Method for reconstructing subsurface element field data based on deep network and scattered observation data and application

Abstract

The invention discloses a method for reconstructing subsurface element field data based on a deep network and scattered observation data, belongs to the technical field of marine environment three-dimensional monitoring, and particularly aims to solve the problem that a traditional complex deep network cannot utilize a scattered field observation data set as label data in the background technology. The subsurface temperature and salinity data can be reconstructed by taking a sea surface element framework observed by a satellite as a factor and taking field observation profile data as a label. The method comprises the following main parts: (1) a deep neural network part which predicts sub-surface element field data by inputting a sea surface element field; and (2) a cost function part of a bilinear interpolation operator is used, the bilinear interpolation operator is used in the part, the predicted subsurface element field corresponds to observation data set labels at scattered positions, and the purpose of carrying out subsurface element field data reconstruction by comprehensively utilizing the multi-source ocean data is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of marine environment stereoscopic monitoring, and in particular relates to a method for reconstructing subsurface element field data based on a deep network and scattered observation data. Background Art

[0002] Ocean data is an essential support for the development of modern oceanography and the protection of the marine environment. The observation of ocean data is mainly divided into on-site observation and remote sensing observation. On-site observation data can provide vertical profiles of ocean temperature, salinity and other elements, but the existing on-site observations are still relatively sparse. Even though the development of the Argo observation network has enabled all-weather large-scale observations of the marine environment, it can only provide an average of 10 profiles within a range of 3°×3° every 10 days, while the temporal and spatial resolution of at least 0.25°×0.25° / week is required to reflect the mesoscale phenomena in the ocean. Therefore, except for some targeted marine surveys of local small and medium-scale phenomena, on-site observations cannot effectively reflect the small and medium-scale phenomena that are common in the ocean. Remote sensing observations can provide ocean surface element fields such as sea surface temperature and sea surface height. The existing remote sensing observation data generally reach a temporal and spatial resolution of 0.25° / 1 day, which can effectively identify mesoscale phenomena such as ocean fronts and mesoscale vortices. However, remote sensing observations are limited to the surface of the ocean and cannot effectively reflect the marine environmental information below the sea surface. Overall, observation data from any single source cannot reflect the three-dimensional structure of the ocean.

[0003] The technology of reconstructing high-precision three-dimensional ocean data by comprehensively utilizing multi-source data is called ocean three-dimensional data reconstruction (hereinafter referred to as reconstruction). Reconstruction maps a small amount of ocean information to the entire ocean three-dimensional space, which is a typical inverse problem and has inherent ill-posedness. To address this problem, researchers mostly use empirical methods to model, among which the most widely used technical route is statistical regression. The core idea of ​​statistical regression is to establish a statistical regression relationship between ocean surface elements such as sea surface temperature and sea surface height observed by satellites and underwater field observation profiles. Based on this idea, predecessors established a statistical relationship between sea surface temperature, sea surface height and field observed temperature and salinity observation profiles through linear regression methods, realizing the surface-underwater reconstruction of ocean temperature and salinity data. This technical route is applied to the famous Modular Ocean Data Assimilation System (MODAS) and has been widely used. Based on this technical route, predecessors have tried to combine some nonlinear regression algorithms for reconstruction, such as self-organizing map neural network (SOM), support vector machine (SVM), etc. After the advent of the machine learning craze, some more popular methods have been put into use, such as generalized regression neural network (GRNN), random forest (RF), extreme gradient boosting learning machine (XGBoost), lightweight gradient boosting learning machine (LightGBM), feedforward neural network (FFNN), etc.

[0004] In the era of ocean big data, deep neural networks have been rapidly developed and applied, and the most classic networks are convolutional neural networks (CNN) and recurrent neural networks (RNN). The neural network structure based on CNN is mainly suitable for "surface" data such as pictures and feature fields, that is, the data is required to be a regular grid with continuous space. The neural network structure based on RNN is mainly suitable for "line" data of time series, that is, the data is required to be a time-continuous sequence. Unfortunately, the traditional reconstruction technology route cannot effectively apply deep neural networks. The main reason is that the points of ocean observations represented by Argo buoys are not fixed and are not continuous in time and space, which cannot meet the data format requirements for deep learning operations. Although gridded data can be obtained from the sea surface as regression factors, since the label data comes from scattered and sparse observation profiles, the grid products of the sea surface can only be interpolated to the corresponding observation profile positions for modeling, that is, the reconstruction is based on the "point-to-point" modeling method, but the "point-to-point" modeling method has the following problems: First, in the "point-to-point" modeling process, the element fields of satellite observations are interpolated to the observation positions that are discontinuous in time and space, which castrates the spatiotemporal evolution information of all-weather satellite observations. Second, the "point-to-point" modeling process cannot fully consider physical constraints in the cost function. For example, physical formulas such as geostrophic balance require space or time differentiation, and only field data or time series can meet the addition of physical constraints. Third, the "point-to-point" modeling process is not conducive to model iteration. The progress in the field of deep learning is mostly centered on images or time series. For example, the large model technology that has gained important attention in recent years cannot be directly migrated into the existing reconstruction architecture in the form of modules. Therefore, the "point-to-point" modeling method causes the reconstruction technology to fall into a bottleneck and cannot follow up with the advanced methods of deep learning.

[0005] Therefore, there is an urgent need to design a method to reconstruct subsurface feature field data based on deep networks and scattered observation data. This method breaks through the bottleneck that traditional reconstruction schemes cannot utilize deep networks, and can assimilate the grid feature fields of satellite observations and scattered profiles of field observations to reconstruct high-precision three-dimensional ocean feature fields. Summary of the invention

[0006] In view of the problem in the above-mentioned prior art that traditional complex deep networks cannot use scattered field observation data sets as label data, the purpose of the present invention is to provide a method for reconstructing subsurface element field data based on deep networks and scattered observation data. The method constitutes the architecture OI-Rec-Net (Observation-Informed ReconstructionNet), which uses the sea surface element architecture observed by satellite as factors and field observation profile data as labels to reconstruct subsurface temperature and salinity data. The architecture includes: (1) a deep neural network part, which predicts subsurface element field data by inputting sea surface element fields; (2) a cost function part of a bilinear interpolation operator, which uses a bilinear interpolation operator to interpolate the predicted subsurface element field to the location of the observation data set, thereby realizing deep network training with the observation data set as labels. The present invention aims to solve the problem that traditional complex deep networks cannot use scattered field observation data sets as label data, and achieve the purpose of comprehensively utilizing multi-source ocean data to reconstruct subsurface element field data.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] A method for reconstructing subsurface feature field data based on a deep network and scattered observation data, characterized in that it includes:

[0009] S1. Based on the input data sea surface feature field X and the output data underwater feature field Y, select the corresponding deep neural network F;

[0010] S2. Using the on-site field observation profile information and the grid information of the subsurface three-dimensional reconstruction element field output by the neural network F, a bilinear interpolation matrix H of multiple observation points is constructed;

[0011] S3.y based on the bilinear interpolation matrix H pred , with the profile observation value y obs For the label, construct a loss function;

[0012] S4. Use the constructed loss function to train the deep neural network F, adjust the network weights, and obtain the trained neural network F;

[0013] S5. The sea surface feature field X is used as input data and input into the trained neural network F. The trained neural network F outputs the final subsurface 3D reconstruction feature field Y. 1 .

[0014] Preferably, in step S1, the input layer dimension of the neural network F is the same as the sea surface element field X, the output layer dimension is the same as the subsurface three-dimensional reconstruction element field Y, and the sea surface element field X, the underwater element field Y and the neural network F satisfy the mapping relationship Y=F(X).

[0015] Preferably, the deep neural network F is a convolutional neural network or a transformer network.

[0016] Preferably, in step S2, the process of constructing a plurality of observation point bilinear interpolation matrices H includes:

[0017] S201. Construct bilinear interpolation of multiple observation points;

[0018] S202. Obtain a bilinear interpolation matrix H based on bilinear interpolation of multiple observation points.

[0019] Preferably, the multiple observation points include an observation point P and an observation point Q, and the bilinear interpolation construction method of the observation point P and the observation point Q is:

[0020] Suppose the horizontal coordinate of the observation point P falling within the given grid is (x o ,y o ), there are four grid points around it, and the values ​​corresponding to the four grid points are Y 11 , Y 21 , Y 12 and Y 22 , the corresponding coordinates are (x 1 ,y 1 )、(x 2 ,y 1 )、(x 1 ,y 2 ) and (x 2 ,y 2 );

[0021] Therefore, the bilinear interpolation expression at the observation point P can be expressed as:

[0022]

[0023] Among them, y P is the value interpolated to the observation point;

[0024] Y 11 The factor of Y 12 The factor of Y 21 The factor of Y 22 The factor of Formula (1) can be simplified as:

[0025]

[0026] Different from the observation point P, the observation point Q is assumed to fall on Y 33 , Y 34, Y 43 and Y 44 Within the coordinates, the bilinear interpolation expression at the observation point Q can be abbreviated as:

[0027]

[0028] Among them, y Q is the value interpolated to the observation point.

[0029] Preferably, the process of obtaining the bilinear interpolation matrix H based on the bilinear interpolation of multiple observation points includes: combining formula (2) and formula (3) and converting them into a matrix form to obtain the bilinear interpolation matrix H;

[0030] The expression of the bilinear interpolation matrix H is:

[0031]

[0032] in,

[0033] Y=(Y 11 ,Y 21 ,Y 12 ,Y 22 ,Y 33 ,Y 43 ,Y 34 ,Y 44 ) T .

[0034] Preferably, in step S3,

[0035] The constructed loss function expression is: J = MSE (y pred ,y obs );

[0036] Among them, y obs is the element value of ocean observation, and MSE refers to the mean square error.

[0037] The second object of the present invention is to provide a system for reconstructing subsurface element field data based on a deep network and scattered observation data, characterized in that it includes:

[0038] The network selection module selects the corresponding deep neural network F based on the input data sea surface feature field X and the output data underwater feature field Y;

[0039] The interpolation matrix construction module uses the on-site field observation profile information and the grid information of the subsurface three-dimensional reconstruction element field output by the neural network F to construct a bilinear interpolation matrix H of multiple observation points;

[0040] The loss function construction module is based on the bilinear interpolation matrix H of y pred, with the profile observation value y obs For the label, construct a loss function;

[0041] A training module, which uses the constructed loss function to train the deep neural network F, adjusts the network weights, and obtains the trained neural network F;

[0042] The output module takes the sea surface element field X as input data and inputs it into the trained neural network F, and the trained neural network F outputs the final subsurface three-dimensional reconstruction element field Y1;

[0043] Wherein, the control system is implemented based on the method described in any one of claims 1-7.

[0044] The third object of the present invention is to provide a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the method according to any one of claims 1-7.

[0045] The fourth object of the present invention is to provide an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method described in any one of claims 1 to 7.

[0046] The beneficial effects of the present invention are as follows: the present invention discloses a method for reconstructing subsurface element field data based on a deep network and scattered observation data. Compared with the prior art, the present invention has the following improvements:

[0047] The present invention designs a deep network architecture (OI-Rec-Net) that can assimilate field observation data, which uses scattered observation information as labels to train deep neural networks and realize deep neural network predictions based on pure observations. The OI-Rec-Net architecture consists of a deep network module + a bilinear interpolation module, which breaks through the bottleneck that traditional deep networks cannot assimilate scattered observation profiles, and allows the neural network using pure observations as training data to have the following functions: 1. Taking into account the spatiotemporal evolution information of all-weather satellite observations; 2. Allowing the addition of physical constraints with space or time as differentials; 3. Allowing the deep network module to be iteratively updated according to the latest progress of the neural network. And when the method of the present invention is actually used, the effect of reconstructing subsurface temperature is better than that of the feedforward neural network of the traditional architecture, and can be comparable to the reanalysis data. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a flowchart of the present invention;

[0049] Figure 2 Schematic diagram of the OI-Rec-Net architecture of the present invention;

[0050] Figure 3 It is a schematic diagram of the bilinear interpolation principle of the present invention;

[0051] Figure 4 A schematic diagram of the experimental implementation of the present invention;

[0052] Figure 5 The temperature distribution of the reconstructed element field at a depth of 100 m for the present invention (the selected date is January 1, 2018);

[0053] Figure 6 is the RMSD of the different data of the present invention compared with the measured data;

[0054] Figure 7 Corr is the comparison between different data of the present invention and measured data. DETAILED DESCRIPTION

[0055] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0056] Embodiment 1:

[0057] The specific structure of the network architecture OI-Rec-Net of the present invention is: composed of a deep network module and a bilinear interpolation module; Figure 1-2 As shown in the figure, the left half is the neural network reconstruction part used to reconstruct the three-dimensional element field. This part enters the deep neural network F by inputting the sea surface element field X, and obtains the output sub-surface three-dimensional reconstruction element field Y. It should be noted that there is no fixed combination of the sea surface input field, neural network, and sub-surface three-dimensional reconstruction element field Y. Figure 2 The ones given in the figure are all for reference only, that is, as long as the mapping relationship Y = F (X) can be satisfied, this architecture can be used. For example, X can be sea surface temperature, sea surface height, sea surface salinity, sea surface temperature gradient, sea surface height gradient, longitude and latitude grid, etc.; F can be a convolutional neural network CNN, or a network architecture such as transformer, etc.; Y can be temperature, salinity, or other ocean elements such as flow rate and chlorophyll.

[0058] The steps of OI-Rec-Net architecture to reconstruct subsurface feature field data based on deep network and scattered observation data include:

[0059] S1. Based on the input data sea surface feature field X and the output data underwater feature field Y, select the corresponding deep neural network F;

[0060] Specifically, the selected deep neural network F needs to be able to achieve when the input data is the sea surface feature field X and the output data is the subsurface feature field Y, the neural network F meets the requirements; if the input sea surface feature field is selected as two types of sea surface temperature and sea surface height, and the number of horizontal grid points is 1440×720, then the dimension of X is 1440×720×2, where the number 2 represents two channels, corresponding to the two types of input data, and the output underwater feature field is the ocean temperature at a depth of 200m, the number of horizontal grid points is 1440×720, then the dimension of Y is 1440×720×1; therefore, the selected neural network F should consist of an input layer dimension of 1440×720×2, an output layer dimension of 1440×720×1 and several hidden layers,

[0061] In this embodiment, the neural network F can be selected as a convolutional neural network, a transformer or other network architecture. It should be noted that the sea surface input field X, the deep neural network F, and the subsurface 3D reconstruction element field Y in this embodiment have no fixed combination. Figure 2 The examples given in the figure are for reference only, that is, this architecture can be used as long as the mapping relationship Y = F (X) is satisfied. For example, X can be sea surface temperature, sea surface height, sea surface salinity, sea surface temperature gradient, sea surface height gradient, longitude and latitude grid, etc.; F can be a convolutional neural network, or a network architecture such as transformer, etc.; Y can be temperature, salinity, or other ocean elements such as flow rate and chlorophyll.

[0062] S2. Construct a bilinear interpolation matrix H of multiple observation points using the on-site field observation profile information and the grid information of the subsurface three-dimensional reconstruction element field Y output by the deep neural network F;

[0063] Field observation profile information includes profile location information and profile observation value y obs , in this step, only the profile position information is used;

[0064] S201. Construct bilinear interpolation of multiple observation points;

[0065] Assume that for a certain time, m observations can be collected in the corresponding area, and the subsurface 3D reconstruction element field Y output by the deep neural network F has a total of n grid points, then the constructed bilinear interpolation matrix is ​​H(m×n); please note that the 3D interpolation operation is actually performed by the 2D data of each layer, that is, each layer uses the same matrix H,

[0066] The bilinear interpolation construction process for a single observation point is as follows;

[0067] like Figure 3 As shown, suppose the observation point P falls within the given grid point, then the horizontal coordinate of the observation point P is (x o ,yo ), there are four grid points closest to it, and the values ​​corresponding to the four grid points are Y 11 , Y 21 , Y 12 and Y 22 , the corresponding coordinates are (x 1 ,y 1 )、(x 2 ,y 1 )、(x 1 ,y 2 ) and (x 2 ,y 2 ), therefore, the bilinear interpolation expression at the observation point P can be expressed as:

[0068]

[0069] Among them, y P is the value interpolated to the observation point;

[0070] For simplicity, Y 11 The factor of Y 12 The factor of Y 21 The factor of Y 22 The factor of Among them, Figure 3 As shown, in In the example, the superscript 12 indicates that the grid point index range is between 1 and 2, and the subscript 11 indicates that the coordinates are (x 1 ,y 1 ) The P in the brackets refers to the observation point P. By analogy, formula (1) can be simplified as:

[0071]

[0072] Among them, y P is the value interpolated to the observation point;

[0073] Consider that there is an observation point Q that is different from the observation point P. Assume that the observation point Q falls on Y 33 , Y 34 , Y 43 and Y 44 Within the coordinates, according to the above principle:

[0074]

[0075] S202. Obtain a bilinear interpolation matrix H based on bilinear interpolation of multiple observation points;

[0076] Specifically, formula (2) and formula (3) are combined and converted into a matrix form to obtain a bilinear interpolation matrix H; that is, formula (4):

[0077]

[0078] in,

[0079] Y=(Y 11 ,Y 21 ,Y 12 ,Y 22 ,Y 33 ,Y 43 ,Y 34 ,Y 44 ) T ;

[0080] Therefore, the interpolation from the grid field Y to multiple scattered observations can be realized in the form of matrix H, that is, y pred =HT, please note that Y here is obtained from the sub-surface 3D reconstruction element field Y output by the deep neural network F in step S101;

[0081] In formula (4), the bilinear interpolation matrix H reflects the interpolation of the grid field Y to the two observation points to obtain y pred In the process of , if there are multiple observations within more grid points, then the formula (4) is expanded by analogy. Please note that the tensor dimension of the output of the deep neural network represented by CNN is a fixed grid field, while the number of observations per day in the ocean is a scattered profile with non-fixed dimensions; since the dimension of the output data of the deep neural network does not correspond to the observation data, the traditional deep neural network cannot be directly trained with the observation data as labels. The bilinear interpolation matrix H is constructed to correspond the fixed-dimensional grid field output by the neural network to the observation data of any dimension, realizing the deep network training with the observation data as labels.

[0082] S3.y based on the bilinear interpolation matrix H pred , with the profile observation value y obs For the label, construct the loss function as needed.

[0083] y obs is the profile observation value of the field observation profile information, that is, y obs The value of ocean observation elements, such as ocean temperature values ​​observed by Argo floats, pred The actual observed value y obs The position and dimension of y are exactly the same. For example, if there are 30 observations at a certain depth, then y obs The dimension of is 30×1, and the y interpolated to the observation point position pred The dimension is also 30×1;

[0084] Based on y pred and obs The loss function can be constructed according to the needs of the user. In this step, the loss function can be constructed as J = MSE (y pred ,y obs ), where MSE stands for mean square error, which can be directly implemented in Python via torch.nn.MSEloss.

[0085] S4. Use the constructed loss function to train the deep neural network F, adjust the network weights, and obtain the trained deep neural network F.

[0086] The training of network weights is performed in Python. Specifically, the optimizer, batch division, and loss value backpropagation are performed in the Python torch toolbox. In this step, the optimizer is Adam.

[0087] (torch.optim.Adam), with 64 samples as a batch (batch_size=64), and the loss value backpropagation is implemented through the loss.backward() code; after setting the number of training rounds (epochs), the training is terminated early according to the early stopping strategy. If the training results are not improved in three consecutive rounds, the model training is terminated early.

[0088] S5. The sea surface feature field X is used as input data and input into the trained deep neural network F. The trained deep neural network F outputs the final subsurface 3D reconstruction feature field Y. 1 ;

[0089] like Figure 2 As shown in the figure, the left half is the neural network reconstruction part used to reconstruct the three-dimensional feature field. This part inputs the sea surface feature field X into the deep neural network F and obtains the output sub-surface three-dimensional reconstruction feature field Y 1 .

[0090] The present invention also provides a system for reconstructing subsurface element field data based on a deep network and scattered observation data, which is characterized by comprising:

[0091] The network selection module selects the corresponding deep neural network F based on the input data sea surface feature field X and the output data underwater feature field Y;

[0092] The interpolation matrix construction module uses the on-site field observation profile information and the grid information of the subsurface three-dimensional reconstruction element field output by the neural network F to construct a bilinear interpolation matrix H of multiple observation points;

[0093] The loss function construction module is based on the bilinear interpolation matrix H of y pred , with the profile observation value y obs For the label, construct a loss function;

[0094] A training module, which uses the constructed loss function to train the deep neural network F, adjusts the network weights, and obtains the trained neural network F;

[0095] The output module takes the sea surface element field X as input data and inputs it into the trained neural network F, and the trained neural network F outputs the final subsurface three-dimensional reconstruction element field Y1;

[0096] The control system is implemented based on the method of Example 1.

[0097] The present invention also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to enable the computer to execute the method of embodiment 1.

[0098] The present invention also provides an electronic device, comprising: at least one processor, and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method of Example 1.

[0099] Embodiment 2:

[0100] Example 2 is to reconstruct the subsurface temperature of the northwest Pacific Ocean using the method of Example 1. Specifically, the sea surface temperature data and sea surface height data observed by satellite are used as input to output the subsurface sea temperature data. The specific process is:

[0101] 1. Basic situation of the experiment and data preparation

[0102] The temporal and spatial resolution of the input data is 0.25°×0.25° / day, and the temporal and spatial resolution of the output data is the same as the input data, with a vertical stratification of 59 layers from 10m to 1000m. The test area is the northwest Pacific (130°E-160°E, 30°N-50°N), the training time range is 2010-2017, and the validation time range is 2018-2019.

[0103] The sea surface temperature (SST) data used in the experiment comes from the Optimum Interpolation Sea Surface Temperature (OISST) | National Centers for Environmental Information (NCEI) (noaa.gov), with a temporal and spatial resolution of 0.25°×0.25° / day. The sea surface height (SSH) data comes from the Copernicus Marine Data Store | Copernicus Marine Service, with the product number SEALEVEL_GLO_PHY_L4_MY_008_047, and a temporal and spatial resolution of 0.25°×0.25° / day. The climate state data used is the World Ocean Atlas 2023 (WOA23) World Ocean Atlas-Climate Fields Data Access (noaa.gov), which is a monthly climate state with a spatial resolution of 0.25°×0.25° and a vertical stratification of 0-1500m with a total of 57 layers. The observation profile data used is from the Global Ocean Science Dataset (CODC-GOSD) Ocean and Climate (iap.ac.cn) of the Ocean Data Center of the Chinese Academy of Sciences, which collects multi-source ocean profile data from 1940 to 2023 and performs quality control. The Global Ensemble Reanalysis Dataset (GREP) is used as the comparison data. This data also comes from the Copernicus Marine Data Store|CopernicusMarine Service of the Copernicus Data Center. The product number is GLOBAL_MULTIYEAR_PHY_ENS_001_031, with a temporal and spatial resolution of 0.25°×0.25° / day and a vertical stratification of 75 layers from 0.5m to 5902.1m.

[0104] (2) Data preprocessing

[0105] Horizontal preprocessing: SST, SSH, WOA23, and GREP data are horizontally interpolated onto the SST grid and the regional extent of the northwest Pacific is intercepted.

[0106] Vertical preprocessing: The CODC-GOSD observation profiles, WOA23 and GREP data were vertically interpolated to 59 layers using the Akima interpolation method.

[0107] Abnormal value extraction: SSTA is obtained by deducting the WOA23 climate state of the corresponding month from SST. For example, the SST on January 6 is obtained by deducting the WOA23 climate state of January from SST. SSHA is obtained by deducting the average SSH average of many years. The WOA23 climate state data is horizontally interpolated to the position of the CODC-GOSD profile to obtain the temperature climate state profile T corresponding to the temperature profile T. clim , then the temperature anomaly profile is T anom =TT clim .

[0108] (3) Constructing the reconstruction model

[0109] In the reconstruction process, for the convenience of specific implementation, a layer-by-layer reconstruction method is adopted, that is, the sea surface information and the first layer, the sea surface information and the second layer are modeled layer by layer, and 59 reconstruction models are constructed. The 59 reconstruction models are trained using the same network, and the reconstruction model of the nth layer is trained according to Figure 2 For detailed implementation of the reconstructed network, see Figure 4 Among them, the network for training the reconstruction model is the OI-Rec-Net model. The schematic diagram of the architecture of the OI-Rec-Net model is as follows: Figure 2 As shown,

[0110] like Figure 4 As shown in the figure, a relatively simple CNN model is used to form a three-layer deep neural network. The input layer contains four channels, namely longitude Lon, latitude Lat, sea surface temperature anomaly SSTA and sea surface height anomaly SSHA. According to the selected range, there are 80×120 spatial grid points. Please note that the longitude and latitude here are grid points. The temperature field is reconstructed through the nth layer output by the model, and the bilinear interpolation module corresponds to the field observation information of the nth layer, realizing network training with field measurements as labels.

[0111] (4) Reconstruction model verification

[0112] The verification of the model is mainly considered from three angles: the first is the horizontal distribution of the reconstructed factor field, the second is the root mean square error RMSD compared with the on-site observations, and the third is the Pearson correlation coefficient Corr compared with the on-site observations.

[0113] The calculation formula of RMSD is as follows:

[0114]

[0115] in, is the i-th (i=1, 2, ..., N) observation at depth z, It is the i-th (i=1, 2, ..., N) prediction value at depth z, which is obtained by bilinear interpolation of the reconstructed field of the corresponding layer.

[0116] The calculation formula of Corr is as follows:

[0117]

[0118] Figure 5 The main forms of the reconstructed temperature element field are drawn in . It can be found that the temperature fields of GREP and OI-Rec-Net reflect better temperature forms, and can reflect the front and vortex more clearly. In contrast, the temperature of WOA23 is too smooth, and only the basic form of the sea temperature is difficult to warm and the north is cold can be seen. Although the temperature field of FFNN seems to reflect the vortex, it is easy to see that the reconstructed temperature field has more localized fine structures. These structures are non-physical and appear in the form of noise points. This means that in the traditional "point-to-point" reconstruction architecture, although FFNN can reconstruct a more detailed structure than WOA23, it is easy to produce subsurface horizontal gradient anomalies. The main reason is that in the "point-to-point" reconstruction, the relationship between adjacent grid points is not fully considered. In contrast, the reconstructed temperature form of OI-Rec-Net presents a more reasonable vortex structure, such as the cyclonic vortex entrainment of cold water in the (153°E-156°E, 30°N-35°E) area.

[0119] Figure 6 The vertical profile of RMSD is plotted in Figure 1. It is obvious from the figure that WOA23 has the highest error. Considering that WOA23 only contains monthly signals, the other three data contain small and medium-scale signals within the effective month. The error of the sea temperature field reconstructed by OI-Rec-Net is much lower than that of FFNN and WOA23, and the effect at 500m above the ocean is even slightly better than GREP. It should be emphasized that GREP data is a combination of the calculation results of multiple models and has undergone a complex data assimilation process. Even so, OI-Rec-Net has achieved results that are not inferior to reanalysis data with lightweight computing costs and calculation processes.

[0120] Figure 7 The vertical profile of Corr is drawn in Figure 2. It can be seen that the correlation coefficient of OI-Rec-Net is optimal at a depth of 100m-500m. At a depth of 500m, GREP has the highest Corr. The correlation sparsity of OI-Rec-Net is superior to that of traditional FFNN at all depths.

[0121] It can be seen that although a very simple CNN network is used in this experiment, its reconstruction effect has far exceeded the FFNN under the traditional architecture and is comparable to the GREP reanalysis product. In the future, with the update and iteration of the network structure and the rapid optimization of the cost function, the OI-Rec-Net architecture is expected to achieve higher-precision reconstruction.

[0122] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A method for reconstructing subsurface feature field data based on deep networks and scattered observation data, characterized in that: include: S1. Based on the input data sea surface feature field X and the output data underwater feature field Y, select the corresponding deep neural network F; S2. Using the on-site field observation profile information and the grid information of the subsurface three-dimensional reconstruction element field output by the neural network F, a bilinear interpolation matrix H of multiple observation points is constructed; S3.y based on the bilinear interpolation matrix H pred , with the profile observation value y obs For the label, construct a loss function; S4. Use the constructed loss function to train the deep neural network F, adjust the network weights, and obtain the trained neural network F; S5. The sea surface element field X is input as input data to the trained neural network F, and the trained neural network F outputs the final sub-surface three-dimensional reconstruction element field Y1.

2. The method for reconstructing subsurface element field data based on deep network and scattered observation data according to claim 1, characterized in that: In step S1, the input layer dimension of the neural network F is the same as the sea surface element field X, the output layer dimension is the same as the subsurface three-dimensional reconstruction element field Y, and the sea surface element field X, the underwater element field Y and the neural network F satisfy the mapping relationship Y=F(X).

3. The method for reconstructing subsurface element field data based on deep network and scattered observation data according to claim 1, characterized in that: The deep neural network F is a convolutional neural network or a transformer network.

4. The method for reconstructing subsurface feature field data based on deep network and scattered observation data according to claim 1, characterized in that: In step S2, the process of constructing a bilinear interpolation matrix H of multiple observation points includes: S201. Construct bilinear interpolation of multiple observation points; S202. Obtain a bilinear interpolation matrix H based on bilinear interpolation of multiple observation points.

5. The method for reconstructing subsurface element field data based on deep network and scattered observation data according to claim 4, characterized in that: The multiple observation points include observation point P and observation point Q, and the bilinear interpolation construction method of observation point P and observation point Q is: Suppose the horizontal coordinate of the observation point P falling within the given grid is (x o ,y o ), there are four grid points around it, and the values ​​corresponding to the four grid points are Y 11 , Y 21 , Y 12 and Y 22 , the corresponding coordinates are (x1, y1), (x2, y1), (x1, y2) and (x2, y2); The bilinear interpolation expression at the observation point P is expressed as: Among them, y P is the value interpolated to the observation point; Y 11 The factor of Y 12 The factor of Y 21 The factor of Y 22 The factor of Formula (1) is: Assume that the observation point Q falls on Y 33 , Y 34 , Y 43 and Y 44 Within the coordinates, the bilinear interpolation expression at the observation point Q is obtained as: Among them, y Q is the value interpolated to the observation point.

6. The method for reconstructing subsurface feature field data based on deep network and scattered observation data according to claim 5, characterized in that: Based on the bilinear interpolation of multiple observation points, the process of obtaining the bilinear interpolation matrix H includes: combining formula (2) and formula (3) and converting them into a matrix form to obtain the bilinear interpolation matrix H; The expression of the bilinear interpolation matrix H is: Among them, Y = (Y 11 , Y 21 , Y 12 , Y 22 , Y 33 , Y 43 , Y 34 , Y 44 ) T .

7. The method for reconstructing subsurface feature field data based on deep network and scattered observation data according to claim 1, characterized in that: In step S3, The constructed loss function expression is: J = MSE (y pred ,y obs ); Among them, y obs is the element value of ocean observation, and MSE refers to the mean square error.

8. A system for reconstructing subsurface feature field data based on deep networks and scattered observation data, characterized in that: include: The network selection module selects the corresponding deep neural network F based on the input data sea surface feature field X and the output data underwater feature field Y; The interpolation matrix construction module uses the on-site field observation profile information and the grid information of the subsurface three-dimensional reconstruction element field output by the neural network F to construct a bilinear interpolation matrix H of multiple observation points; The loss function construction module is based on the bilinear interpolation matrix H of y pred , with the profile observation value y obs For the label, construct a loss function; A training module, which uses the constructed loss function to train the deep neural network F, adjusts the network weights, and obtains the trained neural network F; The output module takes the sea surface element field X as input data and inputs it into the trained neural network F, and the trained neural network F outputs the final subsurface three-dimensional reconstruction element field Y1; Wherein, the control system is implemented based on the method described in any one of claims 1-7.

9. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-7.

10. An electronic device, comprising: At least one processor, and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1 to 7.

Images