Multi-element sea surface chlorophyll concentration prediction method based on feature processing

Abstract

The invention relates to the technical field of ocean observation, and provides a multi-element sea surface chlorophyll a concentration prediction method based on feature processing. The method comprises the following steps: acquiring historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data and historical ocean current velocity data in a to-be-predicted region for a continuous period of time, inputting the data into different channels of a convolutional network, extracting five channel feature maps, and performing down-sampling processing; inputting the four channel feature maps subjected to down-sampling processing into a first SELayer layer, calculating the weight of each channel feature map to obtain four weighted channel feature maps, obtaining four spatial feature maps by adopting an attention mixing module, obtaining four weighted spatial feature maps by adopting a second SELayer layer, and fusing the feature maps by adopting a convolutional network and up-sampling processing to obtain a new channel feature map; and outputting the sea surface chlorophyll a concentration data in the to-be-predicted region in a future period of time through the full connection layer.

Description

Technical Field

[0001] The present invention relates to the field of ocean observation technology, and in particular to a multi-factor sea surface chlorophyll a concentration prediction method based on feature processing. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] At present, there are two main methods for detecting sea surface chlorophyll a concentration: process-oriented method and result-oriented method. The process-oriented method represents physical data by fitting indicators into mathematical equations such as power functions and exponential functions. Although it has a wide range of applications, it is not very accurate. The result-oriented method infers chlorophyll a concentration from other measurable ocean variables, usually relying on models such as artificial neural networks (ANN) and random forests (RF). Although the result-oriented method has higher accuracy, its input variables need to be strictly screened and require a large amount of data.

[0004] In addition, existing artificial neural network methods tend to focus only on changes in chlorophyll a concentration, while ignoring the effects of other key ocean parameters (such as sea surface temperature, ocean current speed and salinity) on chlorophyll a concentration. In fact, there is a significant correlation between these ocean parameters and chlorophyll a concentration, and they jointly affect the growth environment of phytoplankton and the distribution of nutrients. For example, changes in sea surface salinity can change water density and flow patterns, affecting the mixing and transportation of nutrients; ocean current speed determines the transport and distribution of nutrients, and upwelling may bring abundant nutrients and promote phytoplankton reproduction; increases or decreases in sea surface temperature directly affect the metabolic rate and growth cycle of phytoplankton.

[0005] However, most current models fail to fully consider the differences in the impact of these different factors on chlorophyll a concentration, greatly reducing the data utilization of the model. This model design that focuses solely on changes in chlorophyll a concentration ignores the complex physical and biochemical processes in the marine ecosystem, limiting the model's comprehensive understanding and prediction capabilities of marine environmental changes. Therefore, future research needs to pay more attention to the comprehensive analysis of multiple factors and the weight allocation of the impact of different factors on chlorophyll a concentration in order to improve the prediction accuracy and reliability of the model. Summary of the invention

[0006] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a multi-factor sea surface chlorophyll a concentration prediction method based on feature processing, designs a prediction model that can process multi-channel data, and can adapt to various influencing factors in the marine environment by receiving and analyzing multiple types of data and explicitly simulating the relationships between these data, thereby significantly improving the accuracy and reliability of the prediction.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] The first aspect of the present invention provides a multi-factor sea surface chlorophyll a concentration prediction method based on feature processing.

[0009] A multi-factor sea surface chlorophyll a concentration prediction method based on feature processing, characterized by comprising:

[0010] Obtain historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data, and historical ocean current velocity data for a continuous period of time in the area to be predicted, input them into different channels of the convolutional network, extract five channel feature maps, and perform downsampling processing;

[0011] SELayer (Squeeze-and-Excitation Layer) is a module used to improve the representation capability of feature maps. It optimizes the expression of feature maps by learning the importance weights of each channel feature map, enhancing important features and suppressing unimportant features.

[0012] The SELayer module works as follows:

[0013] Perform a global average pooling operation on each channel of the input feature map. Through this process, the feature map of each channel is compressed into a scalar, and finally a global description vector is obtained. This operation compresses the feature map from the original C×H×W to a C×1×1 vector, thereby extracting the global information of each channel.

[0014] The importance weights of each channel are learned through two fully connected layers (first reducing the dimension and then increasing the dimension), and the weights are limited to the range of [0,1]. The compressed C×1×1 vector is passed through a fully connected layer to reduce its dimension to C / r. A nonlinear transformation is performed through the ReLU activation function, and then the dimension is restored to C through a fully connected layer, and the weights are limited to the range of [0,1] using the Sigmoid activation function.

[0015] The weight vector obtained by the above operation is multiplied by the original feature map channel by channel. The weight vector is expanded to the same shape C×H×W as the original feature map, and then multiplied element by element. In this way, the feature map is weighted, important features are enhanced and unimportant features are suppressed, thereby optimizing the representation of the feature map.

[0016] The four channel feature maps after downsampling are input into the first SELayer layer, and the weight of each channel feature map is calculated to obtain four weighted channel feature maps;

[0017] Based on the four weighted channel feature maps, an attention mixing module is used to obtain four spatial feature maps;

[0018] Based on the four spatial feature maps, the second SELayer layer is used to obtain four weighted spatial feature maps;

[0019] Based on four weighted spatial feature maps, a convolutional network and upsampling processing are used to fuse the feature maps. After the fully connected layer, the sea surface chlorophyll a concentration data for a period of time in the future in the predicted area is output.

[0020] Furthermore, the processing of the first SELayer layer includes:

[0021] Through global average pooling of each channel feature map, the channel descriptor of each channel is obtained;

[0022] Based on the channel descriptor of each channel, two fully connected layers and a ReLU activation function are used to learn the nonlinear relationship between channels and obtain the weight of each channel;

[0023] Multiply the weight of each channel by the channel feature map corresponding to the channel to obtain a weighted channel feature map.

[0024] Furthermore, the nonlinear relationship between channels is learned by using two fully connected layers and a ReLU activation function to obtain the weight of each channel; the following formula is used to express it:

[0025] s i =σ(W 2 δ(W 1 z i ))

[0026] Among them, W 1 and W 2 is the weight matrix of the fully connected layer, δ is the ReLU activation function, and σ is the sigmoid activation function, which is used to normalize the weights to between [0,1].

[0027] Furthermore, the ReLU activation function is expressed by the following formula:

[0028]

[0029] Here, x is the input of the neuron and α is a small positive value that controls the slope of the function in the negative input region.

[0030] Furthermore, the attention hybrid module includes a cuboid attention block and a downsampling combination module, a residual connection stacked cuboid attention block, and a cuboid attention block and an upsampling combination module; the cuboid attention block and the downsampling combination module are used to simultaneously capture the correlation between features in the spatial dimension and the channel dimension and reduce the spatial dimension; the residual connection stacked cuboid attention block introduces jump connections and multiple cuboid attention blocks; the cuboid attention block and the upsampling combination module are used to restore the spatial dimension of the feature map.

[0031] Furthermore, the historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data and historical ocean current velocity data for a continuous period of time in the area to be predicted are data located at the same longitude and latitude and at the same time.

[0032] The second aspect of the present invention provides a multi-factor sea surface chlorophyll a concentration prediction system based on feature processing.

[0033] A multi-factor sea surface chlorophyll a concentration prediction system based on feature processing, comprising:

[0034] A data acquisition module is configured to: obtain historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data, and historical ocean current velocity data for a continuous period of time in the area to be predicted, input them into different channels of the convolutional network, extract five channel feature maps, and perform downsampling processing;

[0035] A first weighting module is configured to: input the four channel feature maps after downsampling into the first SELayer layer, calculate the weight of each channel feature map, and obtain four weighted channel feature maps;

[0036] An attention mixing module, which is configured to: obtain four spatial feature maps by using the attention mixing module based on the four weighted channel feature maps;

[0037] A second weighting module is configured to: based on the four spatial feature maps, adopt a second SELayer layer to obtain four weighted spatial feature maps;

[0038] The prediction module is configured as follows: based on four weighted spatial feature maps, a convolutional network and upsampling processing are used to fuse the feature maps, and through a fully connected layer, the sea surface chlorophyll a concentration data for a period of time in the future in the area to be predicted is output.

[0039] A third aspect of the present invention provides a computer-readable storage medium.

[0040] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing as described in the first aspect above.

[0041] A fourth aspect of the present invention provides a computer device.

[0042] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing as described in the first aspect above are implemented.

[0043] A fifth aspect of the present invention provides a computer program product or a computer program.

[0044] The present invention provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing as described in the first aspect above.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] Traditional prediction methods often do not fully consider the complexity of the marine environment, such as key factors such as sea surface salinity, ocean current velocity and sea surface temperature, which have a significant impact on the concentration of marine chlorophyll a. Due to the lack of comprehensive consideration of these dynamic variables, traditional deep learning models may not be able to fully capture all relevant factors that affect the changes in chlorophyll a concentration, resulting in their prediction effects not always being optimal. The present invention can fully utilize the correlation between different physical parameters and improve the accuracy and robustness of the prediction model by integrating multi-dimensional data such as historical sea surface chlorophyll a concentration, temperature, salinity and ocean current velocity.

[0047] In addition to chlorophyll a concentration, the present invention adds multiple elements (such as sea surface salinity, ocean current velocity, and sea surface temperature, etc.) to the data set, uses the fusion of deep learning technology, and improves the efficiency and accuracy of feature learning through global information squeezing and inter-channel collaborative operations, thereby achieving deep mining and efficient use of complex multidimensional data. It captures complex, long-distance dependencies from multiple data (sea surface salinity, ocean current velocity, sea surface temperature, sea surface chlorophyll a concentration), while maintaining efficient parallel computing capabilities, and achieves accurate prediction of sea surface chlorophyll a concentration. When using the same data set, it can provide faster and more accurate predictions.

[0048] The present invention uses the SELayer layer to calculate the weight of each channel feature map, which can emphasize the features that are more important to the prediction task and suppress unimportant features, thereby enhancing the model's ability to capture key information. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0050] Figure 1 It is a flow chart of a multi-factor sea surface chlorophyll a concentration prediction method based on feature processing shown in the present invention;

[0051] Figure 2 It is a structural schematic diagram of the prediction model shown in the present invention;

[0052] Figure 3 It is a schematic diagram of the data logic structure shown in the present invention;

[0053] Figure 4 The present invention shows the scatter plots of chlorophyll concentration predicted by the Chldiff model throughout the year and in each season; A is a scatter plot of data for the whole year, and B, C, D, and E are scatter plots of data for each season of spring, summer, autumn, and winter, respectively;

[0054] Figure 5 The scatter plots of chlorophyll concentration predicted by the SEChlDiff model for the whole year and each season are shown in the present invention, wherein A is a scatter plot of data for the whole year, and B, C, D, and E are scatter plots of data for each season of spring, summer, autumn, and winter, respectively. DETAILED DESCRIPTION

[0055] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0056] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0057] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0058] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and systems according to various embodiments of the present disclosure. It should be noted that each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, program segment, or a part of a code may include one or more executable instructions for implementing the logical functions specified in each embodiment. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of boxes in the flowchart and / or block diagram can be implemented using a dedicated hardware-based system that performs a specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0059] Embodiment 1

[0060] like Figure 1 As shown, this embodiment provides a multi-factor sea surface chlorophyll a concentration prediction method based on feature processing. This embodiment uses the method applied to a server as an example. It can be understood that the method can also be applied to a terminal, and can also be applied to a terminal, a server, and a system, and is implemented through the interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0061] Step 1: Obtain historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data, and historical ocean current velocity data for a continuous period of time in the area to be predicted, input them into different channels of the convolutional network, extract five channel feature maps, and perform downsampling processing;

[0062] In this embodiment, the time range for obtaining historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data and historical ocean current velocity data for a continuous period of time in the area to be predicted is January 1, 2015 to December 31, 2022, and the spatial range is the area of ​​23°N-41°N, 116°E-132°E.

[0063] In some embodiments, after acquiring the data, the data needs to be preprocessed, including:

[0064] (1) Interpolation processing

[0065] Extract key parameters from data such as sea surface temperature, ocean current velocity and various salinities, including sampling date, longitude and latitude. Take the corresponding linear interpolation of ocean surface chlorophyll a concentration as an example, the specific method is as follows:

[0066] Taking the historical ocean surface chlorophyll a concentration data as the center, the key parameters of the data are determined, including longitude, latitude, sampling date, etc., and other data elements, including sea surface temperature, ocean current velocity and various salinity data, are interpolated with the same date, longitude and latitude to achieve uniform distribution of data in each channel.

[0067] (2) Normalization

[0068] Before dividing the data set, the interpolated sea surface chlorophyll a concentration data was normalized. The normalization formula is:

[0069]

[0070] Among them, Z represents the normalized sea surface chlorophyll a concentration value; x is the actual observed value of sea surface chlorophyll a concentration after interpolation processing; μ and σ represent the mean and standard deviation of the observed values, respectively.

[0071] (3) Constructing a dataset

[0072] The longitude, latitude, year, sea surface chlorophyll a concentration data, sea surface temperature data and various salinity data after the above processing are integrated into the input feature matrix of the Squeeze-and-Excitation Layer Chlorophylls a latent Diffusion spatio-temporal predictive learning Model (SEChldiff), and the sea surface chlorophyll a concentration data is used as the target output of the model. In the data set construction stage, each row of data represents an independent observation sample with a time scale of 1 day, which contains all the above input features and the corresponding sea surface chlorophyll a concentration data.

[0073] In addition to the chlorophyll a concentration itself, many other factors will affect the chlorophyll a concentration, such as sea surface salinity, ocean current velocity, sea surface temperature, etc. Changes in salinity can affect the density and fluidity of the water body, and thus affect the distribution and supply of nutrients. For example, in the study of Bohai Sea, the chlorophyll a concentration in summer was negatively correlated with salinity. This may be because the input of diluted water from the Yellow River in summer led to a low-salinity environment, while carrying rich nutrients, which promoted the growth of phytoplankton. The velocity of ocean currents affects the transport and distribution of nutrients, thereby indirectly affecting the concentration of chlorophyll a. Ocean currents can transport nutrients from the deep sea or the coast to other areas, providing phytoplankton with the nutrients they need for growth. In the Bohai Sea, the diluted water from the Yellow River is affected by the southeast wind in summer and moves northeast. The nutrients carried by the river water may promote the growth of phytoplankton in the central sea area. Temperature is one of the main environmental factors affecting the growth of phytoplankton. In the winter in the Bohai Sea, temperature is the main factor affecting the growth of phytoplankton. In the summer, due to the influence of the water and sediment regulation of the Yellow River, the peak monthly runoff of the Yellow River is advanced to summer, resulting in the replenishment of nutrients in summer, which in turn leads to a significant increase in chlorophyll a concentration.

[0074] In the field of chlorophyll concentration prediction, multiple factors such as sea surface salinity, ocean current velocity, and sea surface temperature are introduced, which have a significant impact on the concentration of chlorophyll a. These environmental factors affect the growth and distribution of phytoplankton, thereby affecting the concentration of chlorophyll a. For example, changes in sea surface salinity can affect the density and fluidity of the water body, thereby affecting the distribution and supply of nutrients; the velocity of ocean currents affects the transportation and distribution of nutrients; and sea surface temperature is one of the main environmental factors affecting the growth of phytoplankton.

[0075] In one or more implementations, the SEChldiff model is obtained by adding a SELayer layer to the ChlDiff model.

[0076] The ChlDiff model is an improved deep learning model based on the PreDiff architecture, which improves the model's learning representation of ocean data features and its ability to fit data under complex changes. The ChlDiff model is applied to the prediction of marine environmental variables, especially the prediction of sea surface chlorophyll a concentration.

[0077] like Figure 3As shown, in this embodiment, the SEChlDiff model first receives data features from different channels as input, which may include chlorophyll a concentration, sea surface salinity, ocean current velocity, sea surface temperature, etc. In order to extract useful information from these raw data, the model uses a multimodal encoder. The multimodal encoder consists of a 2D-CNN convolutional layer and a pooling layer, which can learn and extract high-level representations from the raw data. Among them, the 2D-CNN convolutional layer learns local features from the input data through convolution operations, while the pooling layer reduces the spatial size of the data through downsampling operations while retaining the most important information.

[0078] Step 2: Input the four channel feature maps after downsampling into the first SELayer layer, calculate the weight of each channel feature map, and obtain four weighted channel feature maps;

[0079] The core idea of ​​the SELayer layer is to perform weight calculations on the features of each channel to emphasize the features that have a greater impact on the prediction results and suppress unimportant features.

[0080] For each channel feature map F i (where i represents the channel index), the SELayer layer first obtains a channel descriptor z through global average pooling (GAP) i :

[0081]

[0082] Among them, H and W are the height and width of the feature map respectively.

[0083] Next, two fully connected layers (FC) and a ReLU activation function are used to learn the nonlinear relationship between channels and obtain the weight s of each channel. i :

[0084] s i =σ(W 2 δ(W 1 z i ))

[0085] Among them, W 1 and W 2 is the weight matrix of the fully connected layer, δ is the ReLU activation function, and σ is the sigmoid activation function, which is used to normalize the weights to between [0,1].

[0086] Finally, the learned weights s i Multiply by the corresponding feature map F i , and get the weighted feature map

[0087]

[0088] The ReLU activation function is expressed by the following formula:

[0089]

[0090] Here, x is the input of the neuron and α is a small positive value that controls the slope of the function in the negative input region.

[0091] Step 3: Based on the four weighted channel feature maps, an attention mixing module is used to obtain four spatial feature maps;

[0092] In some embodiments, the attention hybrid module includes a cuboid attention block and a downsampling combination module, a residual connection stacked cuboid attention block, and a cuboid attention block and an upsampling combination module; the cuboid attention block and the downsampling combination module are used to simultaneously capture the correlation between features in the spatial dimension and the channel dimension and reduce the spatial dimension; the residual connection stacked cuboid attention block introduces jump connections and multiple cuboid attention blocks; the cuboid attention block and the upsampling combination module are used to restore the spatial dimension of the feature map.

[0093] Weighted feature map It is fed into several cuboid attention blocks and downsampling modules. The cuboid attention blocks can capture the correlation between features in both spatial and channel dimensions. The downsampling modules are used to further reduce the spatial dimension of the data.

[0094] After being processed by several cuboid attention blocks and downsampling modules, the data is fed into the residual connection stacked cuboid attention block. The residual connection stacked cuboid attention block further enhances the model's feature learning ability by introducing skip connections and multiple cuboid attention blocks. Skip connections allow input information to be passed directly to subsequent layers, thus avoiding the problem of gradient vanishing and gradient exploding.

[0095] The output of the residual connected stacked cuboid attention blocks is fed again into several cuboid attention blocks and upsampling modules. The upsampling modules are used to gradually restore the spatial dimensions of the data to match the spatial resolution of the original input data.

[0096] In the residual connection stacked cuboid attention block, each cuboid attention block may contain one or more convolutional layers, activation functions, attention layers and other components. These components work together to further extract and transform the input features. Finally, the output of the module is sent to the SELayer layer again for weight calculation.

[0097] Figure 2In the ChlDiff model, a variational autoencoder (VAE) encodes high-dimensional observations into a low-dimensional latent space, and then trains a conditional latent diffusion model on this compressed latent space. This model is based on the Earthformer-UNet architecture, which is a hierarchical Transformer encoder-decoder based on cuboid attention. The length of the input sequence is T, and the length of the target sequence is K. "×D" means stacking D cuboid attention blocks with residual connections. "M×" means having M layers of hierarchy.

[0098] Step 4: Based on the four spatial feature maps, the second SELayer layer is used to obtain four weighted spatial feature maps;

[0099] After being processed by several cuboid attention blocks and upsampling modules, the data is sent to the SELayer layer again for weight calculation. This step is similar to the previous SELayer layer, but it is based on feature maps that have been transformed and learned more.

[0100] Step 5: Based on the four weighted spatial feature maps, a convolutional network and upsampling processing are used to fuse the feature maps. After the fully connected layer, the sea surface chlorophyll a concentration data for a period of time in the future in the area to be predicted is output.

[0101] After the SELayer layer is processed, the data is fed into the 2D-CNN+ upsampling module, which consists of a series of convolutional layers and upsampling layers to gradually restore the spatial dimensions of the data and generate the final prediction results. The upsampling layer usually increases the spatial size of the data through methods such as transposed convolution or nearest neighbor interpolation.

[0102] After the 2D-CNN+upsampling module, the SECHldiff model also includes one or more fully connected layers to map the extracted features to the final predicted values. Finally, the SECHldiff model outputs the predicted results of the chlorophyll a concentration on the ocean surface.

[0103] The present invention adds the SELayer layer, and the SEChlDiff model can generate prediction results that are more consistent with the actual physical process. This structure enables the model to pay more attention to the features that have a greater impact on the prediction results, thereby improving the interpretability and credibility of the model.

[0104] In the SECHldiff model, data features of different channels are used as input, and a multimodal encoder is used to extract high-level representations of each mode. On this basis, the SELayer layer is introduced after the 2D-CNN+ downsampling module and before the 2D-CNN+ upsampling module. Through this layer, the model can calculate the weights of each channel, so as to more effectively learn the characteristics of the data of each channel in the ocean. This method not only improves the efficiency and accuracy of feature learning, but also significantly improves the model's prediction accuracy of the ocean surface chlorophyll concentration.

[0105] In general, the SEChlDiff model achieves a more accurate and interpretable prediction of chlorophyll a concentration by comprehensively considering multiple environmental factors and utilizing the SELayer layer.

[0106] By adjusting model parameters and optimizing algorithms, the SEChlDiff model can not only generate prediction results that meet physical constraints, but also more effectively capture and utilize complex features in the data, thereby improving the accuracy and reliability of the prediction.

[0107] The SEChlDiff model achieves a more accurate and interpretable prediction of chlorophyll a concentration by comprehensively considering multiple environmental factors and introducing SELayer layers, cuboid attention blocks, and residual connections in the latent space of the ChlDiff model. The SELayer layer emphasizes important features and suppresses unimportant features by weighting the features of each channel, while the cuboid attention block further enhances the model's feature learning ability by capturing the correlation between features. This method not only improves the efficiency and accuracy of feature learning, but also significantly improves the model's prediction accuracy for ocean surface chlorophyll concentration.

[0108] The data set is divided into training set and test set with a ratio of 8:2, and the preprocessed training set is input into the model to predict the chlorophyll a concentration data.

[0109] The batch size is set to 1, the training epoch is set to 500, the optimizer used is Adam, and the initial learning rate of training is 0.0001. The absolute error loss (L1Loss) is used as the loss function to minimize the absolute error between the predicted value and the true value.

[0110] The mean absolute error (MAE) and root mean square error (RMSE) are used to evaluate the model accuracy. The following are the formulas for each indicator.

[0111] RMSE is the square root of the average of the squares of the differences between the predicted values ​​and the true values. It emphasizes the impact of larger errors and is calculated as follows:

[0112]

[0113] Among them, y i represents the true value of the i-th sample, Represents the predicted value of the i-th sample.

[0114] MAE is the average of the absolute values ​​of the differences between the predicted values ​​and the true values. It gives equal weight to all errors and does not emphasize large errors like RMSE. The calculation formula is as follows:

[0115]

[0116] The linearly interpolated dataset is first convolved with a 2×1 convolution kernel to extract local features; then it is processed by batch normalization, activation, and Dropout before passing through the Transformer encoder. The input and output dimensions of the encoder are 64, and 4 attention heads are used for multi-head self-attention calculation to capture local dependencies in the data; then the data flows through another one-dimensional convolution layer with a 4×1 convolution kernel to continue feature extraction; then after batch normalization, activation, and Dropout, it passes through the second Transformer encoder to further deepen the understanding of long-term dependencies. The input and output dimensions of the second encoder are increased to 128, and 8 attention heads are used to further understand the long-term dependencies in the data. Then, after the data passes through the convolution layer of the 4×1 convolution kernel, it is processed by batch normalization, activation, and Dropout, and then passes through the convolution layer of the 3×1 volume set. Finally, the data output by the module is integrated into the input feature matrix of the model and output to the transmission layer to generate the final prediction results.

[0117] In order to verify the prediction accuracy of the method of the present invention, the above-mentioned evaluation parameters are compared with several existing models. The results are shown in Table 1. The model includes the ChlDiff model and the SEChlDiff model combined with the SELayer layer, and the sea surface chlorophyll a concentration data predicted by the model and the actual observation data are processed by annual average and each season, and the mean absolute error (MAE), root mean square error (RMSE) and two statistical indicators are used for comparative analysis, as shown in Table 1.

[0118] Table 1 Comparison of evaluation parameters of several existing models

[0119]

[0120] As can be seen from Table 1, compared with the traditional ChlDiff model, the SELayer layer has significantly improved the annual average, spring and winter predictions of the SEChlDiff model. This shows that the SELayer layer can effectively improve the prediction accuracy of the model for the ocean surface chlorophyll concentration.

[0121] Figure 4 The present invention shows the scatter plots of chlorophyll concentration predicted by the Chldiff model throughout the year and in each season; A is a scatter plot of data for the whole year, and B, C, D, and E are scatter plots of data for each season of spring, summer, autumn, and winter, respectively; Figure 5 It is the scatter plot of chlorophyll concentration predicted by the SEChlDiff model shown in the present invention throughout the year and in each season. Wherein A is the scatter plot of data throughout the year, and B, C, D, and E are the scatter plots of data in each season of spring, summer, autumn, and winter respectively. It can be seen that the SEChlDiff model not only has excellent prediction effect in a wide range of sea areas, but also performs well in the prediction accuracy of specific key areas, especially in the fit of extreme value areas, which is better than other models.

[0122] Embodiment 2

[0123] This embodiment provides a multi-factor sea surface chlorophyll a concentration prediction system based on feature processing.

[0124] A multi-factor sea surface chlorophyll a concentration prediction system based on feature processing, comprising:

[0125] A data acquisition module is configured to: obtain historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data, and historical ocean current velocity data for a continuous period of time in the area to be predicted, input them into different channels of the convolutional network, extract five channel feature maps, and perform downsampling processing;

[0126] A first weighting module is configured to: input the four channel feature maps after downsampling into the first SELayer layer, calculate the weight of each channel feature map, and obtain four weighted channel feature maps;

[0127] An attention mixing module, which is configured to: obtain four spatial feature maps by using the attention mixing module based on the four weighted channel feature maps;

[0128] A second weighting module is configured to: based on the four spatial feature maps, adopt a second SELayer layer to obtain four weighted spatial feature maps;

[0129] The prediction module is configured as follows: based on four weighted spatial feature maps, a convolutional network and upsampling processing are used to fuse the feature maps, and through a fully connected layer, the sea surface chlorophyll a concentration data for a period of time in the future in the area to be predicted is output.

[0130] In some embodiments, the nonlinear relationship between channels is learned by using two fully connected layers and a ReLU activation function to obtain the weight of each channel; the following formula is used to express it:

[0131] s i =σ(W 2 δ(W 1 z i ))

[0132] Among them, W 1 and W 2 is the weight matrix of the fully connected layer, δ is the ReLU activation function, and σ is the sigmoid activation function, which is used to normalize the weights to between [0,1].

[0133] In some embodiments, the ReLU activation function is expressed by the following formula:

[0134]

[0135] Here, x is the input of the neuron and α is a small positive value that controls the slope of the function in the negative input region.

[0136] In some embodiments, the attention hybrid module includes a cuboid attention block and a downsampling combination module, a residual connection stacked cuboid attention block, and a cuboid attention block and an upsampling combination module; the cuboid attention block and the downsampling combination module are used to simultaneously capture the correlation between features in the spatial dimension and the channel dimension and reduce the spatial dimension; the residual connection stacked cuboid attention block introduces jump connections and multiple cuboid attention blocks; the cuboid attention block and the upsampling combination module are used to restore the spatial dimension of the feature map.

[0137] In some embodiments, the historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data, and historical ocean current velocity data for a continuous period of time in the area to be predicted are data located at the same longitude and latitude and at the same time.

[0138] Embodiment 3

[0139] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing as described in the first embodiment above are implemented.

[0140] Embodiment 4

[0141] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing as described in the first embodiment above are implemented.

[0142] Embodiment 5

[0143] This embodiment provides a computer program product or a computer program, which includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing described in the first embodiment.

[0144] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0145] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0146] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A multi-factor sea surface chlorophyll a concentration prediction method based on feature processing, characterized in that: include: Obtain historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data, and historical ocean current velocity data for a continuous period of time in the area to be predicted, input them into different channels of the convolutional network, extract five channel feature maps, and perform downsampling processing; The four channel feature maps after downsampling are input into the first SELayer layer, and the weight of each channel feature map is calculated to obtain four weighted channel feature maps; Based on the four weighted channel feature maps, an attention mixing module is used to obtain four spatial feature maps; Based on the four spatial feature maps, the second SELayer layer is used to obtain four weighted spatial feature maps; Based on four weighted spatial feature maps, a convolutional network and upsampling processing are used to fuse the feature maps. After the fully connected layer, the sea surface chlorophyll a concentration data for a period of time in the future in the predicted area is output.

2. The multi-factor sea surface chlorophyll a concentration prediction method based on feature processing according to claim 1 is characterized in that: The processing of the first SELayer layer includes: Through global average pooling of each channel feature map, the channel descriptor of each channel is obtained; Based on the channel descriptor of each channel, two fully connected layers and a ReLU activation function are used to learn the nonlinear relationship between channels and obtain the weight of each channel; Multiply the weight of each channel by the channel feature map corresponding to the channel to obtain a weighted channel feature map.

3. The multi-factor sea surface chlorophyll a concentration prediction method based on feature processing according to claim 2 is characterized in that: The nonlinear relationship between channels is learned by using two fully connected layers and a ReLU activation function to obtain the weight of each channel; It is expressed by the following formula: s i =σ(W2δ(W1z i )) Among them, W1 and W2 are the weight matrices of the fully connected layer, δ is the ReLU activation function, and σ is the sigmoid activation function, which is used to normalize the weights to between [0,1].

4. The multi-factor sea surface chlorophyll a concentration prediction method based on feature processing according to claim 3 is characterized in that: The ReLU activation function is expressed by the following formula: Here, x is the input of the neuron and α is a small positive value that controls the slope of the function in the negative input region.

5. The multi-factor sea surface chlorophyll a concentration prediction method based on feature processing according to claim 1 is characterized in that: The attention hybrid module includes a cuboid attention block and a downsampling combination module, a residual connection stacked cuboid attention block, and a cuboid attention block and an upsampling combination module; the cuboid attention block and the downsampling combination module are used to simultaneously capture the correlation between features in the spatial dimension and the channel dimension and reduce the spatial dimension; the residual connection stacked cuboid attention block introduces jump connections and multiple cuboid attention blocks; the cuboid attention block and the upsampling combination module are used to restore the spatial dimension of the feature map.

6. The multi-factor sea surface chlorophyll a concentration prediction method based on feature processing according to claim 1 is characterized in that: The historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data and historical ocean current velocity data for a continuous period of time in the area to be predicted are data located at the same longitude and latitude and at the same time.

7. A multi-factor sea surface chlorophyll a concentration prediction system based on feature processing, characterized in that: include: A data acquisition module is configured to: obtain historical sea surface chlorophyll a concentration data, historical sea surface temperature data, historical sea surface salinity data, and historical ocean current velocity data for a continuous period of time in the area to be predicted, input them into different channels of the convolutional network, extract five channel feature maps, and perform downsampling processing; A first weighting module is configured to: input the four channel feature maps after downsampling into the first SELayer layer, calculate the weight of each channel feature map, and obtain four weighted channel feature maps; An attention mixing module, which is configured to: obtain four spatial feature maps by using the attention mixing module based on the four weighted channel feature maps; A second weighting module is configured to: based on the four spatial feature maps, adopt a second SELayer layer to obtain four weighted spatial feature maps; The prediction module is configured as follows: based on four weighted spatial feature maps, a convolutional network and upsampling processing are used to fuse the feature maps, and through a fully connected layer, the sea surface chlorophyll a concentration data for a period of time in the future in the area to be predicted is output.

8. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps in the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing as described in any one of claims 1 to 6 are implemented.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing as described in any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the steps in the multi-factor sea surface chlorophyll a concentration prediction method based on feature processing as described in any one of claims 1 to 6 are implemented.

Images