Underwater temperature field satellite remote sensing prediction reconstruction method based on Transform architecture

Abstract

The invention relates to an underwater temperature field satellite remote sensing prediction reconstruction method based on a Transform architecture, and the method comprises the steps: carrying out the joint inversion of various sea surface parameter information (position information, time information, sea surface temperature and sea level height abnormity), which is established through the satellite remote sensing data, and the known temperature profile information; reconstruction of an underwater temperature field by using ocean surface parameters is realized, random known temperature information corresponding to 0-3 depth layers is added to training data, and the accuracy of a reconstruction result is improved compared with that of a traditional method. According to the method, the complex ocean dynamics process does not need to be known and analyzed, the reconstructed temperature field data can be directly obtained only by using the corresponding observation data, the calculated amount is small, the process is easy to implement, and the result is accurate.

Description

Technical Field

[0001] The present invention belongs to the fields of ocean physics, ocean engineering and hydroacoustic engineering, and relates to a method for satellite remote sensing prediction and reconstruction of underwater temperature fields based on a Transformer architecture, which is suitable for predicting and reconstructing underwater temperature fields using satellite remote sensing data. Background Art

[0002] Although there are many underwater temperature field reconstruction methods that have been used in engineering practice, such as empirical function regression method, self-organizing neural network method and deep evidence regression network method, the reconstruction of underwater temperature field in deep and complex waters still faces serious technical challenges. The root cause is mainly because the existing temperature profile reconstruction methods have certain defects when facing the strong nonlinear process of complex marine environment, or the reconstruction accuracy is not enough, lack of prediction and reconstruction of complete profiles, etc. The specific analysis is as follows:

[0003] (1) Single empirical orthogonal function method. This method represents the temperature profile by an empirical orthogonal function and its coefficients, and uses the correlation between sea surface remote sensing parameters and the coefficients of the empirical orthogonal function to establish a regression relationship between the two. The reconstruction accuracy of the temperature profile is related to factors such as ocean dynamics, spatial resolution, temporal resolution, observation accuracy of sea surface remote sensing parameters, and combination of sea surface remote sensing parameters, among which ocean dynamics and sea surface remote sensing parameters are the key. In sea areas with frequent vortices and strong fronts, there is usually no significant correlation between sea surface remote sensing parameters and empirical orthogonal function coefficients, which will cause significant errors in the reconstruction of the temperature profile.

[0004] (2) Self-organizing neural network method. This method represents the temperature profile by empirical orthogonal functions and empirical orthogonal function coefficients, and uses sea surface remote sensing parameters and empirical orthogonal function coefficients to establish a regression relationship between the two through a self-organizing neural network. The reconstruction accuracy of the temperature profile is related to factors such as ocean dynamics, spatial resolution, temporal resolution, observation accuracy of sea surface remote sensing parameters, and combination of sea surface remote sensing parameters, among which ocean dynamics and sea surface remote sensing parameters are the key. Although this method has good prediction results in sea areas with frequent vortices and strong fronts, there are still large prediction errors in some cases.

[0005] (3) Deep evidence regression network method. This method uses a deep evidence regression network to combine the location information, time information, and corresponding sea surface remote sensing parameters (such as sea surface temperature and sea surface height) of each profile in the study area to reconstruct the underwater temperature field. The temperature profile reconstruction at different time and space scales is achieved, and the uncertainty estimate, confidence level, and confidence interval of the reconstructed profile are given. Although the prediction results of this method are consistent with the actual results, there are still many prediction results that deviate from the actual results by about ±2°C, or even more, and the prediction accuracy needs to be improved.

[0006] In short, the single empirical orthogonal function method, self-organizing neural network method, deep evidence regression network method, etc. have relatively large errors when reconstructing temperature profiles in areas with active ocean dynamic processes, and there is still a gap in actual needs. Therefore, it is necessary to seek new principles and technical approaches to improve the reconstruction accuracy using field observation data. Underwater unmanned systems such as underwater robots or underwater platforms such as submarines can obtain temperature information at certain depths underwater in real time. These real temperature information has a good effect on improving the reconstruction accuracy of the underwater temperature field and reducing errors. Summary of the invention

[0007] Technical issues to be solved

[0008] In order to avoid the shortcomings of the prior art, the present invention proposes a method for satellite remote sensing prediction and reconstruction of underwater temperature field based on Transformer architecture to make up for the shortcomings of the prior art in estimating prediction results. The present invention combines the temperature information on some known depth layers to improve the prediction accuracy, and is suitable for the prediction and reconstruction of underwater temperature field in deep-sea multi-scale marine environments.

[0009] Technical Solution

[0010] A method for satellite remote sensing prediction and reconstruction of underwater temperature field based on Transformer architecture, characterized by the following steps:

[0011] Step 1: Establish a data set of the input feature network of the multi-dimensional information set, including the location information, time information of the reconstructed profile, and the corresponding sea surface remote sensing parameter data;

[0012] The sea surface remote sensing parameters include sea surface temperature data and sea level height anomaly data;

[0013] The location information includes longitude and latitude;

[0014] The time information is month;

[0015] Perform sine and cosine encoding on the location information data and time information data, and convert them into data that can be understood and learned by the model as input data;

[0016] Step 2: Encapsulate the input data into a PyTorch dataset, and the input features and temperature profiles of the dataset are stacked into two-dimensional tensors;

[0017] Randomly select m known temperature values ​​of the temperature layer, save the temperature value and its corresponding temperature layer index, process the data in the batch and mask the positions of the known temperatures;

[0018] Step 3: Input embedding and position encoding: Map the input features to a high-dimensional space through a linear layer, add position encoding to each position, use sine function encoding for even positions and cosine function encoding for odd positions; after the known temperature value is mapped to the high-dimensional space, it is merged with the position encoding and feature embedding;

[0019] Step 4: Input the encapsulated PyTorch dataset into the network in batches to load the Transformer model. Through the unique multi-head attention mechanism of the Transformer model, the model is trained using historical data.

[0020] Step 5: The location information, time information, and corresponding sea surface remote sensing parameters such as sea surface temperature and sea level height anomaly of the profile to be reconstructed are input into the Transformer network to obtain the temperature field reconstruction value and its corresponding MSE:

[0021]

[0022] Where n represents the number of samples, y i represents the true value, It represents the predicted value; the original temperature profile and the reconstructed temperature profile are obtained by visualizing the test set data.

[0023] The m temperature values ​​of the n known temperature layers are 1-3.

[0024] The encoding process of step 3 is as follows:

[0025]

[0026] Among them, PE represents position, which is used to map the position in the input sequence to the index of the position matrix, pos represents the position of the input data in the sequence, and i represents the dimension index, ranging from 0 to d model Represents the dimension of the model embedding.

[0027] Step 4 uses historical data to complete the training of the model. The parameters during training are set as follows: the loss function uses the mean square error loss function to calculate the difference between the model predicted temperature and the actual temperature; the optimizer uses AdamW, and the learning rate is set to 10 -4 , weight decay is set to 10 -5 .

[0028] The learning rate scheduler uses ReduceLROnPlateau to automatically adjust the learning rate according to the loss of the test set.

[0029] The specific settings of the learning rate scheduler are as follows: when the test set loss does not decrease within 3 training rounds, the learning rate is reduced to half of the previous one; the entire training round of the model is set to 50 times with an early stopping mechanism. During the training process, the best loss of the test set up to the current training round is updated. When the best loss does not improve within 5 consecutive training rounds, the early stopping mechanism is triggered to stop training to prevent overfitting; in addition, the batch size is set to 16.

[0030] During the model training process, information of 1 to 3 known temperature layers is randomly added.

[0031] An electronic device, characterized in that it includes a processor and a memory, wherein the processor is used to implement the steps of data migration of the underwater temperature field satellite remote sensing prediction and reconstruction method based on Transformer architecture when executing the computer program stored in the memory.

[0032] A readable storage medium, characterized in that a computer program is stored on the readable storage medium, and when the computer program is executed by a processor, the steps of data migration of the underwater temperature field satellite remote sensing prediction and reconstruction method based on Transformer architecture are implemented.

[0033] A computer program product, characterized in that it includes computer executable instructions, which, when executed, are used to implement the underwater temperature field satellite remote sensing prediction and reconstruction method based on Transformer architecture.

[0034] Beneficial Effects

[0035] The present invention proposes a method for satellite remote sensing prediction and reconstruction of underwater temperature fields based on Transformer architecture, which utilizes a variety of sea surface parameter information (position information, time information, sea surface temperature, sea level height anomaly) established by satellite remote sensing data and previously known temperature profile information for joint inversion, thereby realizing the reconstruction of the underwater temperature field using ocean surface parameters. Random known temperature information corresponding to 1-3 depth layers is added to the training data, and the accuracy of the reconstruction result is improved compared with the traditional method.

[0036] The present invention utilizes the Transformer architecture to establish a mapping network between sea surface parameters (position information, time information, sea surface temperature, sea level height anomaly) and underwater temperature profiles based on a large amount of remote sensing data. The self-attention mechanism introduced by the Transformer architecture calculates the similarity between each parameter in the input features and other parameters to assign different weights, thereby achieving refined processing of the input features, allowing the model to focus on important information in the input features. Compared with traditional methods, it is more suitable for nonlinear mapping from sea surface parameters to underwater temperature profiles.

[0037] The present invention does not need to understand and analyze complex ocean dynamic processes, and can directly obtain reconstructed temperature field data by using corresponding observation data. The amount of calculation is small, the process is simple to implement, and the result is relatively accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 :(a) The location of the measured temperature profile near the Philippine Sea during 2001-2018 (longitude range: 123-135°E, latitude range: 15-25°N); (b) The temperature profile used in the construction of the Transformer network model in this sea area. (c) The location of the measured temperature profile near the sea east of Japan during 2001-2018 (longitude range: 145-165°E, latitude range: 30-45°N); (d) The temperature profile used in the construction of the Transformer network model in this sea area is given.

[0039] Figure 2 : Flowchart of the temperature field reconstruction framework based on the Transformer model: The flowchart on the left side of the figure is to train the Transformer network through training samples; the flowchart on the right side of the figure is to predict the temperature field through known necessary information.

[0040] Figure 3 :(a), (b), (c), and (d) show the comparison between the predicted profile and the observed profile when the number of known temperature points near the Philippine Sea is 0, 1, 2, and 3, respectively.

[0041] Figure 4 Figures (a), (b), (c), and (d) show the comparison between the predicted profile and the observed profile when the number of known temperature points in the waters east of the Sea of ​​Japan is 0, 1, 2, and 3, respectively. DETAILED DESCRIPTION

[0042] The present invention will now be further described with reference to the embodiments and the accompanying drawings:

[0043] A method for predicting and reconstructing underwater temperature fields based on a Transformer architecture is characterized in that: in the research sea area, the underwater temperature field is used to combine the location information, time information, and corresponding sea surface remote sensing parameters such as sea surface temperature (SST) and sea level anomaly (SLA) data of each profile in the research sea area to establish a data set of the Transformer feature input network of a multidimensional information set, and the temperature data of each profile is used as the ideal output result of the input data set. After the training of the Transformer model is completed, the location information, time information, and corresponding sea surface remote sensing parameters such as sea surface temperature and sea level anomaly information of the profile to be reconstructed are input to invert the reconstructed temperature profile and its mean square error (MSE) information with the real profile data. In order to improve the accuracy of the reconstruction result, the temperature information of 1-3 known depth layers is randomly added during the training process. The entire process and principle of its reconstruction can be divided into the following five steps:

[0044] Step 1: Establish a data set of the input feature network of the multi-dimensional information set, including the location information, time information, and corresponding sea surface remote sensing parameters (sea surface temperature and sea level height anomaly) of the profile. Perform sine and cosine encoding on the location information (here longitude and latitude) and time information (here month) and convert them into data that can be understood and learned by the model to capture the periodic characteristics in the data so that the model can better understand the changing patterns of the data. Standardize these input features to speed up model training and improve generalization capabilities.

[0045] Step 2: Encapsulate the input data as a PyTorch dataset so that it can be loaded in batches during training and testing. The input features and temperature profiles are stacked into two-dimensional tensors, and the temperature values ​​of 1 to 3 known temperature layers are randomly selected and saved with the corresponding temperature layer index. The data in the batch is processed and the positions of the known temperatures are masked.

[0046] Step 3: Input embedding and position encoding. The input features are mapped to a high-dimensional space through a linear layer. Since the Transformer has no implicit order information, position encoding needs to be added for each position. In the Transformer architecture, even positions are encoded using a sine function, and odd positions are encoded using a cosine function. After the known temperature value is mapped to a high-dimensional space, it is merged with the position encoding and feature embedding. The encoding process is as follows:

[0047]

[0048] Among them, PE represents position, which is used to map the position in the input sequence to the index of the position matrix, pos represents the position of the input data in the sequence, and i represents the dimension index, ranging from 0 to d model Represents the dimension of the model embedding.

[0049] Step 4: Build a network model based on the input features, and use historical data to train the network parameters through the unique multi-head attention mechanism of the Transformer model.

[0050] Step 5: According to the location information, time information, and corresponding sea surface remote sensing parameters such as sea surface temperature and sea level height anomaly of the profile to be reconstructed, the Transformer network is input to obtain the temperature field reconstruction value and its corresponding MSE, which can be expressed as:

[0051]

[0052] Where n represents the number of samples, y i represents the true value, Is the predicted value. The original temperature profile and the reconstructed temperature profile are obtained by visualizing the test set data.

[0053] The parameters in network training are set as follows: The loss function uses the mean square error loss function to calculate the difference between the model's predicted temperature and the actual temperature. The optimizer uses AdamW, and the learning rate is set to 10 -4 , weight decay is set to 10 -5 The learning rate scheduler uses ReduceLROnPlateau, which automatically adjusts the learning rate according to the loss of the test set. The specific setting is: when the loss of the test set does not decrease within 3 training rounds, the learning rate is reduced to half of the previous one. The entire training round of the model is set to 50 times with an early stopping mechanism. During the training process, the best loss of the test set up to the current training round is updated. When the best loss does not improve within 5 consecutive training rounds, the early stopping mechanism is triggered to stop training to prevent overfitting. In addition, the batch size is set to 16.

[0054] During the training process, the model will randomly add information of 1-3 known temperature layers. The larger the value, the better. The hope is to reconstruct the temperature field with less information of known temperature layers. The 0-3 here is to unify with the 0-3 of the test set, because we want to know whether the test set results are significantly different under different numbers of known temperature layers. Due to randomness, only a part of the data is not added to the known temperature layer during the training process. If the value of 0 is not appropriate here, changing it to 1-3 will indeed be better for the results.

[0055] Figure 1 In the figure, (a) gives the location of the measured temperature profile near the Philippine Sea during the period 2001-2018, with a depth range of 0-1950m; (b) gives the location of the measured temperature profile near the waters east of Japan during the period 2001-2018, with a depth range of 0-1950m; (c) gives the temperature profile used in constructing the Transformer network near the Philippine Sea with some outliers removed; (d) gives the temperature profile used in constructing the Transformer network near the waters east of Japan.

[0056] Figure 2 A framework flow chart based on the Transformer network is given. The implementation process is divided into two steps:

[0057] (1) Sample training: The sea surface information (including location information, time information, sea surface temperature, and sea height anomaly) is used as input, and the corresponding profile data is used as the expected output result and put into the Transformer network for training. The total sample data used is about 10,146 in the Philippine Sea and about 22,114 in the waters east of Japan. 80% of the samples are randomly selected as training samples, and the remaining 20% ​​of the samples are used as test samples to verify the reconstruction performance of the algorithm.

[0058] (2) Reconstructing the temperature field: The sea surface information (including location information, time information, sea surface temperature, and sea height anomaly) is used as input to obtain the reconstructed temperature field data. The reconstructed data is compared with the real profile temperature data to calculate the mean square error between the reconstructed data and the real data.

[0059] Figure 3 (a), (b), (c), and (d) show the comparison diagrams of the predicted profile and the observed profile when the number of known temperature points near the Philippine Sea is 0, 1, 2, and 3 respectively. The results show that there is good consistency between the reconstructed temperature field and the real temperature field data. The mean square error value of the reconstruction in this sea area is about 0.2937.

[0060] Figure 4 (a), (b), (c), and (d) show the comparison diagrams of the predicted profiles and the observed profiles when the number of known temperature points in the waters east of the Sea of ​​Japan is 0, 1, 2, and 3 respectively. The results show that there is good consistency between the reconstructed temperature field and the real temperature field data. The mean square error value obtained by reconstruction in this sea area is about 0.4062.

[0061] The present invention has achieved good results in tests near the Philippine Sea and the sea east of Japan. Affected by the differences in the original data sets caused by the geographical location and environment of the sea area, the mean square error of the reconstruction results in the two sea areas is different, which are 0.2937 and 0.4062 respectively. The average error temperature is within 0.6℃ to 0.7℃, respectively, which is more accurate than the traditional method. The underwater temperature field reconstruction method based on the Transformer network does not require the understanding and analysis of complex ocean dynamics processes. It only needs to use the corresponding observation data to directly obtain the reconstructed temperature field data. The calculation amount is small, the process is simple to implement, and the results are relatively accurate. It is suitable for large-scale applications and is suitable for using satellite remote sensing observation data within a certain sea area to realize the reconstruction of the underwater temperature field in the area.

[0062] This method establishes a Transformer network with multi-dimensional information such as position information, time information, sea surface temperature, and sea level height anomaly. After the network training is completed, the reconstructed temperature field information under different depth conditions is obtained through the position information, time information of the profile to be reconstructed, and the corresponding sea surface remote sensing parameters (sea surface temperature, sea level height anomaly) information.

[0063] Based on a large amount of remote sensing data, this paper establishes a mapping network between sea surface parameters (position information, time information, sea surface temperature, sea level height anomaly) and underwater temperature profiles, which can realize nonlinear mapping from sea surface parameters to underwater temperature profiles.

[0064] This method is based on big data training of the model. It does not need to understand and analyze complex ocean dynamic processes. It only uses existing observation data and analyzes the correlation between different sea surface parameter data to achieve the reconstruction of the underwater temperature field. It has a small amount of calculation, a simple process, and relatively accurate results. It is more suitable for using satellite remote sensing observation data in a certain sea area to realize the reconstruction of the underwater temperature field in the region.

Claims

1. A method for underwater temperature field satellite remote sensing prediction and reconstruction based on Transformer architecture, characterized in that Here are the steps: Step 1: Establish a data set of the input feature network of the multi-dimensional information set, including the location information, time information of the reconstructed profile, and the corresponding sea surface remote sensing parameter data; The sea surface remote sensing parameters include sea surface temperature data and sea level height anomaly data; The location information includes longitude and latitude; The time information is month; Perform sine and cosine encoding on the location information data and time information data, and convert them into data that can be understood and learned by the model as input data; Step 2: Encapsulate the input data into a PyTorch dataset, and the input features and temperature profiles of the dataset are stacked into two-dimensional tensors; Randomly select m known temperature values ​​of the temperature layer, save the temperature value and its corresponding temperature layer index, process the data in the batch and mask the positions of the known temperatures; Step 3: Input embedding and position encoding: Map the input features to a high-dimensional space through a linear layer, add position encoding to each position, use sine function encoding for even positions and cosine function encoding for odd positions; after the known temperature value is mapped to the high-dimensional space, it is merged with the position encoding and feature embedding; Step 4: Input the encapsulated PyTorch dataset into the network in batches to load the Transformer model. Through the unique multi-head attention mechanism of the Transformer model, the model is trained using historical data. Step 5: The location information, time information, and corresponding sea surface remote sensing parameters such as sea surface temperature and sea level height anomaly of the profile to be reconstructed are input into the Transformer network to obtain the temperature field reconstruction value and its corresponding MSE: Where n represents the number of samples, y i represents the true value, It represents the predicted value; the original temperature profile and the reconstructed temperature profile are obtained by visualizing the test set data.

2. According to the Transformer architecture-based underwater temperature field satellite remote sensing prediction and reconstruction method of claim 1, it is characterized by: The m temperature values ​​of the n known temperature layers are 1-3.

3. According to the Transformer architecture-based underwater temperature field satellite remote sensing prediction and reconstruction method of claim 1, it is characterized by: The encoding process of step 3 is as follows: Among them, PE represents position, which is used to map the position in the input sequence to the index of the position matrix, pos represents the position of the input data in the sequence, and i represents the dimension index, ranging from 0 to d model Represents the dimension of the model embedding.

4. According to claim 1, the underwater temperature field satellite remote sensing prediction and reconstruction method based on Transformer architecture is characterized by: Step 4 uses historical data to complete the training of the model. The parameters during training are set as follows: the loss function uses the mean square error loss function to calculate the difference between the model predicted temperature and the actual temperature; the optimizer uses AdamW, and the learning rate is set to 10 -4 , weight decay is set to 10 -5 .

5. According to claim 4, the underwater temperature field satellite remote sensing prediction and reconstruction method based on Transformer architecture is characterized by: The learning rate scheduler uses ReduceLROnPlateau to automatically adjust the learning rate according to the loss of the test set.

6. The underwater temperature field satellite remote sensing prediction and reconstruction method based on the Transformer architecture according to claim 5 is characterized by: The specific settings of the learning rate scheduler are as follows: when the test set loss does not decrease within 3 training rounds, the learning rate is reduced to half of the previous one; the entire training round of the model is set to 50 times with an early stopping mechanism. During the training process, the best loss of the test set up to the current training round is updated. When the best loss does not improve within 5 consecutive training rounds, the early stopping mechanism is triggered to stop training to prevent overfitting; in addition, the batch size is set to 16.

7. According to claim 1, the underwater temperature field satellite remote sensing prediction and reconstruction method based on Transformer architecture is characterized by: During the model training process, information of 1 to 3 known temperature layers is randomly added.

8. An electronic device, characterized in that: It includes a processor and a memory, and the processor is used to implement the data migration steps of the underwater temperature field satellite remote sensing prediction and reconstruction method based on the Transformer architecture as described in any one of claims 1 to 7 when executing the computer program stored in the memory.

9. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed by a processor, implements the steps of data migration of the underwater temperature field satellite remote sensing prediction and reconstruction method based on the Transformer architecture as described in any one of claims 1 to 7.

10. A computer program product, characterized in that It includes computer executable instructions, which, when executed, are used to implement the underwater temperature field satellite remote sensing prediction and reconstruction method based on Transformer architecture as described in any one of claims 1 to 7.

Images