before time t, and t+1 represents the moment after time t, where t∈(0, L), L is the length of the monitoring data time series.

[0046] In this embodiment, for the missing data part, that is, the meteorological monitoring data, the Cressman filling algorithm is used to optimize the generated data matrix to obtain a new optimized matrix, and the original matrix is ​​overwritten.

[0047] S26: At the current time t, the air quality concentration data of each air quality monitoring station is mapped according to the first mapping table D 1 Mapping to matrix V t At the corresponding position, the humidity, temperature, wind direction and wind speed data of each meteorological monitoring station are mapped according to the second mapping table D 2 Mapping to the matrix Ht 、T t , At the corresponding position;

[0048] Among them, V t Represents the air mass concentration data matrix at time t, H t 、T t , They are the humidity, temperature, wind direction, and wind speed data matrices at time t, respectively, where t∈(0, L), and L is the length of the monitoring data time series.

[0049] In the embodiment of the present invention, for any time t, the PM2.5 data of each air quality monitoring station at time t can be mapped according to the first mapping table D 1 Mapping to matrix V t At the corresponding position, the humidity, temperature, wind direction, and wind speed data of each weather monitoring station at time t are calculated according to the second mapping table D 2 Mapping to the matrix H t 、T t , The corresponding position, where V t It means that the monitoring value of PM2.5 at time t is mapped to the target area divided into fishing nets with a scale of 1km×1km. Similarly, H t 、T t , That is, the monitoring values ​​of humidity, temperature, wind direction and wind speed at time t are mapped to the target area and divided into fishing nets at a scale of 1km×1km.

[0050] S3: Generate a dynamic wind field map using the wind direction data and wind speed data in the meteorological monitoring data in the grid matrix, and calculate the wind field adjacency matrix using the Dijkstra algorithm;

[0051] In the embodiment of the present invention, step S3 may include:

[0052] S31: Using the wind direction data matrix And wind speed data matrix Generate dynamic wind map matrix in, represents the wind field distance from the i-th grid point to the j-th grid point in the grid matrix, where θ ij It represents the wind direction angle from the i-th grid point to the j-th grid point, t∈(0,L), where L is the length of the monitoring data time series;

[0053] S32: According to the dynamic wind field matrix DW t , use Dijkstra's algorithm to calculate the shortest path a from the i-th grid point to the j-th grid point in the grid matrix ij, calculate the shortest path for each grid point in the grid matrix, and form the wind field adjacency matrix A with all the shortest paths t .

[0054] S4: construct a mask matrix at each moment according to the air mass concentration data in the grid matrix, and construct a feature vector set Z at each moment according to the wind field adjacency matrix, the mask matrix and the meteorological monitoring data;

[0055] In the embodiment of the present invention, step S4 includes the following steps:

[0056] S41: Initialize the first mask matrix mask at time t t and the second mask matrix mask2 t , when the value of the air mass concentration data matrix at time t When not empty, otherwise make And from mask2 t Randomly select K grid points with a value of 1 and reset them to 0, and update the second mask matrix, where i and j are the horizontal and vertical coordinates of the mask matrix;

[0057] S42: Extract V and W at time t and each of the previous F times s , W d , H, T, A, mask and mask2 matrices, construct the eigenvector cardinality D of the first F moments before time t f , denoted as D f = [x1, x2, x3, x4, x5, x6, x7];

[0058] S43: construct a feature vector set Z according to the length of the monitoring data time series and the feature vector cardinality, so that Z = [D1, D2, ..., D n , …, D J ];

[0059] Among them, x1, x2, x3, x4, x5, x6, x7 represent the air pollution concentration matrix V*mask2 and wind speed matrix W of the first F time nodes at time t respectively. s , wind direction matrix W d , humidity matrix H, temperature matrix T, wind field adjacency matrix A, mask matrix; f, t, F are integers, f∈[1, J], t∈(0, L), F<L; J represents the dimension of the feature vector set; (n, J)∈(0, LF), L is the length of the monitoring data time series.

[0060] Figure 2 A schematic diagram of generating a mask matrix in an embodiment of the present invention is given, such as Figure 2As shown, in the embodiment of the present invention, it is first necessary to generate a first mask matrix according to the air mass concentration data matrix at time t, wherein the air mass concentration data matrix V at time t t is a multidimensional data, which may be empty or have actual values. The present invention needs to determine the matrix coordinates with actual values, set the matrix coordinates to 1 at the corresponding position of the first mask matrix, and set the corresponding position of the empty matrix to 0, thereby forming the first mask matrix mask t Considering that the first mask matrix is ​​completely determined by the air mass concentration data, its anti-interference ability is weak. Therefore, the present invention partially randomizes the matrix coordinates set to 1 in the first mask matrix, selects some matrix coordinates to be reset to 0, thereby forming a second mask matrix mask2 t , these two mask matrices will play different roles in the subsequent graph convolutional network.

[0061] It can be understood that in the embodiment of the present invention, the first mask matrix is ​​obtained based on the wind field adjacency matrix, and the second mask matrix is ​​obtained based on the first mask matrix and random initialization. The embodiment of the present invention respectively uses the first mask matrix and the second mask matrix before and after random initialization to enhance the robustness of the graph convolutional network model and make the model's gridding effect on air quality more stable.

[0062] Figure 3 A schematic diagram of generating a feature vector set in an embodiment of the present invention is given. Figure 3 As shown, in the embodiment of the present invention, it is first necessary to obtain the air pollution concentration matrix V*mask2 and the wind speed matrix W of the previous F moments of each moment t. s , wind direction matrix W d , humidity matrix H, temperature matrix T, wind field adjacency matrix A, mask matrix, integrate the matrices at each time t to form a set of feature vectors with a length of J.

[0063] S5: Generate a target matrix Y at each moment according to the mask matrix and the air quality concentration data;

[0064] In the embodiment of the present invention, step S5 includes the following steps:

[0065] S51: Generate a target matrix at time t based on the second mask matrix and the air mass concentration data matrix, expressed as Y t =V t mask2 t ;

[0066] S52: Integrate the target matrices at different times to construct a target matrix set Y = [Y1, Y2, ..., Y n , …, YJ ];

[0067] Among them, V t Indicates mapping the air mass concentration data at time t to the grid position of the target area, mask2 t Represents the second mask matrix at time t; t∈(0,L), J represents the dimension of the feature vector set; (n,J)∈(0,LF), L is the length of the monitoring data time series.

[0068] Figure 4 A schematic diagram of generating the target matrix in an embodiment of the present invention is given, such as Figure 4 As shown, in the embodiment of the present invention, it is first necessary to obtain the air mass concentration data matrix V at each time t t and the second mask matrix mask2 t , after dot multiplication of the above two matrices, the resulting matrix is ​​the target matrix Y at time t t .

[0069] S6: Input the feature vector set Z matrix into the trained graph convolutional neural network model to obtain an estimated matrix P of the target matrix.

[0070] Figure 5 The graph convolutional neural network model in the embodiment of the present invention is given, such as Figure 5 As shown, in an embodiment of the present invention, the graph convolutional neural network model includes a temporal convolutional network layer, a spectrum-based graph convolutional network layer, a space-based graph convolutional network layer and a convolutional neural network output layer which are established in sequence; the temporal convolutional network layer extracts the time series characteristics of the air pollution concentration matrix, the wind speed matrix, the wind direction matrix, the humidity matrix and the temperature matrix; the spectrum-based graph convolutional network layer performs noise perturbation denoising on the wind field adjacency matrix; the space-based graph convolutional network layer maps the interaction between the time series characteristics of the temporal convolutional network layer and the wind field characteristics of the spectrum-based graph convolutional network layer; the convolutional neural network output layer processes the interaction relationship and the mask matrix of the space-based graph convolutional network layer, and restores the implicit features to the target matrix.

[0071] In the present invention, the embodiments of the present invention can use TCN to extract the sequence relationship of time features, better characterize the change trend of data in the gridding process, filter the random noise in the gridding process based on spectrum GCN, simulate the influence of wind on air quality pollutants based on space GCN, generate the target matrix by CNN, and fully consider the influence of humidity and temperature on air quality gridding, so that the gridding effect is more accurate. At the same time, the graph convolutional neural network based on the present invention adds random disturbances to the environmental data gridding model during the training process, so that the model is more stable. Compared with the general statistical model, the model of the present invention can effectively capture the spatiotemporal characteristics of the data to be gridded, and the gridding of environmental data is more efficient and stable after the training is completed.

[0072] In an embodiment of the present invention, the training process of the graph convolutional neural network model includes taking the feature vector set Z as the model input, obtaining an estimated matrix P of the target matrix, bitwise multiplying the estimated matrix P of the target matrix with the mask matrix, comparing the bitwise multiplication result with the target matrix Y, and optimizing the graph convolutional neural network model parameters according to the comparison result until the error is within a set range.

[0073] It can be understood that in the embodiment of the present invention, only the training process of the model needs to compare the bitwise multiplication result with the target matrix Y, and optimize the graph convolutional neural network model parameters according to the comparison results. In the application process of the model, that is, when the air quality monitoring data of N air quality monitoring stations and the meteorological monitoring data of M meteorological monitoring stations in the target area are directly processed, it is only necessary to output the estimated matrix P of the target matrix. The estimated matrix P is a dense matrix, and the target matrix is ​​a sparse matrix. In the present invention, the sparse matrix is ​​estimated to complete the grid processing of the environmental data.

[0074] In this example, in step S6, the model for gridding environmental data based on graph convolutional neural network uses MAE as the loss function, and the calculation formula is: The error threshold is set to 5, and the model training stops when the average loss of the model is less than 5.

[0075] In the description of the present invention, it is necessary to understand that the terms "coaxial", "bottom", "one end", "top", "middle", "the other end", "upper", "one side", "top", "inside", "outside", "front", "center", "both ends" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.

[0076] In the present invention, unless otherwise clearly stipulated and limited, the terms such as "installation", "setting", "connection", "fixation" and "rotation" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral one; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium; it can be the internal connection of two elements or the interaction relationship between two elements. Unless otherwise clearly defined, ordinary technicians in this field can understand the specific meanings of the above terms in the present invention according to the specific circumstances.

[0077] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A grid processing method for environmental data based on graph convolutional network, characterized in that: The steps include: S1: Acquire air quality monitoring data of N air quality monitoring stations and meteorological monitoring data of M meteorological monitoring stations in the target area; the air quality monitoring data includes air quality concentration data, and the meteorological monitoring data includes humidity, temperature, wind direction and wind speed data; S2: Preprocess the missing parts of all monitoring data, and map the monitoring data of the corresponding stations to the grid matrix divided by scale in the target area according to the coordinates of the monitoring stations; S3: Generate a dynamic wind field map using the wind direction data and wind speed data in the meteorological monitoring data in the grid matrix, and calculate the wind field adjacency matrix using the Dijkstra algorithm; S4: construct a mask matrix at each moment according to the air mass concentration data in the grid matrix, and construct a feature vector set Z at each moment according to the wind field adjacency matrix, the mask matrix and the meteorological monitoring data; S5: Generate a target matrix Y at each moment according to the mask matrix and the air quality concentration data; S6: Input the feature vector set Z matrix into the trained graph convolutional neural network model to obtain an estimated matrix P of the target matrix; The graph convolutional neural network model in step S6 includes a temporal convolutional network layer, a spectrum-based graph convolutional network layer, a space-based graph convolutional network layer and a convolutional neural network output layer which are sequentially established; the temporal convolutional network layer extracts the time series features of the air pollution concentration matrix, the wind speed matrix, the wind direction matrix, the humidity matrix and the temperature matrix; the spectrum-based graph convolutional network layer performs noise disturbance denoising on the wind field adjacency matrix; the space-based graph convolutional network layer maps the interaction between the time series features of the temporal convolutional network layer and the wind field features of the spectrum-based graph convolutional network layer; the convolutional neural network output layer processes the interaction and mask matrix of the space-based graph convolutional network layer and restores the implicit features to the target matrix; Among them, the training process of the graph convolutional neural network model includes taking the feature vector set Z as the model input, obtaining the estimated matrix P of the target matrix, multiplying the estimated matrix P of the target matrix with the mask matrix in bitwise manner, comparing the bitwise multiplication result with the target matrix Y, and optimizing the graph convolutional neural network model parameters according to the comparison result until the error is within the set range.

2. According to claim 1, a grid point processing method for environmental data based on graph convolutional network is characterized in that: The step S1 includes respectively obtaining the air quality concentration data of the air quality monitoring site at each moment in the first frequency period and the humidity, temperature, wind direction and wind speed data of the meteorological monitoring site in the first frequency period; sorting the corresponding air quality monitoring data and meteorological monitoring data in chronological order, and aligning the corresponding monitoring data by time.

3. The grid point processing method for environmental data based on graph convolutional network according to claim 2 is characterized in that: The step S2 comprises the following steps: S21: Obtain the boundary of the target area, divide the target area into fishing grid points or grids according to a certain scale, and obtain the coordinates of the center point of each grid. ij , forming the center point coordinate matrix P, i.e., the grid matrix, where i and j are the row and column coordinates of the grid matrix respectively; S22: According to the coordinates and longitude and latitude distance of the air quality monitoring station, traverse and calculate the first distance from the air quality monitoring station to each grid point in the grid matrix, determine the grid point position when the first distance is the shortest, and record it in the first mapping table D 1 middle; S23: Read the monitoring data of all air quality monitoring stations, and for the same grid point, fill in the default monitoring data at the current moment with the average value of the monitoring data of the monitoring station before and after the monitoring station or the average value of the monitoring data of all moments of the monitoring station; S24: According to the coordinates and longitude and latitude distance of the meteorological monitoring station, traverse and calculate the second distance from the meteorological monitoring station to each grid point in the grid matrix, determine the grid point position when the second distance is the shortest, and record it in the second mapping table D 2 middle; S25: Reading meteorological monitoring data of all meteorological monitoring stations, for the same grid point, supplementing the default monitoring data at the current moment with the average value of the monitoring data of the monitoring station before and after the moment or the average value of the monitoring data of the monitoring station at all moments; S26: At the current time t, the air quality concentration data of each air quality monitoring station is mapped according to the first mapping table D 1 Mapping to matrix V t At the corresponding position, the humidity, temperature, wind direction and wind speed data of each meteorological monitoring station are mapped according to the second mapping table D 2 Mapping to matrix H t 、T t , At the corresponding position; Among them, V t Represents the air mass concentration data matrix at time t, H t , T t , They are the humidity, temperature, wind direction, and wind speed data matrices at time t, respectively, where t∈(0, L), and L is the length of the monitoring data time series.

4. The grid point processing method for environmental data based on graph convolutional network according to claim 3 is characterized in that: The step S3 comprises the following steps: S31: Using the wind direction data matrix And wind speed data matrix Generate dynamic wind field matrix DW t , in, Represents the wind field distance from the i-th grid point to the j-th grid point in the grid matrix, θ ij represents the wind direction angle from the i-th grid point to the j-th grid point, Represents the wind speed from the i-th grid point to the j-th grid point in the wind speed data matrix at time t, t∈(0,L), where L is the length of the monitoring data time series; S32: According to the dynamic wind field matrix DW t , use Dijkstra's algorithm to calculate the shortest path a from the i-th grid point to the j-th grid point in the grid matrix ij , calculate the shortest path for each grid point in the grid matrix, and form the wind field adjacency matrix A with all the shortest paths t .

5. The grid point processing method for environmental data based on graph convolutional network according to claim 4 is characterized in that: The step S4 comprises the following steps: S41: Initialize the first mask matrix mask at time t t and the second mask matrix mask2 t , when the value of the air mass concentration data matrix at time t When not empty, otherwise make And from mask2 t Randomly select K grid points with a value of 1 and reset them to 0, and update the second mask matrix; S42: Extract V and W at time t and each of the previous F times s , W d , H, T, A, mask and mask2 matrices, construct the eigenvector cardinality D of the first F moments before time t f , denoted as D f = [x1, x2, x3, x4, x5, x6, x7]; S43: construct a feature vector set Z according to the length of the monitoring data time series and the feature vector cardinality, so that Z = [D1, D2, ..., D n , …, D J ]; Among them, x1, x2, x3, x4, x5, x6, x7 represent the air pollution concentration matrix V*mask2 and wind speed matrix W of the first F time nodes at time t respectively. s , wind direction matrix W d , humidity matrix H, temperature matrix T, wind field adjacency matrix A, mask matrix; f, t, F are integers, f∈[1, J], t∈(0, L), F<L; J represents the dimension of the feature vector set; (n, J)∈(0, LF), L is the length of the monitoring data time series.

6. The grid point processing method for environmental data based on graph convolutional network according to claim 5, characterized in that: The step S5 comprises the following steps: S51: Generate a target matrix at time t based on the second mask matrix and the air mass concentration data matrix, expressed as Y t =V t mask2 t ; S52: Integrate the target matrices at different times to construct a target matrix set Y = [Y1, Y2, ..., Y n , …, Y J ]; Among them, V t Indicates mapping the air mass concentration data at time t to the grid position of the target area, mask2 t Represents the second mask matrix at time t; t∈(0,L), J represents the dimension of the feature vector set; (n,J)∈(0,LF), L is the length of the monitoring data time series.

Images