Air pollutant concentration interpolation prediction method and device based on GAI-BRI and medium

Abstract

The invention relates to a GAI-BRI-based air pollutant concentration interpolation prediction method and device and a medium, and the method comprises the steps: carrying out the interpolation prediction of the concentration of air pollutants according to the extracted air pollutant concentration, meteorological data and satellite data of a plurality of distribution sites in a given region within a given time range; constructing an air pollutant concentration diagram for describing pollutant conditions in a given time range and a given area range; inputting the numerical value of the air pollutant concentration diagram into a GAT-BRI prediction model to obtain an air pollutant concentration interpolation prediction result in a period of time in the future; wherein the GAT-BRI prediction model comprises a graph attention network GAT which is arranged at a bottom layer and is used for extracting spatial features from an air pollutant concentration graph, and a bidirectional cyclic interpolation network BRI which is arranged at a middle layer and is used for extracting time sequence features from the spatial features output by the graph attention network; and the feed-forward network is arranged at a high layer and is used for outputting an interpolation prediction result. Compared with the prior art, the method has the advantages that the spatial dependence and time dependence relationship among multiple cities can be fully mined, and the prediction accuracy is high.

Description

Technical Field

[0001] The present invention relates to the field of air pollutant concentration prediction, and in particular to a method, device and medium for interpolation prediction of air pollutant concentration based on GAI-BRI. Background Art

[0002] In recent years, air pollution has brought many impacts on the environment and human health, and this problem has attracted widespread attention from all walks of life. It is very important to predict the concentration of air pollution in advance and announce it to the public in a timely manner. If the prediction time can be extended while ensuring accuracy, people can do early warning and prevention work earlier, which can effectively reduce the adverse effects of air pollution on people and the environment. Using big data thinking combined with deep learning technology to conduct effective data analysis on air pollution and then make accurate predictions is a frontier hot issue in the current interdisciplinary research of environmental science and computer science.

[0003] The interpolation method of air pollutant concentration is a method that uses existing air pollutant monitoring data to predict, estimate or fill in the air pollutant concentration values ​​during unmonitored time. The commonly used method is linear interpolation. With the deepening of research, machine learning methods have been applied to interpolation work. Compared with traditional statistical methods, machine learning methods can adaptively process nonlinear and high-dimensional data and can better capture the correlation between data. With the widespread application of deep learning methods in prediction tasks, the prediction effect of air pollutant concentration has been significantly improved. Therefore, the use of prediction methods to interpolate missing values ​​in real data sets has attracted widespread attention from researchers. Compared with the traditional GAI-BRI-based air pollutant concentration interpolation prediction method, the deep learning method has stronger nonlinear modeling capabilities.

[0004] Although traditional existing interpolation methods for urban air pollution concentration can achieve certain interpolation effects, they all face the following problems: they ignore the temporal changes in air pollutant concentration data. If there is a large area of ​​missing data, these interpolation methods will fill in a large number of identical values, resulting in large data deviations, which in turn affects the prediction results; they rely on numerical models to learn the statistical characteristics of the data and do not have the ability to jointly extract the spatiotemporal correlation of air pollutant concentrations; they only focus on one aspect of the complex spatiotemporal correlation of air pollutant concentrations, and lack research on combining it with models suitable for extracting spatial correlations of data. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a GAI-BRI-based air pollutant concentration interpolation prediction method, device and medium with higher prediction accuracy.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to a first aspect of the present invention, there is provided a method for interpolation prediction of air pollutant concentration based on GAI-BRI, comprising:

[0008] Based on the extracted air pollutant concentrations, meteorological data and satellite data of multiple distribution stations in a given area within a given time range, an air pollutant concentration map describing the pollutant conditions within a given area within a given time range is constructed;

[0009] The numerical values ​​of the air pollutant concentration map are input into the GAT-BRI prediction model to obtain the interpolation prediction results of the air pollutant concentration for a period of time in the future; wherein, the GAT-BRI prediction model includes a graph attention network GAT set at the bottom layer for extracting spatial features from the air pollutant concentration map, a bidirectional recurrent interpolation network BRI set at the middle layer for extracting temporal features from the spatial features output by the graph attention network, and a feedforward network set at the high layer for outputting the interpolation prediction results.

[0010] Preferably, the pollutants involved in the air pollutant concentration include PM2.5, PM10, SO2, NO2, O3 and CO; the meteorological factors involved in the meteorological data include temperature, humidity, wind speed and air pressure; and the satellite data include boundary layer height, shortwave radiation, sunshine hours, specific humidity and sensible heat flux.

[0011] Preferably, the step of constructing a spatiotemporal graph describing the pollutant situation in a given area within a given time range and constructing an air pollutant concentration graph describing the pollutant situation in a given area within a given time range specifically includes:

[0012] Use the distance between regions as the basis for the existence of edges between nodes to initialize the adjacency matrix;

[0013] Use the attention mechanism in the graph attention network to learn the correlation weights between cities;

[0014] Based on the data collected at each moment from all distributed stations in the region, a spatiotemporal diagram is constructed to describe the pollutant situation within a given area within a given time range.

[0015] Preferably, the graph attention network arranged at the bottom layer for extracting spatial features from the air pollutant concentration map updates the node characteristics by assigning different attention coefficients to adjacent nodes, thereby extracting the nonlinear spatial dependency of pollutant propagation, specifically including:

[0016] Input the feature vector in the air pollutant concentration map into the graph attention network;

[0017] The feature vectors of the target node and each of its adjacent nodes are concatenated and input into a feedforward network to obtain the attention weights between nodes;

[0018] The feature vectors of all neighboring nodes of the target node are weighted summed with the corresponding attention weights to obtain the aggregated feature vector;

[0019] The aggregated feature vectors are transformed nonlinearly to obtain the spatial characteristics of air pollutant concentrations.

[0020] Preferably, the graph attention network uses attention scores between nodes to characterize the magnitude of the influence of air pollutant concentrations between regions. The specific network training process includes:

[0021] The self-attention mechanism is used to calculate the similarity coefficient e between the target node i and its adjacent node j. ij ; Among them, the self-attention mechanism introduces the LeakyReLU activation function;

[0022] The similarity coefficient e between the target city node i and its adjacent city node j ij Normalization processing is performed to obtain the normalized attention score a of the target city node i and its adjacent city node j ij ,;

[0023] Aggregate the information of all adjacent nodes of target node i to obtain the node feature h of target node i i :

[0024] The feature vector of all neighboring nodes of target node i is weighted and summed with the corresponding attention weights to obtain the feature vector

[0025] Preferably, the graph attention network adopts a multi-head attention mechanism to capture different features through different graph attention layers, and then connects multiple captured features to obtain a multi-head attention calculation result.

[0026] Preferably, a tensor transposition is performed on the spatial features of the air pollutant concentration map extracted by the graph attention network GAT to obtain the first data, the spatial features of the air pollutant concentration map extracted by the graph attention network GAT are concatenated with the original data to obtain the second data, the first data and the second data are jointly input into a bidirectional recurrent interpolation network, and the time dependency of the propagation and diffusion of air pollutants is extracted through a gating mechanism to obtain the time characteristics of the air pollutant concentration.

[0027] Preferably, the use of a bidirectional cyclic interpolation network to extract the temporal characteristics of air pollutant concentrations includes two stages of extraction, specifically:

[0028] The first stage is extraction, judging the current input data X tIs there any missing value? If not, input the current time into data X. t As the filling value, input the bidirectional recurrent interpolation network, otherwise the estimated value of the previous moment As a fill value, it is input into the bidirectional loop interpolation network to obtain a complete first estimate based on historical observations. and fill value τ is the time window length;

[0029] The second stage extracts the fill value obtained in the first stage. As input, a second estimate based on other features is calculated by setting the diagonal of the parameter matrix to 0.

[0030] The first estimate and the second estimate are weighted and summed to obtain the combined vector

[0031] The current moment data X t+1 If it does not exist, the combined vector is used Fill in the next moment data X t+1 The value of the first vector

[0032] The first vector Input into the bidirectional recurrent interpolation network to extract the time information in two directions and obtain the hidden vector {h t ,h t+1 ,h t+2 , ..., h t+τ};

[0033] According to the hidden vector {h t ,h t+1 ,h t+2 , ..., h t+τ}, which is used to obtain the predicted value of air pollutant concentration of the target city in the final given area through a feedforward network.

[0034] According to a second aspect of the present invention, there is provided an electronic device, comprising a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the program, any one of the methods described above is implemented.

[0035] According to a third aspect of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, any one of the methods described above is implemented.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] (1) The present invention makes predictions based on the air pollutant concentration data and meteorological data of multiple cities, constructs a GAT-BRI prediction model, and uses a graph neural network at the bottom layer to extract the spatial dependencies of the graph structure composed of multiple cities. It can make full use of the attention mechanism to obtain the influence between multiple cities and mine the spatial dependencies between multiple cities. Then, a bidirectional cyclic interpolation network is used to fully extract the spatial features obtained by the graph neural network, mine spatiotemporal information, and finally output the prediction results through a feedforward network, which can realize the refined assignment of real air pollutant concentration data.

[0038] (2) The bidirectional recurrent interpolation network uses a bidirectional long short-term memory network, which can describe the degree of change of the input latent vector. When the input data changes drastically, the memory unit can be reset in time. It can fully mine the spatiotemporal information in the air pollutant concentration data in a bidirectional manner, thereby improving the accuracy of pollutant concentration prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a flow chart of the steps of the present invention;

[0040] Figure 2 It is a schematic diagram of a flow chart in an embodiment of the present invention;

[0041] Figure 3 This is the air pollutant concentration interpolation model proposed in the present invention. DETAILED DESCRIPTION

[0042] 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 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 should fall within the scope of protection of the present invention.

[0043] Example

[0044] First, define the interpolation of air pollutant concentration:

[0045] Air pollutant concentration interpolation: It is a method to predict, estimate or fill in the concentration values ​​of air pollutants such as PM2.5 and PM10 in the future for a certain period of time based on historical pollutant and meteorological information. It is one of the key research topics in environmental science, meteorological science, computer science, etc., and therefore has a certain degree of interdisciplinary nature.

[0046] Traditional GAI-BRI-based interpolation prediction method for air pollutant concentration: Non-deep learning-based GAI-BRI-based interpolation prediction method for air pollutant concentration. Most of the traditional air pollutant concentration interpolation models mainly include: linear interpolation model, sliding average data filling model, previous value interpolation model, traditional machine learning model based on small sample data, etc., all of which are traditional prediction models.

[0047] like Figures 1 to 3 As shown, this embodiment provides an air pollutant concentration interpolation prediction method based on GAI-BRI, including:

[0048] Based on the extracted air pollutant concentrations, meteorological data and satellite data of multiple distribution stations in a given area within a given time range, an air pollutant concentration map describing the pollutant conditions within a given area within a given time range is constructed;

[0049] The numerical values ​​of the air pollutant concentration map are input into the GAT-BRI prediction model to obtain the interpolation prediction results of the air pollutant concentration for a period of time in the future; wherein, the GAT-BRI prediction model includes a graph attention network GAT set at the bottom layer for extracting spatial features from the air pollutant concentration map, a bidirectional recurrent interpolation network BRI set at the middle layer for extracting temporal features from the spatial features output by the graph attention network, and a feedforward network set at the high layer for outputting the prediction results.

[0050] Next, the method of this embodiment is introduced in detail.

[0051] 1. Data acquisition

[0052] In this embodiment, the following data are selected from all detection sites in the prediction area:

[0053] The pollutants involved in the air pollutant concentration include PM2.5, PM10, SO2, NO2, O3 and CO. The data comes from the National Urban Air Quality Real-time Release Platform of the China National Environmental Monitoring Center. The air pollutant data collected by the ground monitoring station is collected every 1 hour.

[0054] Meteorological factors involved in meteorological data include temperature, humidity, wind speed and air pressure, and the data comes from the National Climatic Data Center (NCDC) of the United States;

[0055] Satellite data include boundary layer height, shortwave radiation, sunshine duration, specific humidity and sensible heat flux, which are selected through impact experiments and come from the European Centre for Medium-Range Weather Forecasts.

[0056] Specifically, taking PM2.5 as an example, the linear and nonlinear correlations between PM2.5 and satellite features are extracted from the data set, and the Pearson correlation coefficient is used to calculate the correlation:

[0057]

[0058] Among them: X represents PM2.5 data, represents the mean of PM2.5 data, Y represents different satellite data, represents the mean of satellite data, t represents the observed values ​​of samples X and Y at the corresponding time t, and n represents the number of samples;

[0059] The correlation is then calculated using the Spearman rank correlation coefficient:

[0060]

[0061] Where: d i For X i and Y i The grade difference;

[0062] The above 15 features are determined to be used to predict PM2.5 concentration values, and then the data is normalized. The Z-score normalization method is selected for normalization, and the calculation expression is:

[0063]

[0064] in: represents the normalized data, μ is the mean of the original data, δ is the standard deviation of the original data, and x t Represents the normalized number, which follows a standard normal distribution with a mean of 0 and a variance of 1.

[0065] 2. GAT-BRI prediction model

[0066] A GAT-BRI prediction model was constructed based on the extracted data to interpolate missing values ​​in the real air pollutant concentration dataset.

[0067] The GAT-BRI prediction model includes a graph attention network GAT set at the bottom layer to extract spatial features from the air pollutant concentration map, a bidirectional recurrent interpolation network BRI set at the middle layer to extract temporal features from the spatial features output by the graph attention network, and a feedforward network set at the high layer to output the prediction results.

[0068] 2.1 Graph Attention Network GAT

[0069] like Figure 3As shown in the figure, the input of the graph attention network GAT is the feature vector obtained by modeling the air pollutant concentration graph data. The feature vectors of the target node and each of its neighboring nodes are concatenated and passed through a feedforward network to calculate the attention weights between the nodes. The feature vectors of all neighboring nodes of the target node are weighted and summed with the corresponding attention weights to obtain the aggregated feature vector. Finally, the aggregated feature vector is transformed through a nonlinear transformation to obtain a new feature vector. In this way, GAT can extract the spatial information of the air pollutant concentration graph structure data, and can also adaptively calculate the attention weights of each node and its neighboring nodes.

[0070] The output of the graph attention network GAT contains an implicit graph representation of neighbor node information, which is jointly input into the BRI part with the original input data through tensor transposition, splicing and other operations, and the time dependence of pollutant propagation and diffusion is extracted through the gating mechanism. Specifically, the spatial features of the air pollutant concentration map extracted by the graph attention network GAT are tensor transposed to obtain the first data, the spatial features of the air pollutant concentration map extracted by the graph attention network GAT are spliced ​​with the original data to obtain the second data, and the first data and the second data are jointly input into the BRI part. The original input data is the air pollutant concentration, meteorological data and satellite data of multiple distributed stations in a given area within the original given time range.

[0071] The Graph Attention Network (GAT) part is responsible for learning the representation information of nodes and extracting the spatial dependencies of pollutant concentrations in the graph;

[0072] The training of GAT first calculates the attention coefficient. For the target city i, the similarity coefficients of the surrounding cities with i are calculated one by one:

[0073]

[0074] Among them: attention represents the calculation method of self-attention, vector h is the feature vector of the node, and its subscripts i and j represent different nodes respectively.

[0075] The softmax function is used to normalize node i and its adjacent node j. The value of the attention score reflects the strength of the relationship between the two nodes, that is, the influence of the concentration of air pollutants between cities:

[0076]

[0077] The nonlinear activation function LeakyReLU function is introduced into the attention mechanism, and finally the final attention mechanism calculation method is obtained. The specific operation in GAT is to concatenate the feature vectors of the two nodes together and then do the inner product with the weight vector:

[0078]

[0079] Among them: || represents the splicing operation, is the weight vector.

[0080] After obtaining the normalized attention coefficients between different nodes, the information of adjacent nodes is aggregated, and finally the node features of each node after GAT calculation are obtained:

[0081]

[0082] The aggregated feature vector is obtained by weighting the feature vectors of all neighbor nodes of node i with the corresponding attention weights.

[0083] The final result of multi-head attention is as follows:

[0084]

[0085] Among them: K means that K attention calculations need to be performed, k means the calculation of the kth attention, and the kth attention is a k , the linear transformation weight of the feature under the kth attention mechanism is W k .

[0086] 2.2 Bidirectional Circular Interpolation Network

[0087] Traditional recurrent neural network: h t =σ(W h h t-1 +U h X t +b h ), where σ is the sigmoid function, W h , U h and b h are the network parameters and ht is the hidden state at the current time step.

[0088] In this embodiment, a bidirectional long short-term memory network Bi-LSTM is selected using a bidirectional cyclic interpolation network, and the time characteristics of air pollutant concentration are extracted in two stages, specifically:

[0089] The first stage is extraction, judging the current input data X t Is there any missing value? If not, input the current time into data X. t As the filling value, input the bidirectional recurrent interpolation network, otherwise the estimated value of the previous moment As a fill value, it is input into the bidirectional loop interpolation network to obtain a complete first estimate based on historical observations. and fill value τ is the time window length;

[0090] The second stage extracts the fill value obtained in the first stage. As input, a second estimate based on other features is calculated by setting the diagonal of the parameter matrix to 0. in, W z is the parameter matrix, b z is the bias value;

[0091] The first estimate and the second estimate are weighted and summed to obtain the combined vector The specific calculation expression is:

[0092]

[0093]

[0094] Where: β t ∈[0, 1] D As a combination and The weight of It can be either an observed value or a predicted value;

[0095] The current moment data X t+1 If it does not exist, the combined vector is used Fill in the next moment data X t+1 The value of the first vector

[0096] The first vector Input into the bidirectional recurrent interpolation network to extract the time information in two directions and obtain the hidden vector {h t ,h t+1 ,h t+2 , ..., h t+τ}, the calculation expression is:

[0097]

[0098]

[0099] in: As the final fill value and can be used to fill X t+1 The value of h t is the hidden state at time t;

[0100] According to the hidden vector {h t ,h t+1 ,h t+2 , ..., h t+τ}, which is used to obtain the predicted value of air pollutant concentration of the target city in the final given area through a feedforward network.

[0101] 2.3 Model Training

[0102] The model training process specifically includes:

[0103] 1) Input the air pollutant concentration data after feature extraction by the graph neural network GAT:

[0104] X={X1,X2,…,X T};

[0105] 2) Convert the data dimension into a time series distributed according to the time dimension and merge it with the original data as the input of the bidirectional cyclic interpolation network. Use the bidirectional long short time neural network Bi-LSTM to perform stage prediction on the input data. Specifically: As a variable, use Bi-LSTM to reverse the above formula on the input sequence to get the forward estimated sequence and the reverse estimated series and the forward loss sequence {l1, l2, ..., l t} and the reverse loss sequence

[0106] 3) Introduce consistency loss to enforce that the predictions at each stage are consistent in both directions:

[0107]

[0108] 4) The training goal is to minimize the reconstruction error of all stages in both directions. The loss function is as follows:

[0109]

[0110] in, is the reconstruction error in two directions.

[0111] For the graph attention network (GAT), it is set to 2 layers, and the feature vectors of the nodes after processing in the first and second layers are both set to 10; for the bidirectional long short-term memory network (Bi-LSTM), it is set to 2 layers, and the number of hidden neurons is set to 64; the fully connected layer is set to 2 layers, the learning rate is set between 0.001-0.1, and the number of iterations is 200.

[0112] The loss function in the model training phase is defined as the root mean square error function, and the back propagation algorithm is used to minimize the value of the model loss function and reduce the error between the predicted value and the true value.

[0113] The root mean square error RMSE is calculated as follows:

[0114]

[0115] After training, the model hyperparameters are adjusted using the data in the validation set until the model is optimal. For the test set, it is input into the model to obtain the prediction results.

[0116] Using mean square error SE, mean absolute error MAE, and coefficient of determination R 2 To measure the model error and prediction performance, the calculation expressions are:

[0117]

[0118]

[0119]

[0120] Where: Y t represents the true value, y t represents the predicted value, Y t represents the average value of the true value, and n is the test sample size.

[0121] Combining these four commonly used air pollutant concentration evaluation indicators, if the above four indicators are better than the traditional methods, it means that the model meets expectations and the performance of the proposed model can be evaluated from multiple angles.

[0122] The verified spatiotemporal joint interpolation model is used to refine the actual air pollutant concentration data.

[0123] In summary, the prediction model based on graph attention network-bidirectional recurrent interpolation network designed in the present invention can accurately and effectively predict the PM2.5 concentration in the air within a certain period of time in the future.

[0124] Aiming at the shortcomings of the existing GAI-BRI-based air pollutant concentration interpolation prediction method, the present invention makes full use of various spatiotemporal big data and existing research results, and uses the graph attention network (GAT) and the improved long short-term memory network (Bi-LSTM) to construct a GAT-BRI spatiotemporal prediction model.

[0125] The spatiotemporal graph sequence is constructed by considering the spatial distance and feature similarity between graph nodes, which can exclude some interfering neighbor nodes while retaining valid neighbors.

[0126] With the Graph Attention Network (GAT) as the bottom layer, different attention coefficients can be assigned to adjacent nodes to update the node's own features, thereby extracting the nonlinear spatial dependency of pollutant propagation; the output of the GAT network is used as the input of the high-level BiLSTM network to extract the time series characteristics of air pollutant data.

[0127] The improved BiLSTM unit can describe the degree of change of the input latent vector and reset the memory unit in time when the input data changes drastically.

[0128] In terms of the time series prediction model, a bidirectional recurrent interpolation network is used as a feature extractor, which fully utilizes the time series feature acquisition capability of the recurrent neural network, fully extracts the information obtained by the graph attention network GAT in both directions, and mines spatiotemporal information.

[0129] The experimental results show that compared with the existing methods, the present invention obtains more accurate prediction results and can perform refined assignment of real air pollutant concentration data, thus having practical application value and prospects.

[0130] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0131] Multiple components in the device are connected to the I / O interface, including: input units, such as keyboards, mice, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunication networks.

[0132] The processing unit performs the various methods and processes described above. For example, in some embodiments, the method can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed on the device via a ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the method described above can be performed. Alternatively, in other embodiments, the CPU can be configured to perform the method in any other appropriate manner (e.g., by means of firmware).

[0133] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip systems (SOCs), load programmable logic devices (CPLDs), and the like.

[0134] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code can be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0135] In the context of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0136] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for interpolation prediction of air pollutant concentration based on GAI-BRI, characterized in that: include: Based on the extracted air pollutant concentrations, meteorological data and satellite data of multiple distribution stations in a given area within a given time range, an air pollutant concentration map describing the pollutant conditions within a given area within a given time range is constructed; The numerical values ​​of the air pollutant concentration map are input into the GAT-BRI prediction model to obtain the interpolation prediction results of the air pollutant concentration for a period of time in the future; wherein, the GAT-BRI prediction model includes a graph attention network GAT set at the bottom layer for extracting spatial features from the air pollutant concentration map, a bidirectional recurrent interpolation network BRI set at the middle layer for extracting temporal features from the spatial features output by the graph attention network, and a feedforward network set at the high layer for outputting the interpolation prediction results.

2. The method for interpolation prediction of air pollutant concentration based on GAI-BRI according to claim 1 is characterized in that: The pollutants involved in the air pollutant concentration include PM2.5, PM10, SO2, NO2, O3 and CO; the meteorological factors involved in the meteorological data include temperature, humidity, wind speed and air pressure; the satellite data include boundary layer height, shortwave radiation, sunshine hours, specific humidity and sensible heat flux.

3. The method for interpolation prediction of air pollutant concentration based on GAI-BRI according to claim 1 is characterized in that: The step of constructing a spatiotemporal graph describing the pollutant situation within a given area within a given time range and constructing an air pollutant concentration graph describing the pollutant situation within a given area within a given time range specifically includes: Use the distance between regions as the basis for the existence of edges between nodes to initialize the adjacency matrix; Use the attention mechanism in the graph attention network to learn the correlation weights between cities; Based on the data collected at each moment from all distributed stations in the region, a spatiotemporal diagram is constructed to describe the pollutant situation within a given area within a given time range.

4. The method for interpolation prediction of air pollutant concentration based on GAI-BRI according to claim 1 is characterized in that: The graph attention network set at the bottom layer is used to extract spatial features from the air pollutant concentration map. By assigning different attention coefficients to adjacent nodes to update the node's own features, the nonlinear spatial dependency of pollutant propagation is extracted, specifically including: Input the feature vector in the air pollutant concentration map into the graph attention network; The feature vectors of the target node and each of its adjacent nodes are concatenated and input into a feedforward network to obtain the attention weights between nodes; The feature vectors of all neighboring nodes of the target node are weighted summed with the corresponding attention weights to obtain the aggregated feature vector; The aggregated feature vectors are transformed nonlinearly to obtain the spatial characteristics of air pollutant concentrations.

5. The method for interpolation prediction of air pollutant concentration based on GAI-BRI according to claim 4 is characterized in that: The graph attention network uses the attention scores between nodes to represent the influence of air pollutant concentrations between regions. The specific network training process includes: The self-attention mechanism is used to calculate the similarity coefficient e between the target node i and its adjacent node j. ij ; Among them, the self-attention mechanism introduces the LeakyReLU activation function; The similarity coefficient e between the target city node i and its adjacent city node j ij Normalization processing is performed to obtain the normalized attention score a of the target city node i and its adjacent city node j ij ,; Aggregate the information of all adjacent nodes of target node i to obtain the node feature h of target node i i : The feature vector of all neighboring nodes of target node i is weighted and summed with the corresponding attention weights to obtain the feature vector 6. The method for interpolation prediction of air pollutant concentration based on GAI-BRI according to claim 5 is characterized in that: The graph attention network adopts a multi-head attention mechanism to capture different features through different graph attention layers, and then connects multiple captured features to obtain the multi-head attention calculation results.

7. The method for interpolation prediction of air pollutant concentration based on GAI-BRI according to claim 1 is characterized in that: The bidirectional recurrent interpolation network set in the middle layer for extracting temporal features from the spatial features output by the graph attention network specifically includes: The spatial features of the air pollutant concentration map extracted by the graph attention network GAT are tensor-transposed to obtain the first data, the spatial features of the air pollutant concentration map extracted by the graph attention network GAT are concatenated with the original data to obtain the second data, the first data and the second data are jointly input into a bidirectional recurrent interpolation network, and the time dependence of the propagation and diffusion of air pollutants is extracted through a gating mechanism to obtain the time characteristics of the air pollutant concentration.

8. The method for interpolation prediction of air pollutant concentration based on GAI-BRI according to claim 7 is characterized in that: The use of a bidirectional cyclic interpolation network to extract the temporal characteristics of air pollutant concentrations includes two stages of extraction, specifically: The first stage is extraction, judging the current input data X t Is there any missing value? If not, input the current time into data X. t As the filling value, input the bidirectional recurrent interpolation network, otherwise the estimated value of the previous moment As a fill value, it is input into the bidirectional loop interpolation network to obtain a complete first estimate based on historical observations. and fill value τ is the time window length; The second stage extracts the fill value obtained in the first stage. As input, a second estimate based on other features is calculated by setting the diagonal of the parameter matrix to 0. The first estimate and the second estimate are weighted and summed to obtain the combined vector The current moment data X t+1 If it does not exist, the combined vector is used Fill in the next moment data X t+1 The value of the first vector The first vector Input into the bidirectional recurrent interpolation network to extract the time information in two directions and obtain the hidden vector {h t ,h t+1 ,h t+2 ,...,h t+τ }; According to the hidden vector {h t ,h t+1 ,h t+2 ,...,h t+τ }, which is used to obtain the predicted value of air pollutant concentration of the target city in the final given area through a feedforward network.

9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images