Marine surface salinity prediction method and system for long time sequence

Abstract

The invention belongs to the technical field of microwave remote sensing and detection, and particularly relates to an ocean surface salinity prediction method and system for a long time sequence. Monthly salinity data is input into an input gate and a forgetting gate to obtain an input value and a forgetting value; generating a multi-scale periodic coding vector, inputting the multi-scale periodic coding vector into a full-connection neural network, and performing dimension expansion to obtain a gating tensor; extracting trend features from the front cell state; obtaining a current hidden state corresponding to the current cell state according to the trend characteristics and the gating tensor, and carrying out restoration reconstruction to obtain current salinity data; inputting the current salinity data into the prediction model for training, modulating the prediction model according to the loss function to obtain prediction model parameters when the loss function is minimum, and adjusting the prediction model according to the prediction model parameters to obtain an optimized prediction model; and inputting the monthly salinity data set into the optimization prediction model to obtain a predicted salinity value. The accuracy of ocean surface salinity prediction can be improved without a large number of parameters.

Description

Technical Field

[0001] The present invention belongs to the field of microwave remote sensing and detection technology, and in particular relates to a method and system for predicting ocean surface salinity for long time series. Background Art

[0002] In recent years, with global climate change and the increasing complexity of marine ecosystems, accurate prediction of ocean salinity has become increasingly important. Sea surface salinity (SSS) is a key indicator of the interaction between the ocean and the atmosphere. Changes in salinity affect seawater density, thereby influencing global ocean circulation and the thermohaline circulation system. Predicting SSS can improve climate change forecasts. SSS prediction is of great significance in climate research, extreme weather forecasting, and marine ecological protection.

[0003] In related technologies, ocean surface salinity prediction primarily relies on physical and statistical models. Physical models, based on fundamental principles of oceanography, fluid dynamics, and thermodynamics, simulate a large number of ocean parameters to predict the distribution and variation of salinity. Statistical models, on the other hand, do not directly consider physical processes but instead analyze historical observational data (such as satellite remote sensing and buoy data) to establish statistical relationships between salinity and related variables (such as temperature, wind patterns, and precipitation).

[0004] Regarding the aforementioned technologies, physical models require a large number of oceanographic parameters when constructing them. These parameters vary significantly across different ocean regions, depths, temperatures, and salinity conditions, making accurate measurement challenging. Statistical models rely on historical data to identify patterns and establish relationships, and data quality and quantity impact model performance. Data with errors, omissions, or non-representativeness can lead to poor model accuracy and generalization capabilities. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and system for predicting ocean surface salinity over a long time series, which can improve the accuracy of ocean surface salinity prediction without requiring a large number of parameters.

[0006] A method for predicting ocean surface salinity for long time series, comprising: Obtaining the daily brightness temperature data within a preset time period and preprocessing the daily brightness temperature data to obtain a daily brightness temperature dataset; The daily brightness temperature dataset is inverted to obtain the corresponding monthly salinity data. The monthly salinity data at different times constitute a monthly salinity dataset, which includes the salinity data corresponding to each geographical grid position within the set regional longitude and latitude within one month; Combining the monthly salinity data sets according to time steps to obtain a salinity time series, and inputting the salinity time series into an input gate and a forget gate in sequence according to time steps to obtain an input value and a forget value; generating a multi-scale periodic encoding vector according to the time step; Inputting the multi-scale periodic encoding vector into a fully connected neural network to obtain a periodic auxiliary gate vector, and performing dimension expansion on the periodic auxiliary gate vector to obtain a gated tensor; Get the previous cell state of the previous time step. The cell state is used to describe the degree of forgetting and retention of the salinity time series. extracting trend features from the pre-cell state; Obtaining a current cell state according to the trend feature, the gating tensor, the input value, and the forgetting value; activating the current cell state to obtain a current hidden state, and restoring and reconstructing the current hidden state to obtain current salinity data; Inputting the current salinity data into a prediction model to obtain a predicted output value, calculating a loss value based on the predicted output value and a training value, adjusting the prediction model based on the loss value to obtain a prediction model parameter at which the loss function is minimized, and adjusting the prediction model based on the prediction model parameter to obtain an optimized prediction model; The actual monthly salinity data set is input into the optimized prediction model to obtain the predicted salinity value.

[0007] Optionally, the inversion of the daily brightness temperature dataset to obtain corresponding monthly salinity data, and the monthly salinity data at different times forming the monthly salinity dataset include: Correcting the daylight temperature data to obtain corrected data; Performing region selection and normalization operations on the correction data to obtain normalized data; The corresponding monthly salinity data are obtained by inverting the normalized data, and the monthly salinity data at different times constitute the monthly salinity dataset.

[0008] Optionally, combining the monthly salinity datasets by time step to obtain a salinity time series, inputting the salinity time series into an input gate and a forget gate in sequence by time step to obtain input values ​​and forget values ​​includes: Set the time step; Combining the monthly salinity datasets by time step to obtain a salinity time series; Get the hidden state of the previous time step; The salinity time series and the hidden state of the previous time step are input into the input gate and the forget gate in sequence according to the time step to obtain the input value and the forget value, which are expressed as: in, is the input value, represents the forgetting value, represents the convolution operation, Represents the Sigmoid activation function, with an output range of [0,1], 、 、 and They are the convolution kernels gated by input and historical state, is the hidden state of the previous time step, and is the bias term, is the salinity time series.

[0009] Optionally, generating a multi-scale periodic coding vector according to the time step includes: Obtain multi-scale periodic coding vector formula; The time step is input into the multi-scale periodic coding vector formula to obtain a multi-scale periodic coding vector. The multi-scale periodic coding vector formula is expressed as: in, is the defined period, k is the scale number, k=1, 2 or 3, t represents the difference between the current year and the starting year, is a multi-scale periodic coding vector, and the starting year is the earliest year in the monthly salinity dataset.

[0010] Optionally, inputting the multi-scale periodic encoding vector into a fully connected neural network to obtain a periodic auxiliary gate vector, and dimensionally expanding the periodic auxiliary gate vector to obtain a gated tensor includes: The multi-scale periodic encoding vector is input into a two-layer fully connected neural network to obtain a periodic auxiliary gate vector, and the periodic auxiliary gate vector is obtained, which is expressed as: in, and is a fully connected network mapping, and is the bias item that matches the current spatial state, is the Sigmoid activation function, Used for nonlinear transformation, so that the model can process periodic encoding information more flexibly, is a multi-scale periodic coding vector; The dimension of the periodic auxiliary gate vector is expanded to obtain a gated tensor.

[0011] Optionally, the trend feature extracted from the previous cell state is expressed as: in, is the convolution kernel used for trend extraction, For the pre-cell state, is the hyperbolic tangent activation function, with an output range of [-1, 1], As a trend feature, is the bias term.

[0012] Optionally, the current cell state is expressed as follows according to the trend feature and the gating tensor: in, For the pre-cell state, represents the forgetting value, For periodic response, is element-wise multiplication, is the input value.

[0013] A long-term ocean surface salinity prediction system, comprising: The first acquisition module is used to acquire the daily brightness temperature data within a preset time period and pre-process the daily brightness temperature data to obtain a daily brightness temperature dataset; The second acquisition module is used to invert the daily brightness temperature dataset to obtain corresponding monthly salinity data. The monthly salinity data at different times constitute a monthly salinity dataset. The monthly salinity data includes salinity data corresponding to each geographic grid position within a set regional longitude and latitude range within a month; A first calculation module is used to combine the monthly salinity data sets according to time steps to obtain a salinity time series, and input the salinity time series into an input gate and a forget gate in sequence according to time steps to obtain an input value and a forget value; A second calculation module is used to generate a multi-scale periodic code vector according to the time step; a gated tensor generation module, configured to input the multi-scale periodic encoding vector into a fully connected neural network to obtain a periodic auxiliary gate vector, and perform dimension expansion on the periodic auxiliary gate vector to obtain a gated tensor; The third acquisition module is used to obtain the previous cell state of the previous time step. The cell state is used to describe the degree of forgetting and retention of the salinity time series; an extraction module, configured to extract trend features from the pre-cell state; A third calculation module is used to obtain the current cell state according to the trend feature, the gating tensor, the input value and the forgetting value; A reconstruction module is used to activate the current cell state to obtain a current hidden state, and restore and reconstruct the current hidden state to obtain current salinity data; an optimization module, configured to input the current salinity data into a prediction model to obtain a predicted output value, calculate a loss value based on the predicted output value and a training value, adjust the prediction model based on the loss value, obtain prediction model parameters when the loss function is minimized, and adjust the prediction model based on the prediction model parameters to obtain an optimized prediction model; The prediction module is used to input the actual monthly salinity data set into the optimized prediction model to obtain a predicted salinity value.

[0014] A terminal device includes a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor loads and executes the computer program, a method for predicting ocean surface salinity for a long time series is adopted.

[0015] A computer-readable storage medium stores a computer program. When the computer program is loaded and executed by a processor, a method for predicting ocean surface salinity for a long time series is adopted.

[0016] The beneficial effects of the present invention are: By inputting monthly salinity data into an input gate and a forget gate to obtain input values ​​and forget values, a multi-scale periodic encoding vector is generated according to the time step; the multi-scale periodic encoding vector is input into a fully connected neural network to obtain a periodic auxiliary gate vector, and the dimension is expanded to obtain a gated tensor; trend features are extracted from the previous cell state; the current cell state is obtained based on the trend features and the gated tensor; the current cell state is activated to obtain the current hidden state and restored and reconstructed to obtain the current salinity data; the current salinity data is input into a prediction model to obtain a predicted output value; a loss function is calculated based on the predicted output value and the actual value; the prediction model is modulated according to the loss function to obtain the prediction model parameters when the loss function is minimized; the prediction model is adjusted according to the prediction model parameters to obtain an optimized prediction model; the actual monthly salinity data is input into the optimized prediction model to obtain a predicted salinity value. This application can improve the accuracy of ocean surface salinity prediction without requiring a large number of parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 Schematic diagram of the structure of the convolution-memory fusion prediction network of the present invention; Figure 2 This is the forecast result diagram and actual salinity distribution diagram for July 2020 of the present invention. DETAILED DESCRIPTION

[0018] A method for predicting ocean surface salinity for long time series, comprising: S1. Acquire the daylight temperature data within a preset time period and preprocess the daylight temperature data to obtain a daylight temperature dataset; S2. Inverting the daily brightness temperature dataset to obtain corresponding monthly salinity data. The monthly salinity data at different times constitute a monthly salinity dataset, wherein the monthly salinity data includes salinity data corresponding to each geographic grid position within a set regional longitude and latitude range within a month; The corresponding monthly salinity data are obtained by inverting the daily brightness temperature dataset. The monthly salinity data at different times constitute the monthly salinity dataset, which includes: Correct the daylight temperature data to obtain the corrected data; Perform region selection and normalization operations on the correction data to obtain normalized data; The corresponding monthly salinity data are obtained by inverting the normalized data, and the monthly salinity data at different times constitute the monthly salinity dataset.

[0019] Specifically, the daily brightness temperature data of recent years observed by current satellites are obtained. The data include the longitude and latitude corresponding to the grid points in the target area, the ocean and land mask, and the brightness temperature information. The daily brightness temperature data are preprocessed, including data correction, region selection, and normalization operations, to form a brightness temperature dataset for the target ocean area.

[0020] The existing inversion model is used to invert the salinity of each month in the past years into a monthly salinity dataset, which is then cleaned, transformed, sorted and normalized to form a three-dimensional monthly salinity dataset organized by month for the target ocean area. Specifically, a single sample of the sea surface salinity dataset for to Year Monthly salinity data for a month. The size of a single monthly salinity data is X (latitude coordinate index) × Y (longitude coordinate index) × 1 (time step). Each element in the monthly salinity data represents the sea surface salinity value for a specific geographic location and each day of that month. in, is the monthly salinity dataset, y is the year, T is the time step, X is the latitude coordinate index, Y is the longitude coordinate index, and R is a real number set.

[0021] S3. Combine the monthly salinity data sets by time step to obtain a salinity time series. Input the salinity time series into the input gate and the forget gate in sequence by time step to obtain the input value and the forget value. Combine the monthly salinity data sets by time step to obtain the salinity time series. Input the salinity time series into the input gate and forget gate in sequence according to the time step. The input values ​​and forget values ​​obtained include: Set the time step; Based on the time step, the monthly salinity datasets are combined according to the time step to obtain the salinity time series; Get the hidden state of the previous time step; The salinity time series and the hidden state of the previous time step are input into the input gate and the forget gate in sequence according to the time step to obtain the input value and the forget value, which are expressed as: in, is the input value, represents the forgetting value, represents the convolution operation, Represents the Sigmoid activation function, with an output range of [0,1], 、 、 and They are the convolution kernels gated by input and historical state, is the hidden state of the previous time step, is the salinity time series.

[0022] At each time step, the input gate and forget gate are respectively input with the current frame With historical hidden state As input, after convolution operation and sigmoid activation function Generate gated tensors and .in, Controls the degree to which new input information is written into the current memory state. The threshold is [0,1]. The larger the value is, the closer it is to 1, indicating that the current input The more important it is, the smaller the value is, the closer it is to 0, indicating that the current input Ignored; Then determine the memory state of the previous time step The greater the value, the more the current historical memory will be passed on and retained. The smaller the value, the more the current time step "forgets" the past content and discards ; Represents the convolution operation; 、 、 and They are the convolution kernels of the input and historical state gates, and the sizes of the convolution kernels are 、 、 and , here the convolution kernel size k=3; and is the bias term, is the number of gated output channels, C is the number of hidden channels; is a hidden state obtained by cell state transitions. Through the gating mechanism, the model can dynamically adjust the fusion of new and old information and effectively model the temporal continuity and spatial evolution pattern of the salinity field.

[0023] At each time step, input With the previous hidden state Together they determine the gating mechanism (input gate , Forget Gate , periodic gate and candidate status Then, the gating mechanism controls the inflow and forgetting of information to generate the updated cell state. Serves as the long-term memory basis for the current time step and is used to generate new hidden states , thus providing support for the next time step.

[0024] Specifically, a convolution-memory fusion prediction network was constructed by innovatively combining the spatiotemporal distribution characteristics of long-term sea surface salinity data. This network structure integrates the mechanisms of convolutional neural networks (CNN) and long short-term memory networks (LSTM). Compared with the traditional single-layer ConvLSTM model, this structure closes the output gate while maintaining the integrity of the spatial structure, and innovatively introduces an additional gating mechanism that combines multi-scale periodic auxiliary information to process the long-term trend information of salinity changes. This mechanism introduces time auxiliary information in the design and embeds it into the gating calculation process, so that the model has the ability to perceive time phase and periodic modulation, thereby being more accurate and stable in predicting long-term sea surface salinity trends, and more suitable for high-precision prediction of long-term series of ocean variables. The structure of the convolution-memory fusion prediction network is as follows: Figure 1 shown.

[0025] Input layer: Reconstructs the input sea surface salinity dataset to match the network’s input format requirements. Specifically, the monthly salinity dataset , needs to be combined into a single-channel data sequence according to the time series, and the reconstructed input size is The size of the sea surface salinity sequence can facilitate the subsequent simultaneous extraction of time and space dimension features, so that the output layer can generate a single channel salinity prediction value corresponding to the target that needs to be predicted. .

[0026] Multimodal spatiotemporal fusion layer: This layer consists of multiple spatiotemporal fusion units (i.e., optimized ConvLSTM units). Unlike traditional ConvLSTM units, each unit disables the output gate mechanism and adds a gating mechanism that incorporates multi-scale periodic auxiliary information. Specifically, this layer processes the reconstructed input sea surface salinity sequence as follows: 1. Time step input: Input the three-dimensional salinity sequence Each time step is fed into the spatiotemporal fusion unit frame by frame. A two-dimensional salinity sequence is input at each time step , whose size is [1, X, Y], and combined with the hidden state of the previous time step (of size [C,X,Y], where C is the number of hidden channels) with the cell state (size is [C, X, Y]) and participate in the status update at the current moment.

[0027] 2. Gating mechanism calculation and state update: Within each time step, each spatiotemporal fusion unit completes the following operations in sequence: ① The calculation input gate and forget gate control the degree of adoption of new memory and the degree of preservation of historical state by the current input, and the control gate controls the current input Whether it is adopted, the forget gate determines the memory of the previous moment Whether to continue to retain.

[0028] Specifically, Indicates the input value, which controls how much the current input is written. Indicates the forgetting value, which controls how much historical memory is retained.

[0029] S4, generating a multi-scale periodic coding vector according to the time step; Generating a multi-scale periodic encoding vector based on the time step includes: Obtain multi-scale periodic coding vector formula; Input the time step into the multi-scale periodic coding vector formula to obtain the multi-scale periodic coding vector. The multi-scale periodic coding vector formula is expressed as: in, is the defined period, k is the scale number, k=1, 2 or 3, t represents the difference between the current year and the starting year, is a multi-scale periodic coding vector, and the starting year is the earliest year in the monthly salinity dataset.

[0030] Specifically, when K=1, it means the cycle changes year by year. =1, when K=2, it indicates mid-term changes , when K=3, it indicates a long-term trend of change = the year span of the entire data.

[0031] S5. Input the multi-scale periodic encoding vector into a fully connected neural network to obtain a periodic auxiliary gate vector, and expand the dimension of the periodic auxiliary gate vector to obtain a gated tensor; The multi-scale periodic encoding vector is input into the fully connected neural network to obtain the periodic auxiliary gate vector. The periodic auxiliary gate vector is dimensionally expanded to obtain the gated tensor including: The multi-scale periodic encoding vector is input into the two-layer fully connected neural network to obtain the periodic auxiliary gate vector, and the periodic auxiliary gate vector is obtained, which is expressed as: in, and is a fully connected network mapping, and is the bias item that matches the current spatial state, is the Sigmoid activation function, Used for nonlinear transformation, so that the model can process periodic encoding information more flexibly, is a multi-scale periodic coding vector; The dimension of the periodic auxiliary gate vector is expanded to obtain the gated tensor.

[0032] Specifically, calculate the cycle auxiliary gate vector , multi-scale periodic coding vector Input two layers of fully connected neural network to get periodic auxiliary gate vector , this vector can reflect the extent to which each channel should retain period trend information.

[0033] S6. Obtain the previous cell state of the previous time step. The cell state is used to describe the degree of forgetting and retention of the salinity time series. Specifically, the initial hidden state Set to 0 tensor, with cell state (size is [C,X,Y], where the initial cell state Set to 0 tensor.

[0034] S7, extract trend features from the previous cell state; Extracting trend features from the previous cell state is expressed as: in, is the convolution kernel used for trend extraction, For the pre-cell state, is the hyperbolic tangent activation function, with an output range of [-1, 1], As a trend feature, is the bias term.

[0035] S8. Obtaining a current cell state according to the trend feature, the gating tensor, the input value, and the forgetting value; Specifically, the periodic gate is weightedly fused with the current candidate state to form a new cell state .

[0036] According to the trend characteristics and the gating tensor, the current cell state is expressed as: in, For the pre-cell state, represents the forgetting value, For periodic response, is element-wise multiplication, is the input value.

[0037] Specifically, use periodic auxiliary gating tensors Adjust the trend characteristics to obtain the periodic response ,in, is the element-wise multiplication (Hadamard Product).

[0038]

[0039] S9, activating the current cell state to obtain a current hidden state, and restoring and reconstructing the current hidden state to obtain current salinity data; Specifically, the current cell state is activated to obtain the current hidden state, which is expressed as: in, is the current cell state, is the current hidden state, is the activation function Specifically, as time steps progress, the network model gradually accumulates temporal and spatial information, ultimately outputting a hidden state sequence of size [T, C, X, Y]—the dynamic representation of multimodal fusion. This sequence serves as the input feature for subsequent output layers, supporting sea surface salinity prediction at the target time step.

[0040] In order to decode the specific salinity value from the learned hidden features, a two-dimensional convolutional layer (Conv2D) is used as the output layer. The function of this layer is to convert each time step The hidden state of C channels , nonlinearly mapped and restored to a single channel, representing the current salinity data of the salinity series to be predicted , its operation is expressed as: in, is the current salinity data.

[0041] S10, inputting the current salinity data into a prediction model to obtain a predicted output value, calculating a loss value based on the predicted output value and a training value, adjusting the prediction model based on the loss value, obtaining a prediction model parameter when the loss function is minimized, and adjusting the prediction model based on the prediction model parameter to obtain an optimized prediction model; The loss value is calculated by the loss function, which is expressed as: in, is the loss value, yes The predicted salinity value at the coordinate corresponding to the location, yes The actual salinity label value of the coordinate corresponding to the position, is the mask matrix at position The land mask is 0 and the ocean mask is 1.

[0042] In the sea surface salinity prediction task, the data grid after region selection includes both ocean areas and land areas. The salinity data on land has no physical meaning. If all salinity data is used indiscriminately for loss calculation, the network model will learn invalid regional features, affecting the prediction performance. Using this loss function can make the network only learn the salinity data of the ocean area, thereby improving training accuracy and model generalization ability.

[0043] This loss function indirectly acts on the cell state at each time step in the model through the back propagation mechanism , thereby optimizing the entire spatiotemporal memory process and improving the accuracy of predictions of future sea surface salinity.

[0044] S11. Input the actual monthly salinity data set into the optimized prediction model to obtain the predicted salinity value.

[0045] Specifically, in this embodiment, brightness temperature data (including latitude and longitude grid information, land and sea masks, and brightness temperature observations) of the SMOS satellite in June, July, and August from 2015 to 2010 are acquired to form a continuous daily brightness temperature dataset.

[0046] The daily brightness temperature data is preprocessed, specifically by correcting the brightness temperature data and removing abnormal or distorted data; performing regional selection to extract the brightness temperature data of the target sea area; and performing normalization to map the brightness temperature values ​​to a unified scale to facilitate subsequent model training.

[0047] A continuous daily brightness temperature dataset for the target area is generated. The data structure is a three-dimensional array of [longitude, latitude, time]. The existing sea surface salinity inversion model (based on daily brightness temperature data) is used to invert the sea surface salinity data corresponding to June, July, and August from 2015 to 2020 on a monthly basis. This constructs a monthly salinity dataset for June, July, and August from 2015 to 2020.

[0048] The salinity data were normalized according to time and divided into training set, validation set and test set. 80% of the monthly salinity data from June and August 2015 to 2020 were used as the training set; 20% of the monthly salinity data from June and August 2015 to 2019 were used as the validation set; the monthly salinity data from July 2015 to 2020 were used as the test set, and the data from July 2015 to 2019 were used to predict the data in July 2020.

[0049] A convolution-memory fusion prediction network was constructed for training. The monthly salinity data was reconstructed into a sequence input format of [T, 1, X, Y]. The monthly salinity sequence of the training set was input and fed into the network time-step by time for feature learning. The mean squared error (MSE) was used as the loss function to calculate the error between the predicted salinity and the true salinity label. The network parameters were iteratively updated through the back-propagation algorithm. Training continued until the loss function converged or the set number of training rounds was reached.

[0050] Using the trained convolution-memory fusion prediction model, the salinity data sequence without training network structure, that is, the data from July 2015 to 2019, was input to predict the sea surface salinity distribution in July 2020. The prediction results were saved after denormalization and compared with the actual salinity observation data in July 2020. The mean square error (MSE) between the predicted value and the actual value was calculated, and a comparative distribution map of the predicted and actual values ​​of sea surface salinity in July 2020 was drawn.

[0051] Figure 2 The following graphs show the forecast results and actual salinity distribution for July 2020. These graphs reveal that while there are some numerical differences between the forecast and actual values, the overall trend is consistent. This demonstrates that the forecast model accurately captures salinity distribution trends, extracts the most important features from the data, and makes accurate predictions based on these features.

[0052] A system for predicting ocean surface salinity over long time series, including.

[0053] The first acquisition module is used to acquire the daily brightness temperature data within a preset time period and pre-process the daily brightness temperature data to obtain a daily brightness temperature dataset; The second acquisition module is used to invert the daily brightness temperature dataset to obtain corresponding monthly salinity data. The monthly salinity data at different times constitute a monthly salinity dataset. The monthly salinity data includes salinity data corresponding to each geographic grid position within a set regional longitude and latitude range within a month; A first calculation module is used to combine the monthly salinity data sets according to time steps to obtain a salinity time series, and input the salinity time series into an input gate and a forget gate in sequence according to time steps to obtain an input value and a forget value; A second calculation module is used to generate a multi-scale periodic code vector according to the time step; a gated tensor generation module, configured to input the multi-scale periodic encoding vector into a fully connected neural network to obtain a periodic auxiliary gate vector, and perform dimension expansion on the periodic auxiliary gate vector to obtain a gated tensor; The third acquisition module is used to obtain the previous cell state of the previous time step. The cell state is used to describe the degree of forgetting and retention of the salinity time series; an extraction module, configured to extract trend features from the pre-cell state; A third calculation module is used to obtain the current cell state according to the trend feature, the gating tensor, the input value and the forgetting value; A reconstruction module is used to activate the current cell state to obtain a current hidden state, and restore and reconstruct the current hidden state to obtain current salinity data; an optimization module, configured to input the current salinity data into a prediction model to obtain a predicted output value, calculate a loss value based on the predicted output value and a training value, adjust the prediction model based on the loss value, obtain prediction model parameters when the loss function is minimized, and adjust the prediction model based on the prediction model parameters to obtain an optimized prediction model; The prediction module is used to input the actual monthly salinity data set into the optimized prediction model to obtain a predicted salinity value.

[0054] An embodiment of the present application also discloses a terminal device, including a memory and a processor. The memory stores a computer program that can be run on the processor. When the processor loads and executes the computer program, a method for predicting ocean surface salinity for a long time series is adopted.

[0055] Among them, the terminal device can be a computer device such as a desktop computer, a laptop computer or a cloud server, and the terminal device includes but is not limited to a processor and a memory. For example, the terminal device can also include input and output devices, network access devices and buses, etc.

[0056] Among them, the processor can adopt a central processing unit (CPU). Of course, according to actual usage, other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. can also be adopted. The general-purpose processor can adopt a microprocessor or any conventional processor, etc., and this application does not impose any restrictions on this.

[0057] Among them, the memory can be an internal storage unit of the terminal device, such as the hard disk or memory of the terminal device, or it can be an external storage device of the terminal device, such as a plug-in hard disk, smart memory card (SMC), secure digital card (SD) or flash memory card (FC) equipped on the terminal device, etc., and the memory can also be a combination of the internal storage unit and the external storage device of the terminal device. The memory is used to store computer programs and other programs and data required by the terminal device. The memory can also be used to temporarily store data that has been output or is to be output. This application does not impose any restrictions on this.

[0058] Among them, through this terminal device, a method for predicting ocean surface salinity for a long time series in the above embodiment is stored in the memory of the terminal device, and is loaded and executed on the processor of the terminal device for easy use.

[0059] An embodiment of the present application further discloses a computer-readable storage medium, and the computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, a method for predicting ocean surface salinity for a long time series in the above embodiment is adopted.

[0060] Among them, the computer program can be stored in a computer-readable medium, the computer program includes computer program code, the computer program code can be in the form of source code, object code, executable file or certain middleware, etc. The computer-readable medium includes any entity or device that can carry computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that computer-readable medium includes but is not limited to the above-mentioned components.

[0061] Among them, through this computer-readable storage medium, a method for predicting ocean surface salinity for a long time series in the above embodiment is stored in a computer-readable storage medium, and is loaded and executed on a processor to facilitate the storage and application of the above method.

[0062] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0063] The one or more embodiments of this application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application should be included in the scope of protection of this application.

Claims

1. A method for predicting ocean surface salinity for long time series, characterized by: include: Obtaining the daily brightness temperature data within a preset time period and preprocessing the daily brightness temperature data to obtain a daily brightness temperature dataset; The daily brightness temperature dataset is inverted to obtain the corresponding monthly salinity data. The monthly salinity data at different times constitute a monthly salinity dataset, which includes the salinity data corresponding to each geographical grid position within the set regional longitude and latitude within one month; Combining the monthly salinity data sets according to time steps to obtain a salinity time series, and inputting the salinity time series into an input gate and a forget gate in sequence according to time steps to obtain an input value and a forget value; generating a multi-scale periodic encoding vector according to the time step; Inputting the multi-scale periodic encoding vector into a fully connected neural network to obtain a periodic auxiliary gate vector, and performing dimension expansion on the periodic auxiliary gate vector to obtain a gated tensor; Get the previous cell state of the previous time step. The cell state is used to describe the degree of forgetting and retention of the salinity time series. extracting trend features from the pre-cell state; Obtaining a current cell state according to the trend feature, the gating tensor, the input value, and the forgetting value; activating the current cell state to obtain a current hidden state, and restoring and reconstructing the current hidden state to obtain current salinity data; Inputting the current salinity data into a prediction model to obtain a predicted output value, calculating a loss value based on the predicted output value and a training value, adjusting the prediction model based on the loss value to obtain a prediction model parameter at which the loss function is minimized, and adjusting the prediction model based on the prediction model parameter to obtain an optimized prediction model; The actual monthly salinity data set is input into the optimized prediction model to obtain the predicted salinity value.

2. The method for predicting ocean surface salinity for a long time series according to claim 1, wherein: The corresponding monthly salinity data is obtained by inverting the daily brightness temperature dataset. The monthly salinity data at different times constitute the monthly salinity dataset, which includes: Correcting the daylight temperature data to obtain corrected data; Performing region selection and normalization operations on the correction data to obtain normalized data; The corresponding monthly salinity data are obtained by inverting the normalized data, and the monthly salinity data at different times constitute the monthly salinity dataset.

3. The method for predicting ocean surface salinity for a long time series according to claim 1, wherein: Combining the monthly salinity data sets according to time steps to obtain a salinity time series, inputting the salinity time series into an input gate and a forget gate in sequence according to time steps to obtain input values ​​and forget values ​​includes: Set the time step; Combining the monthly salinity datasets by time step to obtain a salinity time series; Get the hidden state of the previous time step; The salinity time series and the hidden state of the previous time step are input into the input gate and the forget gate in sequence according to the time step to obtain the input value and the forget value, which are expressed as: in, is the input value, represents the forgetting value, represents the convolution operation, Represents the Sigmoid activation function, with an output range of [0,1], 、 、 and They are the convolution kernels gated by input and historical state, is the hidden state of the previous time step, and is the bias term, is the salinity time series.

4. The method for predicting ocean surface salinity for a long time series according to claim 1, wherein: Generating a multi-scale periodic coding vector according to the time step includes: Obtain multi-scale periodic coding vector formula; The time step is input into the multi-scale periodic coding vector formula to obtain a multi-scale periodic coding vector. The multi-scale periodic coding vector formula is expressed as: in, is the defined period, k is the scale number, k=1, 2 or 3, t represents the difference between the current year and the starting year, is a multi-scale periodic coding vector, and the starting year is the earliest year in the monthly salinity dataset.

5. The method for predicting ocean surface salinity for a long time series according to claim 1, wherein: Inputting the multi-scale periodic encoding vector into a fully connected neural network to obtain a periodic auxiliary gate vector, and performing dimension expansion on the periodic auxiliary gate vector to obtain a gated tensor includes: The multi-scale periodic encoding vector is input into a two-layer fully connected neural network to obtain a periodic auxiliary gate vector, which is expressed as: in, and is a fully connected network mapping, and is the bias item that matches the current spatial state, is the Sigmoid activation function, Used for nonlinear transformation, so that the model can process periodic encoding information more flexibly, is a multi-scale periodic coding vector; The dimension of the periodic auxiliary gate vector is expanded to obtain a gated tensor.

6. The method for predicting ocean surface salinity for a long time series according to claim 1, wherein: The trend feature extracted from the previous cell state is expressed as: in, is the convolution kernel used for trend extraction, For the pre-cell state, is the hyperbolic tangent activation function, with an output range of [-1, 1], As a trend feature, is the bias term.

7. The method for predicting ocean surface salinity for a long time series according to claim 1, wherein: The current cell state is expressed as follows according to the trend characteristics and the gating tensor: in, For the pre-cell state, represents the forgetting value, For periodic response, is element-wise multiplication, is the input value.

8. A long-term ocean surface salinity prediction system, characterized by: include: The first acquisition module is used to acquire the daily brightness temperature data within a preset time period and pre-process the daily brightness temperature data to obtain a daily brightness temperature dataset; The second acquisition module is used to invert the daily brightness temperature dataset to obtain corresponding monthly salinity data. The monthly salinity data at different times constitute a monthly salinity dataset. The monthly salinity data includes salinity data corresponding to each geographic grid position within a set regional longitude and latitude range within a month; A first calculation module is used to combine the monthly salinity data sets according to time steps to obtain a salinity time series, and input the salinity time series into an input gate and a forget gate in sequence according to time steps to obtain an input value and a forget value; A second calculation module is used to generate a multi-scale periodic code vector according to the time step; a gated tensor generation module, configured to input the multi-scale periodic encoding vector into a fully connected neural network to obtain a periodic auxiliary gate vector, and perform dimension expansion on the periodic auxiliary gate vector to obtain a gated tensor; The third acquisition module is used to obtain the previous cell state of the previous time step. The cell state is used to describe the degree of forgetting and retention of the salinity time series; an extraction module, configured to extract trend features from the pre-cell state; A third calculation module is used to obtain the current cell state according to the trend feature, the gating tensor, the input value and the forgetting value; A reconstruction module is used to activate the current cell state to obtain a current hidden state, and restore and reconstruct the current hidden state to obtain current salinity data; an optimization module, configured to input the current salinity data into a prediction model to obtain a predicted output value, calculate a loss value based on the predicted output value and a training value, adjust the prediction model based on the loss value, obtain prediction model parameters when the loss function is minimized, and adjust the prediction model based on the prediction model parameters to obtain an optimized prediction model; The prediction module is used to input the actual monthly salinity data set into the optimized prediction model to obtain a predicted salinity value.

9. A terminal device comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor. When the processor loads and executes the computer program, the method according to any one of claims 1 to 7 is adopted.

10. A computer-readable storage medium storing a computer program, wherein: When the computer program is loaded and executed by a processor, the method according to any one of claims 1 to 7 is adopted.

Images