Multichannel sea surface chlorophyll concentration prediction method and system based on knowledge control

Abstract

The invention provides a multi-channel sea surface chlorophyll a concentration prediction method and system based on knowledge control. The method comprises the following steps: constructing a data set by using preprocessed sea surface data, and dividing the data set into a training set, a first test set and a second test set; respectively carrying out training and precision evaluation on the KC-ChlDiff model by utilizing the training set and the first test set; the KC-ChlDiff model is composed of a variational auto-encoder and a knowledge control module, the variational auto-encoder is used for encoding high-dimensional observation data to low-dimensional observation data, and the knowledge control module adjusts the transition probability in the generation process through training learning so as to generate a prediction result of the sea surface chlorophyll a; and inputting the second test set into the KC-ChlDiff model which is qualified in evaluation to predict the concentration of the sea surface chlorophyll a. According to the method, a sample can be generated for an extreme condition under the guidance of knowledge control in combination with priori knowledge, so that accurate prediction of the concentration of the sea surface chlorophyll a is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of sea surface chlorophyll concentration prediction, and in particular relates to a multi-channel sea surface chlorophyll a concentration prediction method and system based on knowledge control. 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] Chlorophylls a (Chl-a) is one of the basic indicators for characterizing the distribution of plankton and the degree of eutrophication of water bodies. As an important component of marine water bodies, chlorophyll concentration can effectively reflect the content of plankton in the ocean, measure the primary productivity of the ocean, and also serve as an indicator for monitoring the quality of marine water.

[0004] The research on Chl-a concentration prediction methods is currently divided into two categories: based on physical and chemical analysis and based on machine learning. Among them, the research method based on physical and chemical analysis is to infer the changes in the Chl-a concentration of the water body based on its own physical and chemical properties; however, due to the mutual influence of various factors in the water body, the relevant physical and chemical model structure is complex and diverse, and the calculation complexity is high, especially when dealing with large-scale data.

[0005] Diffusion models (DM) are a class of generative models that have become increasingly popular in recent years. DMs work by adding noise to the data in a forward process and then approximating the reverse process to remove the noise and learn the data distribution. Latent diffusion models (LDMs) are a variant of DMs that are trained on the latent vector output of a variational autoencoder. Studies have shown that LDMs are more efficient to train than the original DMs and are able to generate higher quality images. However, in recent years, in the field of deep learning, on the one hand, although diffusion models have achieved excellent results in image and video generation, they have not yet been applied to deep learning for spatiotemporal prediction of sea surface chlorophyll; on the other hand, deep learning applications for sea surface chlorophyll prediction have made significant progress, but existing methods still lack the ability to incorporate prior knowledge, which is essential for controllable generation; therefore, the existing technology is not ideal for the prediction accuracy of sea surface chlorophyll a concentration. Summary of the invention

[0006] In order to overcome the shortcomings of the above-mentioned prior art, the present invention provides a multi-channel sea surface chlorophyll a concentration prediction method and system based on knowledge control, which can combine prior knowledge and generate samples for extreme situations under the guidance of knowledge control, thereby achieving accurate prediction of sea surface chlorophyll a concentration.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0008] The first aspect of the present invention provides a multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control.

[0009] The multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control includes:

[0010] Collect sea surface data;

[0011] Preprocessing the obtained sea surface data;

[0012] A data set is constructed using the preprocessed sea surface data, and the data set is divided; the divided data set includes a training set, a first test set, and a second test set;

[0013] The training set is input into the KC-ChlDiff model for model training, and the accuracy of the KC-ChlDiff model is evaluated using the first test set; wherein the KC-ChlDiff model is composed of a variational autoencoder and a knowledge control module, specifically, the variational autoencoder is used to encode high-dimensional observation data into low dimensions, and the knowledge control module adjusts the conversion probability in the generation process through training learning to generate a prediction result for sea surface chlorophyll a;

[0014] The second test set was input into the qualified KC-ChlDiff model to predict the sea surface chlorophyll a concentration.

[0015] Furthermore, the sea surface data includes chlorophyll a concentration data, sea surface temperature, ocean current velocity and salinity data.

[0016] Furthermore, the preprocessing includes difference processing and normalization processing; wherein, when performing difference processing, the key parameters of the data are determined with the ocean surface chlorophyll a concentration data as the center; and difference processing is performed on other sea surface data of the same date, longitude and latitude.

[0017] Furthermore, the sea surface data are used to construct a data set, including: integrating the sea surface data into the input feature matrix of the KC-ChlDiff model, where each row of data represents an independent observation sample with a time scale of one day.

[0018] Furthermore, the variational autoencoder first encodes the high-dimensional observation data into a low-dimensional latent space, and then trains a conditional latent diffusion model on this compressed latent space; wherein the conditional latent diffusion model is based on the Earthformer-UNet network architecture; the Earthformer-UNet network architecture includes a self-cube attention mechanism for processing spatial-temporal data.

[0019] Furthermore, the expected sea surface chlorophyll a concentration is used as the knowledge control to simulate the extremely unbalanced chlorophyll concentration distribution in different regions.

[0020] Furthermore, in the inverse denoising process of the latent diffusion model, the knowledge control module first receives the latent variables and context information of the current time step as input; then, it adjusts the transformation distribution of the latent variables by estimating the deviation between the latent variables and the physical constraints.

[0021] The second aspect of the present invention provides a multi-channel sea surface chlorophyll a concentration prediction system based on knowledge control.

[0022] The multi-channel sea surface chlorophyll a concentration prediction system based on knowledge control includes:

[0023] The data acquisition module is configured to: collect sea surface data;

[0024] The preprocessing module is configured to: preprocess the obtained sea surface data;

[0025] The data set division module is configured to: construct a data set using the preprocessed sea surface data, and divide the data set; the divided data set includes a training set, a first test set, and a second test set;

[0026] The model training module is configured to: input the training set into the KC-ChlDiff model for model training, and use the first test set to evaluate the accuracy of the KC-ChlDiff model; wherein the KC-ChlDiff model is composed of a variational autoencoder and a knowledge control module, specifically, the variational autoencoder is used to encode high-dimensional observation data into low dimensions, and the knowledge control module adjusts the conversion probability in the generation process through training learning to generate a prediction result for sea surface chlorophyll a;

[0027] The concentration prediction module is configured to: input the second test set into the qualified KC-ChlDiff model to predict the sea surface chlorophyll a concentration.

[0028] The third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps in the multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as described in the first aspect of the present invention.

[0029] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as described in the first aspect of the present invention are implemented.

[0030] One or more of the above technical solutions have the following beneficial effects:

[0031] The present invention uses sea surface data to construct a data set and divides it into a training set, a first test set, and a second test set. The training set is input into the KC-ChlDiff model for model training, and the accuracy of the KC-ChlDiff model is evaluated using the first test set; the second test set is input into the KC-ChlDiff model that has passed the evaluation to predict the sea surface chlorophyll a concentration. In the deep learning problem of spatiotemporal prediction of sea surface chlorophyll a, the potential diffusion model is applied, and the uncertainty in the basic data distribution is captured through the denoising diffusion process. Therefore, the present invention can avoid simply averaging all possibilities into fuzzy predictions, and use the expected chlorophyll a concentration as knowledge control to simulate the extremely unbalanced chlorophyll concentration distribution in the region; controllable generation under the guidance of prior knowledge helps to improve the accuracy of deep learning of spatiotemporal prediction of sea surface chlorophyll, and thus achieve accurate prediction of sea surface chlorophyll a concentration.

[0032] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] 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.

[0034] Figure 1 Schematic diagram of the structure of the KC-ChlDiff model in Example 1 of the present invention.

[0035] Figure 2 This is a schematic diagram of the data logic structure in the first embodiment of the present invention.

[0036] Figure 3 Schematic diagram of the structure of the knowledge control module in the first embodiment of the present invention.

[0037] Figure 4 It is a scatter plot of the chlorophyll concentration predicted by the ChlDiff model in Example 1 of the present invention throughout the year and in each season.

[0038] Figure 5 This is a scatter plot of the chlorophyll concentration predicted by the KC-ChlDiff model in Example 1 of the present invention throughout the year and in each season. DETAILED DESCRIPTION

[0039] It should be noted that the following detailed descriptions are exemplary and are 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.

[0040] It should be noted that the terms used herein are for describing specific embodiments only and are not intended to be limiting of exemplary embodiments according to the present invention.

[0041] In the absence of conflict, the embodiments of the present invention and the features of the embodiments may be combined with each other.

[0042] The overall idea of ​​the present invention is: the present invention provides a multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control, using the KC-ChlDiff model as a conditional potential diffusion model, and optimizing the prediction performance through a two-stage process: first, ChlDiff uses the conditional potential diffusion model for probability prediction, and this model is trained in a compressed latent space, which effectively reduces the computational cost; second, KC-ChlDiff introduces an explicit knowledge alignment mechanism, combining the prediction with the physical constraints of a specific field, thereby improving the accuracy and reliability of the prediction. Among them, the KC-ChlDiff model is a multi-channel spatiotemporal prediction deep learning diffusion model based on knowledge control.

[0043] Embodiment 1

[0044] This embodiment discloses a multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control.

[0045] The multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control includes:

[0046] Step S1, collecting sea surface data;

[0047] Step S2, preprocessing the obtained sea surface data;

[0048] Step S3, constructing a data set using the preprocessed sea surface data, and dividing the data set; the divided data set includes a training set, a first test set, and a second test set;

[0049] Step S4, inputting the training set into the KC-ChlDiff model for model training, and using the first test set to evaluate the accuracy of the KC-ChlDiff model; wherein the KC-ChlDiff model is composed of a variational autoencoder and a knowledge control module, specifically, the variational autoencoder is used to encode high-dimensional observation data into low dimensions, and the knowledge control module adjusts the conversion probability in the generation process through training learning to generate a prediction result for sea surface chlorophyll a;

[0050] Step S5: input the second test set into the KC-ChlDiff model that has been evaluated as qualified to predict the sea surface chlorophyll a concentration.

[0051] Based on the above process, the present application can combine prior knowledge and generate samples for extreme situations under the guidance of knowledge control, thereby achieving accurate prediction of sea surface chlorophyll a concentration. In order to facilitate the understanding of the technical solution of the present application, the following further explains and illustrates each specific implementation step in the technical solution of the present invention.

[0052] Step S1, collecting sea surface data.

[0053] Marine ecosystems are one of the largest ecosystems on Earth due to their complex and extensive characteristics, and the content of chlorophyll is a key indicator of their health. Accurate prediction of chlorophyll concentration in the ocean is essential for monitoring ecological changes, promoting scientific research, and supporting decision-making. In addition, chlorophyll prediction technology also plays an important role in monitoring the marine environment, protecting water quality, and managing fishery resources. However, traditional methods for predicting marine chlorophyll concentration only consider changes in chlorophyll concentration, while ignoring the impact of other factors on marine chlorophyll concentration, such as sea surface temperature, ocean current speed, and salinity data. The correlation between these ocean parameters and chlorophyll concentration is reflected in their joint influence on the growth environment of phytoplankton and the distribution of nutrients. 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 to promote phytoplankton reproduction; increases or decreases in sea surface temperature directly affect the metabolic rate and growth cycle of phytoplankton. The seasonal changes and geographic distribution patterns of these physical oceanographic parameters are closely linked to the temporal and spatial variations in chlorophyll concentrations, which together determine the productivity and biodiversity of marine ecosystems.

[0054] In this embodiment, in order to avoid the above problems of the prior art, the sea surface data to be collected include chlorophyll a concentration data, sea surface temperature, ocean current velocity and salinity data. Specifically, satellite remote sensing is used to collect chlorophyll a concentration data, and biogeochemical buoys are used to obtain sea surface temperature, ocean current velocity and salinity data.

[0055] When performing actual data collection, this embodiment collects data from a time range of January 1, 2015 to December 31, 2022, and a spatial range of the eastern and southern regions of China at 23°N-41°N and 116°E-132°E.

[0056] Step S2, preprocessing the obtained sea surface data; wherein the preprocessing includes difference processing and normalization processing.

[0057] Step S2-1: Difference processing

[0058] Specifically, key parameters are extracted from data such as sea surface temperature, ocean current velocity, and various salinities, including sampling date, longitude, and latitude. Taking the corresponding linear interpolation of ocean surface chlorophyll a concentration as an example, the specific method is as follows:

[0059] Firstly, the key parameters of the data are determined with the ocean surface chlorophyll a concentration data as the center. The key parameters include longitude, latitude, sampling date, etc., from which other data elements are obtained. The sea surface data with the same date, longitude and latitude are then interpolated to achieve uniform distribution of data in each channel.

[0060] Step S2-2: Normalization

[0061] Before dividing the data set, the interpolated sea surface chlorophyll a concentration data was normalized. The normalization formula is:

[0062]

[0063] Among them, Z represents the normalized sea surface chlorophyll a concentration value; x represents 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.

[0064] Step S3: construct a data set using the preprocessed sea surface data, and divide the data set; the divided data set includes a training set, a first test set and a second test set.

[0065] The sea surface data are integrated into the input feature matrix of the KC-ChlDiff model, and the sea surface chlorophyll a concentration data is used as the target output of the model. In the dataset construction stage, each row of data represents an independent observation sample with a time scale of one day, which contains all the input features and the corresponding sea surface chlorophyll a concentration data.

[0066] In order to facilitate the subsequent evaluation of the generalization ability of the model, the data set is divided; among them, the data from 2015, 2016, 2017, 2018, and 2019 are used as training sets for model training and parameter optimization; the data from 2020 and 2021 are used as the first test set to evaluate the performance of the model on unseen data; the data from 2022 is used as the second test set for overall evaluation of model performance.

[0067] Step S4: input the training set into the KC-ChlDiff model for model training, and use the first test set to evaluate the accuracy of the KC-ChlDiff model.

[0068] Step S4-1: Model construction

[0069] like Figure 1 As shown in the figure, the KC-ChlDiff model is a multi-channel spatiotemporal prediction deep learning diffusion model based on knowledge control. It integrates a knowledge control module to enhance the model's learning representation of ocean data features and its ability to fit data under complex changes. Among them, the ChlDiff model encodes high-dimensional observation data into a low-dimensional latent space by a variational autoencoder (VAE), 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 UNet structure that contains a self-cube attention mechanism and can effectively process spatial and temporal data and simulate complex dynamic systems. On this basis, the KC-ChlDiff model introduces a knowledge control module, which learns how to adjust the transition probability in the generation process through training so that the generated prediction results can meet the domain-specific physical constraints. In this stage, the transition distribution of the model in the denoising step is adjusted to make the generated latent state more consistent with the laws of physics, thereby improving the physical interpretability and accuracy of the prediction.

[0070] Furthermore, the latent diffusion model consists of three parts, namely, the frame encoder, the frame decoder, and the knowledge alignment mechanism. The frame encoder E is used to encode the observation sequence y into the latent context Z cond , compressing the input data into a low-dimensional latent space for subsequent processing. The frame decoder D will generate the final potential future Z 0 Decode back to pixel space to get the final prediction The role is to restore the data in the latent space to an interpretable output. The knowledge alignment mechanism is used in the denoising process to adjust the transfer distribution at each step by estimating the deviation from the physical constraints, thereby ensuring that the generated predictions meet the domain-specific physical constraints.

[0071] Furthermore, the data flow of the KC-ChlDiff model can be divided into the following stages, namely: (1) input observation sequence y, such as meteorological observation data or image sequence; (2) frame encoder E encodes the input data y into a latent context Z cond ; (3) Initialize Gaussian noise Z T , and gradually denoises through the potential diffusion model to generate the potential future Z at each step t , and at each step, the knowledge alignment mechanism adjusts the transfer distribution according to the physical constraints; (4) The frame decoder D will eventually generate the potential future Z 0 Decoded into predictions in pixel space

[0072] Earth system observation data such as chlorophyll a are usually difficult to describe with simple physical rules, which poses a challenge to directly incorporating physical laws for knowledge control. However, if there is highly flexible knowledge control, knowledge control through auxiliary control becomes feasible. Specifically for the deviation correction prediction of chlorophyll, this embodiment uses the expected chlorophyll a concentration as knowledge control to simulate the extremely unbalanced chlorophyll concentration distribution in the region.

[0073] Specifically, in this embodiment, the average strength of the data sequence is expressed as I(x)∈R + In order to estimate the conditional quantification of future intensity, this embodiment trains a simple probabilistic time series prediction model with a parameter (Gaussian) distribution, namely:

[0074]

[0075] Among them, p τ () represents the conditional probability distribution generated under the given condition I(x); represents a normal distribution; μ τ represents the mean function, which depends on the current state and time step t; σ τ It is a parameter related to the time step t and is used to control the amount of noise added in each step.

[0076] In each context frame (Abbreviated as I(y j )) to predict the distribution of the future average intensity I(x). By defining and The ChlDiff model can be guided by knowledge control for extreme cases (such as expected intensity exceeding 2σ τ ) to generate samples. Among them, represents the predicted value of x, y represents the context information, represents a specific function, and n is used to represent the number of samples.

[0077] Furthermore, in the inverse denoising process of the latent diffusion model, the knowledge control module first receives the latent variable z at the current time step t and context information y as input; then, by estimating the latent variable z tThe knowledge control module adjusts the transformation distribution of the latent variables by taking into account the deviation between the latent states that may violate the physical constraints and the physical constraints. The physical constraints here refer to the specific domain knowledge introduced in the model prediction process to ensure that the prediction results conform to the physical laws and the prior knowledge in the actual application scenarios. This method of integrating data-driven and physical models effectively solves the problem of physically unreliable predictions that may occur in traditional deep learning models when dealing with complex physical systems. Specifically, it reduces the sampling probability of latent states that may violate physical constraints, while increasing the sampling probability of intermediate latent states that conform to the physical laws. After this adjustment, the knowledge control module outputs the new latent variable z t , for use in the subsequent denoising step.

[0078] Furthermore, the knowledge control module trains a neural network to parameterize an energy function, which is used to adjust the transition probability in each denoising step. Specifically, it can be achieved through the forward process, the backward process and the transition probability distribution:

[0079] First, the forward process is to gradually add Gaussian noise to the input data x to obtain a noise sequence, that is:

[0080]

[0081] Among them, q() represents the probability distribution, z T represents the noise sequence, β t represents a predetermined noise coefficient, N() represents a Gaussian distribution, and I represents a time step sequence.

[0082] Then, based on the backward process, the denoising model is learned to gradually restore the data and finally generate the target data z 0 ,Right now:

[0083]

[0084] Among them, p θ (z t-1 |z t ,z cond ) represents the learning denoising model, μ θ and∑ θ denote the mean and covariance learned by the neural network respectively.

[0085] Then, the knowledge alignment mechanism adjusts the transition distribution by estimating the deviation at each step. It calculates the deviation between the current prediction and the physical constraints and adjusts the transition distribution during the denoising process. Specifically, the transition probability distribution can be expressed as:

[0086]

[0087] in, represents the random variable zt The probability distribution of θ and represents the model parameters. θ () represents the random variable z t The probability distribution of F represents the bootstrap scale factor, Indicated by the parameter The control function depends on z t , time step t and condition y. The knowledge control module is applied to the denoising diffusion process of the latent diffusion model part of the ChlDiff model. It plays a role in each denoising step, and ensures that the generated latent state not only conforms to the data distribution but also conforms to the domain-specific physical constraints by adjusting the transition distribution of the latent variables. In this way, the prediction results obtained by the final decoding can better match the domain-related knowledge, improving the credibility and operational utility of the prediction.

[0088] 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. Among them, 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 leads to a low-salinity environment, while carrying rich nutrients, which promotes 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 of Bohai Sea, temperature is the main influencing factor of phytoplankton growth. In summer, due to the influence of water and sand 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. In the field of chlorophyll concentration prediction, this embodiment introduces multiple factors such as sea surface salinity, ocean current velocity, and sea surface temperature, which have a significant impact on the concentration of chlorophyll a. These environmental factors affect the concentration of chlorophyll a by affecting the growth and distribution of phytoplankton. 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; ocean current velocity affects the transportation and distribution of nutrients; and sea surface temperature is one of the main environmental factors affecting the growth of phytoplankton.

[0089] Compared with the prior art, this embodiment not only relies on data-driven learning, but also introduces the knowledge control module alignment in the ChlDiff model, which enables the model KC-ChlDiff of this embodiment to generate prediction results that are more in line 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, and improves the interpretability and credibility of the model through knowledge alignment. In other words, the difference from the prior art is that traditional prediction methods often do not fully consider the impact of these complex factors on the concentration of chlorophyll a, resulting in the prediction effect not being optimal; and the KC-ChlDiff model of this embodiment not only improves the accuracy of the prediction, but also enhances the interpretability of the model by introducing the knowledge control module alignment.

[0090] Furthermore, this embodiment utilizes the KC-ChlDiff model to comprehensively consider multiple environmental factors and combines it with the knowledge control module to achieve a more accurate and interpretable prediction of chlorophyll a concentration.

[0091] Among them, the ELU activation function formula used by the KC-ChlDiff model is as follows:

[0092]

[0093] Among them, x represents the input of the neuron, and α represents a positive decimal value, which is used to control the slope of the function in the negative input region. By adjusting the model parameters and optimizing the algorithm, the KC-ChlDiff 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.

[0094] Step S4-2: Model training parameters

[0095] When the preprocessed training set is input into the model to predict the chlorophyll a concentration data, in this embodiment, the batch size is set to 1 (which means that only one sample is processed each time training, and this setting is particularly suitable for scenarios with small data volume or limited memory).

[0096] The number of training epochs is set to 500, which means that the model will perform 500 complete passes over the training data. At the same time, Adam is selected as the optimizer, which is an optimization algorithm with adaptive learning rate function, which is very suitable for the training of deep learning models. The initial learning rate of the training is set to 0.0001, which helps the model achieve stable convergence in the early stage of training.

[0097] In terms of loss function, the absolute error loss (L1Loss) is used, which aims to minimize the absolute error between the predicted value and the true value. During the training process, the loss value is recorded regularly to observe the convergence of the model. At the same time, the performance of the model is evaluated using the validation set to ensure that the model has good generalization ability on unseen data.

[0098] When evaluating the accuracy of the model, we use the mean absolute error (MAE) and the root mean square error (RMSE). The following are the formulas for each indicator.

[0099] 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:

[0100]

[0101] 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:

[0102]

[0103] like Figure 1 , Figure 2 As shown in the figure, the linear interpolation dataset first passes through a 2×1 convolution kernel to convolve the input data in order to extract local features; then after batch normalization, activation and Dropout processing, it passes 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 processing, 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 output layer to generate the final prediction results.

[0104] Step S5: input the second test set into the KC-ChlDiff model that has been evaluated as qualified to predict the sea surface chlorophyll a concentration.

[0105] 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. Among them, the model includes the ChlDiff model without using the knowledge control network. The present invention uses the KC-ChlDiff model with the knowledge control network enabled to analyze with the above model, and performs annual average processing on the ocean surface chlorophyll a concentration data predicted by the model and the actual observation data, and uses the mean absolute error (MAE), root mean square error (RMSE) and two statistical indicators for comparative analysis, as shown in Table 1:

[0106] Table 1 RMSE and MAE values ​​for sea surface chlorophyll a concentration prediction

[0107]

[0108] As can be seen from Table 1, compared with the traditional ChlDiff model, the ChlDiff model with the introduction of the knowledge control network has significantly improved the annual average, spring and winter predictions. This shows that the use of the knowledge control module can effectively improve the model's prediction accuracy for ocean surface chlorophyll concentration. The specific data sources are shown in Table 2:

[0109] Table 2 Data sources

[0110]

[0111]

[0112] Further, Figure 4 and Figure 5 The scatter plots reflect the chlorophyll concentrations predicted by the ChlDiff model and the KC-ChlDiff model throughout the year and in each season. Figure 4 In the figure, A is a scatter plot of the whole year's data, and B, C, D, and E are scatter plots of the data for spring, summer, autumn, and winter respectively; Figure 5 In the figure, A is a scatter plot of data for the whole year, and B, C, D, and E are scatter plots of data for spring, summer, autumn, and winter, respectively. Figure 4 and Figure 5 It is not difficult to see that in the four seasons, the chlorophyll concentration is more concentrated in summer and autumn, and the extreme values ​​are more prominent; the chlorophyll concentration is relatively average in spring and winter. In low-concentration sea areas, the chlorophyll concentration is relatively higher than that in summer and autumn. The most important thing to note is that the KC-ChlDiff model prediction value and the true value of chlorophyll concentration are significantly better than the ChlDiff model. By comparison, it can be seen that the present invention can combine prior knowledge and generate samples for extreme situations under the guidance of knowledge control, thereby achieving accurate prediction of sea surface chlorophyll a concentration.

[0113] Embodiment 2

[0114] This embodiment discloses a multi-channel sea surface chlorophyll a concentration prediction system based on knowledge control.

[0115] The multi-channel sea surface chlorophyll a concentration prediction system based on knowledge control includes:

[0116] The data acquisition module is configured to: collect sea surface data;

[0117] The preprocessing module is configured to: preprocess the obtained sea surface data;

[0118] The data set division module is configured to: construct a data set using the preprocessed sea surface data, and divide the data set; the divided data set includes a training set, a first test set, and a second test set;

[0119] The model training module is configured to: input the training set into the KC-ChlDiff model for model training, and use the first test set to evaluate the accuracy of the KC-ChlDiff model; wherein the KC-ChlDiff model is composed of a variational autoencoder and a knowledge control module, specifically, the variational autoencoder is used to encode high-dimensional observation data into low dimensions, and the knowledge control module adjusts the conversion probability in the generation process through training learning to generate a prediction result for sea surface chlorophyll a;

[0120] The concentration prediction module is configured to: input the second test set into the qualified KC-ChlDiff model to predict the sea surface chlorophyll a concentration.

[0121] Embodiment 3

[0122] The purpose of this embodiment is to provide a computer-readable storage medium.

[0123] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as described in the first embodiment of the present disclosure.

[0124] Embodiment 4

[0125] The purpose of this embodiment is to provide an electronic device.

[0126] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as described in the first embodiment of the present disclosure are implemented.

[0127] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0128] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0129] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control, characterized in that: include: Collect sea surface data; Preprocessing the obtained sea surface data; Construct a data set using the preprocessed sea surface data and divide the data set; The divided data set includes a training set, a first test set, and a second test set; The training set is input into the KC-ChlDiff model for model training, and the accuracy of the KC-ChlDiff model is evaluated using the first test set; wherein the KC-ChlDiff model is composed of a variational autoencoder and a knowledge control module, specifically, the variational autoencoder is used to encode high-dimensional observation data into low dimensions, and the knowledge control module adjusts the conversion probability in the generation process through training learning to generate a prediction result for sea surface chlorophyll a; The second test set was input into the qualified KC-ChlDiff model to predict the sea surface chlorophyll a concentration.

2. The multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as claimed in claim 1, characterized in that: The sea surface data include chlorophyll a concentration data, sea surface temperature, ocean current velocity and salinity data.

3. The multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control according to any one of claims 1 to 2, characterized in that: The preprocessing includes difference processing and normalization processing; wherein, when performing difference processing, the key parameters of the data are determined with the ocean surface chlorophyll a concentration data as the center; and difference processing is performed on other sea surface data with the same date, longitude and latitude.

4. The multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as claimed in claim 1, characterized in that: The data set is constructed using sea surface data, including: integrating the sea surface data into the input feature matrix of the KC-ChlDiff model, where each row of data represents an independent observation sample with a time scale of one day.

5. The multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as claimed in claim 1, characterized in that: The variational autoencoder first encodes high-dimensional observation data into a low-dimensional latent space, and then trains a conditional latent diffusion model on this compressed latent space; wherein the conditional latent diffusion model is based on the Earthformer-UNet network architecture; the Earthformer-UNet network architecture includes a self-cube attention mechanism for processing spatial-temporal data.

6. The multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as claimed in claim 1, characterized in that: The expected sea surface chlorophyll a concentration is used as the knowledge control to simulate the extremely unbalanced chlorophyll concentration distribution in different regions.

7. The multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as claimed in claim 6, characterized in that: In the inverse denoising process of the latent diffusion model, the knowledge control module first receives the latent variables and context information of the current time step as input; then, it adjusts the transformation distribution of the latent variables by estimating the deviation between the latent variables and the physical constraints.

8. A multi-channel sea surface chlorophyll a concentration prediction system based on knowledge control, characterized in that: include: The data acquisition module is configured to: collect sea surface data; The preprocessing module is configured to: preprocess the obtained sea surface data; The data set division module is configured to: construct a data set using the preprocessed sea surface data, and divide the data set; the divided data set includes a training set, a first test set, and a second test set; The model training module is configured to: input the training set into the KC-ChlDiff model for model training, and use the first test set to evaluate the accuracy of the KC-ChlDiff model; wherein the KC-ChlDiff model is composed of a variational autoencoder and a knowledge control module, specifically, the variational autoencoder is used to encode high-dimensional observation data into low dimensions, and the knowledge control module adjusts the conversion probability in the generation process through training learning to generate a prediction result for sea surface chlorophyll a; The concentration prediction module is configured to: input the second test set into the qualified KC-ChlDiff model to predict the sea surface chlorophyll a concentration.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of the multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps in the multi-channel sea surface chlorophyll a concentration prediction method based on knowledge control as described in any one of claims 1 to 7 are implemented.

Images