A method and system for predicting ocean temperature and salinity

Abstract

The present invention relates to a method and system for predicting ocean temperature and salinity, including using a four-dimensional convolution model to extract features of ocean temperature and salinity data in time and space dimensions, using a residual structure to further explore the spatial characteristics of ocean temperature and salinity, and then using recalibration weighting to quantify the contribution of each regional feature, thereby improving the model quality and realizing ocean temperature and salinity prediction. At the same time, the prediction results of ocean temperature and salinity are visualized in combination with visualization. The present invention combines a four-dimensional convolution model, a residual structure, and recalibration weighting, and considers that ocean temperature and salinity prediction is affected by both time series and internal spatial characteristics of the ocean, thereby overcoming the limitations of existing methods such as insufficient accuracy of ocean internal temperature and salinity prediction due to insufficient extraction of spatial relationships, and has theoretical and practical significance for the prediction of the overall ocean temperature and salinity and the prediction of the location of the thermocline.

Description

Technical Field

[0001] The invention belongs to the technical field of ocean prediction, and in particular to an ocean temperature and salinity prediction method and system. Background Art

[0002] As time goes by, changes in ocean temperature and salinity have a significant impact on marine ecosystems and global climate change. Many phenomena and processes occurring in the ocean are often related to changes in seawater temperature and salinity. Therefore, the study of the changing laws of ocean temperature and salinity occupies an important position in marine science. Studying and mastering the temporal and spatial distribution and changing laws of seawater temperature and salinity is an important part of oceanography, which is of great significance to marine fishing, aquaculture, and maritime operations, and is also important to meteorology, navigation, and hydroacoustics.

[0003] Ocean temperature and salinity prediction has always occupied an important position in the research of ocean-related fields. Current research is mainly based on the temperature and salinity of the ocean surface, but the prediction of the temperature and salinity of the ocean interior is more important in practical applications. At present, most of the research on the prediction of ocean temperature and salinity is based on time series, and few of them consider the dual characteristics of time and space. Therefore, the prediction accuracy of related methods is not high enough, especially for the prediction of thermocline and deep sea positions. Therefore, a prediction model that can fully consider the dual characteristics of time and space is of great significance to improve the accuracy of ocean temperature and salinity prediction. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method and system for predicting ocean temperature and salinity, which uses a four-dimensional convolution model to extract dual features of time and space from ocean temperature or salinity data, and then uses a residual network to further obtain the spatial features of ocean temperature or salinity. Finally, the feature results are input into a recalibration module, and each feature is weighted and summed to achieve ocean temperature or salinity prediction. The proposed method takes into account the three-dimensional rasterization characteristics of the ocean data itself, overcomes the limitations of existing temperature and salinity prediction methods, which usually consider the characteristics of time series but do not extract spatial relationships, etc., which has theoretical and practical significance in the field of ocean temperature and salinity prediction.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0006] A method for predicting ocean temperature and salinity, comprising the following steps:

[0007] Step 1: preprocessing the original ocean temperature or salinity real-time data at a fixed location within a period of time to meet the data dimension of the four-dimensional convolution model;

[0008] Step 2: Establish a four-dimensional convolution model, and perform four-dimensional convolution operations on the pre-processed ocean temperature or salinity real-time data in batches to fully extract the spatiotemporal characteristics of the ocean data;

[0009] Step 3: Establish a residual network composed of multiple basic residual units connected in series to further obtain spatiotemporal data of ocean temperature or salinity with prominent spatial characteristics;

[0010] Step 4, recalibrate and weight the acquired spatial characteristics of ocean temperature or salinity to predict ocean temperature or salinity;

[0011] Step 5, repeatedly iterate the above steps 2-4, select the Adam optimizer, use the mean absolute error as the loss function, and back propagate to adjust the parameters of the ocean temperature or salinity prediction model, so as to optimize the ocean temperature or salinity prediction model;

[0012] Step 6: collect the on-site ocean temperature or salinity of the fixed point to be predicted in real time, pre-process it, and input it into the optimized ocean temperature or salinity prediction model to realize ocean temperature or salinity prediction;

[0013] Step 7: Visually display the results of the ocean temperature or salinity prediction in the form of a two-dimensional plane or a three-dimensional grid image in the form of colors or charts.

[0014] The original ocean temperature or salinity real-time data is three-dimensional tensor temperature or salinity data of longitude, latitude and depth collected by a GPS positioning device, a depth sensor, a temperature sensor and a salinometer.

[0015] The pre-processing comprises the following steps:

[0016] Step 1-1, normalization, the mapping range is [-1, 1];

[0017] Step 1-2: Process the normalized original ocean temperature or salinity real-time data into a six-dimensional form (U×C×S×L×H×W) required for four-dimensional convolution;

[0018] Among them, U is the batch size of the data, C is the number of channels, S is the number of samples in the time series, and the data shape is L×H×W, where L is the number of data layers, H is the data height, and W is the data width;

[0019] The definition and selection principle of the parameters L, H, and W are as follows: in ocean data, the cross-sectional direction is expressed as: L is the depth range, H is the latitude range, and W is the longitude range; the profile direction is expressed as: L, H, and W are the ranges of latitude, depth, and longitude, respectively.

[0020] The four-dimensional convolution model is established, comprising the following steps:

[0021] Step 2-1, set the number of convolution kernels, convolution kernel size, step size parameters, set the activation function of the convolution layer to the relu function, and establish an M-layer four-dimensional convolutional neural network;

[0022] Step 2-2: During the calculation process, the six-dimensional data obtained in step 1 is input into an M-layer convolutional neural network; the four-dimensional convolution operation of each layer is completed in sequence, and finally the temperature or salinity data representing the dual characteristics of time and space of the ocean data is obtained.

[0023] Each layer of the four-dimensional convolutional neural network is built on the basis of the three-dimensional convolutional model, and the formula is as follows:

[0024]

[0025] Among them, b j is the bias term, c is the input channel, j is the output channel, R×X×Y×Z is the size of the convolution kernel, is the weight corresponding to the position (r,x,y,z) in the model, corresponding to input channel c and output channel j. V c is the input of position (s+r,l+x,h+y,w+z) in channel c, is the output of position (s,l,h,w) at channel j.

[0026] The four-dimensional convolution operation of each layer includes the following steps:

[0027] Since three dimensions in the six-dimensional data represent the shape of the data, one of the dimensions is expanded to obtain several five-dimensional data, and a three-dimensional convolution calculation is applied to each five-dimensional data. The calculated data are added according to the corresponding positions to complete a four-dimensional convolution operation; the formula for the four-dimensional convolution calculated by the three-dimensional convolution is as follows:

[0028]

[0029] The method of further obtaining ocean temperature or salinity spatiotemporal data with prominent spatial features by using the residual network comprises the following steps:

[0030] Step 3-1: The basic residual unit includes: activation unit→3D convolution→activation unit→3D convolution;

[0031] Step 3-2: Use the residual network to further obtain the spatial characteristics of ocean temperature or salinity; obtained by the following formula:

[0032] X l+1 =X l +F(X l θ l ) (2)

[0033] Among them, θ l is the set of all learnable parameters in the lth residual unit, X l is the input of the lth residual unit, X l+1is the output of the lth residual unit.

[0034] The spatial characteristics of the acquired ocean temperature or salinity are recalibrated and weighted summed to obtain the predicted temperature or salinity X c ,include:

[0035]

[0036] in, The input of feature k in channel c is the acquired spatiotemporal data of ocean temperature or salinity with prominent spatial features; for The corresponding weights; is an element-wise multiplication; by learning the parameters To quantify the effect of each feature in the space.

[0037] The visualization display of the two-dimensional plane is a temperature or salinity contour map of the ocean horizontal surface composed of longitude and latitude at the same depth, or a temperature and salinity contour map of the ocean profile composed of depth and longitude at the same latitude;

[0038] The visualization display in the form of the three-dimensional grid image is a stereoscopic map of the temperature or salinity contour lines of a certain sea area including the three dimensions of longitude, latitude and depth.

[0039] A system for predicting ocean temperature and salinity includes a sensor device, a processor, and a memory; the sensor device includes a GPS positioning device, a depth sensor, a temperature sensor, and a salinometer, and is used to send the collected three-dimensional tensor temperature or salinity data of longitude, latitude, and depth to the processor; the memory stores a program, and the processor reads the program to execute the method steps described above to realize ocean temperature and salinity prediction.

[0040] The present invention has the following beneficial effects and advantages:

[0041] The present invention provides a method and system for predicting ocean temperature and salinity. By combining a four-dimensional convolution model, a residual structure and a recalibration module, the four-dimensional convolution model is first used to extract dual features of time and space from ocean temperature or salinity data, and then the residual network is used to further obtain the spatial features of ocean temperature or salinity. Finally, the feature results are input into the recalibration module, and each feature is weighted and summed to achieve ocean temperature or salinity prediction. The three-dimensional rasterization characteristics of the ocean data itself are taken into account, and the limitations of existing temperature and salinity prediction methods, such as usually considering the characteristics of time series and insufficient extraction of spatial relationships, are overcome. This method has theoretical and practical significance for the field of ocean temperature and salinity prediction. The visualization display of the present invention can intuitively display the results of ocean temperature and salinity prediction in the form of two-dimensional planes or three-dimensional grid images in the form of colors or charts. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure is a flow chart of the method of the present invention.

[0043] Figure 2 This is a comparison chart of the temperature prediction results and the actual values ​​at 0-10m underwater.

[0044] Figure 3 This is a comparison chart of the temperature prediction results and the actual values ​​at 1000m-1200m underwater.

[0045] Figure 4 This is a comparison chart of the predicted results and actual values ​​of summer ocean profile temperature.

[0046] Figure 5 This is a comparison chart of the predicted results and actual values ​​of winter ocean profile temperature. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation method of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the invention, so the present invention is not limited by the specific implementation disclosed below.

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0049] like Figure 1 Shown is a flow chart of the method of the present invention.

[0050] The ocean temperature and salinity prediction method normalizes the measured ocean temperature or salinity data and forms six-dimensional data, uses four-dimensional convolution and residual network to extract dual features of time and space, and then uses a recalibration module to perform weighted summation on each feature to achieve ocean temperature and salinity prediction. The programming language used in the program execution step of the present invention is not limited to MATLAB, Python, etc.

[0051] The specific steps of the present invention are as follows:

[0052] Step 1: Data processing and normalization:

[0053] Step 1-1: Normalize the measured temperature or salinity data of the ocean using the Min-Max normalization method, with the mapping range being [-1, 1];

[0054] The Min-Max normalization method is used to normalize the ocean temperature or salinity data. The formula is as follows:

[0055]

[0056] Among them, x i It represents the measured data of ocean temperature or salinity. The measured data of ocean temperature or salinity is three-dimensional tensor data including longitude, latitude and depth characteristics.

[0057] Step 1-2: Process the data into six dimensions required for four-dimensional convolution, in the form of (U×C×S×L×H×W). U is the batch size of the data, C is the number of channels, S is the number of samples in the time series, and the data shape is L×H×W, where L is the number of data layers, H is the data height, and W is the data width. In ocean data, the cross-sectional direction is represented as: L is the depth range, H is the latitude range, and W is the longitude range; the profile directions are represented as: L, H, and W are the ranges of latitude, depth, and longitude, respectively. In ocean data, the number of channels is 1. After experiments, the number of samples in the time series is set to 9, and U is 537 (the original 546-day ocean temperature and salinity data). The purpose is to use the temperature or salinity data of the previous 9 days to predict the temperature or salinity of the 10th day, which can effectively extract time-dependent features and will not cause the accumulation of useless information due to the selection of too long time samples. The original 546-day ocean temperature or salinity data was processed into six-dimensional data in 537 batches.

[0058] Step 2: Establish a four-dimensional convolution model, and use the four-dimensional convolution model to perform a four-dimensional convolution operation on the six-dimensional data to fully extract the spatiotemporal characteristics of the ocean data;

[0059] Step 2-1: Use the four-dimensional convolution model to perform four-dimensional convolution operation on the six-dimensional data to fully extract the spatiotemporal characteristics of the ocean data.

[0060] The four-dimensional convolution model in this example is a three-layer four-dimensional convolutional neural network. Each layer is a four-dimensional convolution model based on the three-dimensional convolution model; the construction of the four-dimensional convolution model is obtained through the following formula:

[0061]

[0062] Among them, b j is the bias term, c is the input channel, j is the output channel, R×X×Y×Z is the size of the convolution kernel, is the weight corresponding to the position (r,x,y,z) in the model, corresponding to input channel c and output channel j. V c is the input of position (s+r,l+x,h+y,w+z) in channel c, is the output of position (s,l,h,w) at channel j.

[0063] The four-dimensional convolution can be realized by three-dimensional convolution, which is obtained by the following formula:

[0064]

[0065] Step 2-1, during the calculation process, the six-dimensional data obtained in step 1 is input into the first convolution layer. Since three dimensions in the six-dimensional data represent the shape of the data. Expand one of the dimensions to obtain several five-dimensional data. Apply three-dimensional convolution calculation to each five-dimensional data, and add the calculated data according to the corresponding position, which completes a four-dimensional convolution operation. The addition according to the corresponding position includes: after performing three-dimensional convolution of the same specification on each group of five-dimensional data, a multi-dimensional matrix of the same size will be obtained, and a group of five-dimensional data outputs a group of multi-dimensional matrices, and the obtained several groups of multi-dimensional matrices are added according to the elements of the corresponding positions, and the several groups of multi-dimensional matrices are turned into a group of multi-dimensional matrices, and the four-dimensional convolution operation is realized. In this way, the convolution operation of the three-layer four-dimensional convolutional neural network is completed in sequence, and 537 batches of temperature or salinity data representing the dual characteristics of time and space of ocean data are obtained.

[0066] Example: The four-dimensional convolution model is a three-layer four-dimensional convolutional neural network. The first layer uses 64 convolution kernels for each group of expanded five-dimensional data, with a convolution kernel size of (1, 3, 3) and a step size of (1, 1, 1). The calculated data is added according to the corresponding positions to complete the first layer of four-dimensional convolution. The second layer uses 64 convolution kernels with a convolution kernel size of (3, 3, 3) and a step size of (3, 1, 1). The third layer has the same settings as the second layer. The activation function of each convolution layer is set to relu, and the border_mode of the convolution is set to same.

[0067] Step 3: Establish a residual network and use it to process 537 batches of temperature or salinity data that represent the temporal and spatial characteristics of ocean data:

[0068] Step 3-1: Use activation unit (relu) → 3D convolution → activation unit (relu) → 3D convolution to form a basic residual unit. The number of convolution kernels in the 3D convolution is set to 64, the convolution kernel size is (3, 3, 3), and the border parameter border_mode is set to "same". Add the input and output of the current residual unit as the input of the next residual unit.

[0069] Step 3-2: The residual network consists of 4 basic residual structures connected in series.

[0070] The use of the residual network to further obtain the spatial characteristics of ocean temperature or salinity is obtained by the following formula:

[0071] X l+1 =X l+F(X l θ l ) (4)

[0072] Among them, θ l is the set of all learnable parameters in the lth residual unit, F() is the residual function, which represents all operations in the residual unit. In this method, it represents activation + 3D convolution + activation + 3D convolution, X l is the input of the lth residual unit, X l+1 is the output of the lth residual unit.

[0073] Step 3-3: Input the 537 batches of temperature or salinity data representing the temporal and spatial dual characteristics of the ocean data obtained in step 2 into the residual network to obtain 537 batches of temperature or salinity data with prominent spatial characteristics.

[0074] Step 4: The recalibration module includes:

[0075] Step 4-1: Obtain the temporal and spatial features of the ocean data through the previous step 3, and construct tensor data of the same size as the obtained features. The data represents the weight of each feature element. Generate random tensor data of [0,1) first, and then set it as a learnable variable.

[0076] Step 4-2: Multiply the features obtained in the previous step by the corresponding elements of the weight tensor and perform the summation between the features to achieve weighted summation of different features. This is obtained through the following formula:

[0077]

[0078] The input of feature k in channel c is the 537 batches of temperature or salinity data features with prominent spatial characteristics obtained in step 3; for The corresponding weights, is an element-wise multiplication. By learning the parameters To quantify the degree of effect of each feature in the space. Each layer of the neural network has 64 convolution kernels. After four-dimensional convolution and residual network, 64 corresponding features will be generated. Feature k refers to the 64 features generated in the previous steps. The channel c here refers to the channel of the temperature data output by the model, which is the same as the input, and is 1.

[0079] Step 5. In this example, the batch size of the overall spatiotemporal four-dimensional convolution model is set to 32, the learning rate is set to 0.0005, the optimizer uses Adam, and the loss function uses the mean absolute error (MAE). Repeatedly iterate the above steps 2-4, and back propagate to adjust the parameters of the ocean temperature or salinity prediction model, thereby optimizing the ocean temperature or salinity prediction model. The model consists of three parts: four-dimensional convolution, residual, and recalibration. The model parameters are all the learnable parameters in these three parts, such as θ l .

[0080] Step 6: The ocean temperature or salinity prediction is a univariate prediction. The temperature or salinity data of the previous 9 days at the same location is modeled to predict the temperature or salinity on the 10th day. After the above standardized modeling, a spatiotemporal four-dimensional convolution model is obtained. After the measured ocean temperature or salinity data is input into the standard model, the predicted temperature or salinity data of each location is obtained, which realizes the prediction of the overall ocean temperature or salinity, improves the prediction accuracy of the ocean temperature or salinity, and has theoretical and practical significance for the field of ocean temperature and salinity prediction.

[0081] Step 7: Visualize the results of the ocean temperature or salinity prediction in the form of two-dimensional plane or three-dimensional grid by using colors or charts.

[0082] The visualization of the two-dimensional plane is a temperature and salinity contour map of the ocean horizontal surface composed of longitude and latitude at the same depth, or a temperature or salinity contour map of the ocean profile composed of depth and longitude at the same latitude.

[0083] The visualization display in the form of the three-dimensional grid image is a stereoscopic map of the temperature or salinity contour lines of a sea area including three latitudes: longitude, latitude and depth.

[0084] At the same time, an ocean temperature and salinity prediction system of the present invention includes a sensor device, a processor, and a memory; the sensor device includes a GPS positioning device, a depth sensor, a temperature sensor, and an optical fiber sensor (salinometer), which is used to collect three-dimensional tensor temperature or salinity data of longitude, latitude and depth and send it to the processor; a program is stored in the memory, and the processor reads the program to execute the method steps as described above to achieve ocean temperature and salinity prediction.

[0085] like Figure 2 This is a comparison chart of the temperature prediction results and the actual values ​​at 0-10m underwater. Figure 3 This is a comparison chart of the temperature prediction results and the actual values ​​at 1000m-1200m underwater. Figure 4 This is a comparison chart of the predicted results and actual values ​​of summer ocean profile temperature. Figure 5 The following is a comparison chart of the predicted results and true values ​​of winter ocean profile temperature. It can be seen that this method can effectively predict ocean temperature data.

[0086] In summary, the present invention combines a four-dimensional convolution model, a residual structure and a recalibration module, first uses a four-dimensional convolution model to extract dual features of time and space from ocean temperature or salinity data, then uses a residual network to further obtain the spatial features of ocean temperature or salinity, and finally inputs the feature results into the recalibration module, performs weighted summation on each feature, and realizes ocean temperature or salinity prediction. Taking into account the three-dimensional rasterization characteristics of the ocean data itself, the present invention overcomes the limitations of existing temperature or salinity prediction methods, which usually consider the characteristics of time series but do not extract spatial relationships sufficiently, and has theoretical and practical significance in the field of ocean temperature and salinity prediction.

[0087] The embodiments described in the above description will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

Claims

1. A method for predicting ocean temperature and salinity, characterized in that: The steps include: Step 1: preprocessing the original ocean temperature or salinity real-time data at a fixed location within a period of time to meet the data dimension of the four-dimensional convolution model; Step 2: Establish a four-dimensional convolution model, and perform four-dimensional convolution operations on the pre-processed ocean temperature or salinity real-time data in batches to fully extract the spatiotemporal characteristics of the ocean data; Step 2-1, set the number of convolution kernels, convolution kernel size, step size parameters, set the activation function of the convolution layer to the relu function, and establish an M-layer four-dimensional convolutional neural network; Each layer of the four-dimensional convolutional neural network is built on the basis of the three-dimensional convolutional model. The formula is as follows: Among them, b j is the bias term, c is the input channel, j is the output channel, R×X×Y×Z is the size of the convolution kernel, is the weight corresponding to the position (r, x, y, z) in the model, corresponding to input channel c, output channel j, V c is the input of position (s+r,l+x,h+y,w+z) in channel c, is the output of position (s, l, h, w) in channel j; Step 2-2: During the calculation process, the six-dimensional data obtained in step 1 is input into an M-layer convolution neural network; the four-dimensional convolution operation of each layer is completed in sequence, and finally the temperature or salinity data representing the dual characteristics of time and space of the ocean data is obtained; The four-dimensional convolution operation of each layer includes the following steps: Since three dimensions in the six-dimensional data represent the shape of the data, one of the dimensions is expanded to obtain several five-dimensional data, and a three-dimensional convolution calculation is applied to each five-dimensional data. The calculated data are added according to the corresponding positions to complete a four-dimensional convolution operation; the formula for the four-dimensional convolution calculated by the three-dimensional convolution is as follows: Step 3: Establish a residual network composed of multiple basic residual units connected in series to further obtain spatiotemporal data of ocean temperature or salinity with prominent spatial characteristics; Step 4, recalibrate and weight the acquired spatial characteristics of ocean temperature or salinity to predict ocean temperature or salinity; Step 5, repeatedly iterate the above steps 2-4, select the Adam optimizer, use the mean absolute error as the loss function, and back propagate to adjust the parameters of the ocean temperature or salinity prediction model, so as to optimize the ocean temperature or salinity prediction model; Step 6: collect the on-site ocean temperature or salinity of the fixed point to be predicted in real time, pre-process it, and input it into the optimized ocean temperature or salinity prediction model to realize ocean temperature or salinity prediction; Step 7: Visually display the results of the ocean temperature or salinity prediction in the form of a two-dimensional plane or a three-dimensional grid image in the form of colors or charts.

2. The method for predicting ocean temperature and salinity according to claim 1, characterized in that: The original ocean temperature or salinity real-time data is three-dimensional tensor temperature or salinity data of longitude, latitude and depth collected by a GPS positioning device, a depth sensor, a temperature sensor and a salinometer.

3. The method for predicting ocean temperature and salinity according to claim 1, characterized in that: The pre-processing comprises the following steps: Step 1-1, normalization, the mapping range is [-1, 1]; Step 1-2: Process the normalized original ocean temperature or salinity real-time data into a six-dimensional form (U×C×S×L×H×W) required for four-dimensional convolution; Among them, U is the batch size of the data, C is the number of channels, S is the number of samples in the time series, and the data shape is L×H×W, where L is the number of data layers, H is the data height, and W is the data width; The definition and selection principle of the parameters L, H, and W are as follows: in ocean data, the cross-sectional direction is expressed as: L is the depth range, H is the latitude range, and W is the longitude range; the profile direction is expressed as: L, H, and W are the ranges of latitude, depth, and longitude, respectively.

4. The method for predicting ocean temperature and salinity according to claim 1, characterized in that: The residual network is used to further obtain the spatiotemporal data of ocean temperature or salinity with prominent spatial characteristics, including the following steps: Step 3-1: The basic residual unit includes: activation unit→3D convolution→activation unit→3D convolution; Step 3-2: Use the residual network to further obtain the spatial characteristics of ocean temperature or salinity; obtained by the following formula: X l+1 =X l +F(X l ;θ l )(2) Among them, θ l is the set of all learnable parameters in the lth residual unit, X l is the input of the lth residual unit, X l+1 is the output of the lth residual unit.

5. The method for predicting ocean temperature and salinity according to claim 1, characterized in that: The spatial characteristics of the acquired ocean temperature or salinity are recalibrated and weighted summed to obtain the predicted temperature or salinity X c ,include: in, The input of feature k in channel c is the acquired spatiotemporal data of ocean temperature or salinity with prominent spatial features; for The corresponding weights; is an element-wise multiplication; by learning the parameters To quantify the effect of each feature in the space.

6. The method for predicting ocean temperature and salinity according to claim 5, characterized in that: The visualization display of the two-dimensional plane is a temperature or salinity contour map of the ocean horizontal surface composed of longitude and latitude at the same depth, or a temperature and salinity contour map of the ocean profile composed of depth and longitude at the same latitude; The visualization display in the form of the three-dimensional grid image is a stereoscopic map of the temperature or salinity contour lines of a certain sea area including the three dimensions of longitude, latitude and depth.

7. An ocean temperature and salinity prediction system, characterized in that: It comprises a sensor device, a processor, and a memory; the sensor device comprises a GPS positioning device, a depth sensor, a temperature sensor, and a salinometer, and is used to send the collected three-dimensional tensor temperature or salinity data of longitude, latitude and depth to the processor; the memory stores a program, and the processor reads the program to execute the steps of the method described in any one of claims 1 to 6 to realize ocean temperature and salinity prediction.

Images