CN110648030A - Method and device for predicting seawater temperature - Google Patents

Abstract

The invention discloses a seawater temperature prediction method and device, which receive temperature and salinity data in a time series format; the temperature and salinity data are the temperature and salinity data of seawater at any longitude, latitude and depth; In the seawater temperature prediction model, a seawater temperature prediction result is obtained; wherein, the seawater temperature prediction result is a temperature prediction value at the next moment of the last moment of the time series. It can be seen that the present invention uses two variables, seawater temperature and salinity, as the reference quantities for predicting seawater temperature. Compared with the method of using a single sea surface temperature as the reference quantity for predicting seawater temperature, more accurate seawater temperature prediction results can be obtained, and the present invention can obtain more accurate seawater temperature prediction results. The invention is not only applicable to the sea surface, but also applicable to the prediction of seawater temperature at any position of longitude, latitude and depth, and has a wider application range.

Description

Method and device for predicting seawater temperature

technical field

The invention relates to the technical field of temperature prediction, in particular to a seawater temperature prediction method and device.

Background technique

Since the heat content of the ocean is much greater than that of the atmosphere and the land surface, even if there is an extremely slight change in the temperature of the ocean surface, it is possible that the global average temperature will change greatly through the influence of atmospheric circulation. Global temperature has a significant impact. Therefore, accurate prediction of changes in seawater temperature is very important.

The existing seawater temperature prediction methods are based on historical sea surface temperature and long short-term memory (Long short-term memory, LSTM) recurrent neural network to predict seawater temperature. Since the change of seawater temperature is affected by many factors, if the temperature prediction only considers a single variable of sea surface temperature and ignores the temperature change caused by the influence of other factors, the prediction result of seawater temperature will be inaccurate.

SUMMARY OF THE INVENTION

The invention provides a seawater temperature prediction method and device, which can solve the problem of inaccurate seawater temperature prediction results in the prior art because the temperature prediction only considers a single variable of sea surface temperature and ignores temperature changes caused by other factors. question.

To achieve the above object, the invention provides the following technical solutions:

A method for predicting seawater temperature, including:

Receive temperature and salinity data in a time series format; the temperature and salinity data are temperature and salinity data of seawater at any position of longitude, latitude and depth;

Inputting the temperature and salinity data into a seawater temperature prediction model to obtain a seawater temperature prediction result;

Wherein, the seawater temperature prediction result is the temperature prediction value at the next moment of the last moment of the time series.

Optionally, the training process of the seawater temperature prediction model includes:

Convolve the historical temperature and salinity data in time series format to extract the change characteristics between temperature and salinity at all two adjacent moments;

Taking all the changing features as input in chronological order, carry out long short-term memory LSTM cyclic neural network training, and sequentially output the temperature value at the next moment of each group of two adjacent moments.

Optionally, the temperature and salinity data in the time series format is in the form of a two-dimensional array, the temperature and salinity at the same time are arranged in the same column, and the time series of the same variable are arranged in the same row, finally forming the temperature and salinity format in the time series format. data.

Optionally, the convolution operation is performed on the historical temperature and salinity data in the time series format to extract the variation characteristics between temperature and salinity at all two adjacent moments, including:

A convolution operation with a step size of 1 is performed on the historical temperature and salinity data in the time series format through a filter with a resolution of 2*2, and the change characteristics between temperature and salinity between all two adjacent moments are extracted.

Optionally, the sequentially outputting the temperature values at the next moment of each group of two adjacent moments, including:

The output information is controlled by the output gate of the LSTM cyclic unit in the long short-term memory LSTM cyclic neural network architecture, and the output information is directed to the temperature value at the next moment of each group of two adjacent moments.

A device for predicting seawater temperature, comprising:

a receiving unit for receiving temperature and salinity data in a time series format; the temperature and salinity data is the temperature and salinity data of seawater at any position of longitude, latitude and depth;

a prediction unit for inputting the temperature and salt data into a seawater temperature prediction model to obtain a seawater temperature prediction result;

Wherein, the seawater temperature prediction result is the temperature prediction value at the next moment of the last moment of the time series.

Optionally, the prediction device further includes:

The extraction unit is used to perform a convolution operation on the historical temperature and salinity data in the time series format, and extract the variation characteristics between the temperature and salinity between all two adjacent moments;

The training unit is used for taking all the changing features as input in a chronological order, to perform long-short-term memory LSTM cyclic neural network training, and sequentially outputting the temperature values at the next moment of each group of two adjacent moments.

Optionally, the temperature and salinity data in the time series format is in the form of a two-dimensional array, the temperature and salinity at the same time are arranged in the same column, and the time series of the same variable are arranged in the same row, finally forming the temperature and salinity format in the time series format. data.

Optionally, the extraction unit is configured to perform a convolution operation with a step size of 1 on the historical temperature and salinity data in the time series format through a 2*2 resolution filter, and extract the temperature and salinity at all two adjacent moments. characteristics of change between.

Optionally, the training unit is used to control the output information through the output gate of the LSTM cyclic unit in the long short-term memory LSTM cyclic neural network architecture, and direct the output information to the temperature at the next moment of each group of two adjacent moments. value.

As can be seen from the above technical solutions, the present invention discloses a method and device for predicting seawater temperature, which receives temperature and salinity data in a time series format; The temperature and salinity data are input into the seawater temperature prediction model, and the seawater temperature prediction result is obtained; wherein the seawater temperature prediction result is the temperature prediction value at the next moment of the last moment of the time series. It can be seen that the present invention uses two variables of seawater temperature and salinity as reference quantities for predicting seawater temperature. Compared with the method of using a single sea surface temperature as the reference quantity for predicting seawater temperature, more accurate seawater temperature prediction results can be obtained, and the present invention can obtain more accurate seawater temperature prediction results. The invention is not only applicable to the sea surface, but also applicable to the prediction of seawater temperature at any position of longitude, latitude and depth, and has a wider application range.

Description of drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only It is an embodiment of the present invention. For those of ordinary skill in the art, other drawings can also be obtained according to the provided drawings without creative work.

1 is a flowchart of a method for predicting seawater temperature disclosed in an embodiment of the present invention;

2 is a schematic diagram of a feature extraction process of temperature-salt grid data disclosed in an embodiment of the present invention;

3 is a schematic diagram of the structure of an LSTM cell unit disclosed in an embodiment of the present invention;

Fig. 4 is the LSTM cyclic neural network architecture diagram of the temperature and salt data convolution disclosed in the embodiment of the present invention;

FIG. 5 is a schematic diagram of an apparatus for predicting seawater temperature disclosed in an embodiment of the present invention.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

As known from the background art, the existing seawater temperature prediction method predicts the seawater temperature according to the historical sea surface temperature and a long short-term memory (Long short-term memory, LSTM) recurrent neural network. Since the change of seawater temperature is affected by many factors, if the temperature prediction only considers a single variable of sea surface temperature and ignores the temperature change caused by the influence of other factors, the prediction result of seawater temperature will be inaccurate.

In view of this, the present invention provides a method and device for predicting seawater temperature, which can solve the problem that in the prior art, since temperature prediction only considers a single variable of sea surface temperature and ignores the temperature change caused by the influence of other factors, resulting in the prediction result of seawater temperature. Inaccurate technical issues.

As shown in FIG. 1 , an embodiment of the present invention discloses a method for predicting seawater temperature, including the following steps:

S101. Receive temperature and salinity data in a time series format.

In step S101, the temperature and salinity data are temperature and salinity data of seawater at any position of longitude, latitude and depth.

S102. Input the temperature and salinity data into a seawater temperature prediction model to obtain a seawater temperature prediction result.

In step S102, the seawater temperature prediction result is the temperature prediction value at the next moment after the last moment of the time series.

Optionally, the training process of the seawater temperature prediction model includes:

Convolve the historical temperature and salinity data in time series format to extract the change characteristics between temperature and salinity at all two adjacent moments;

Taking all the changing features as input in chronological order, carry out long short-term memory LSTM cyclic neural network training, and sequentially output the temperature value at the next moment of each group of two adjacent moments.

Optionally, the temperature and salinity data in the time series format is in the form of a two-dimensional array, the temperature and salinity at the same time are arranged in the same column, and the time series of the same variable are arranged in the same row, finally forming the temperature and salinity format in the time series format. data.

Optionally, the convolution operation is performed on the historical temperature and salinity data in the time series format to extract the variation characteristics between temperature and salinity at all two adjacent moments, including:

A convolution operation with a step size of 1 is performed on the historical temperature and salinity data in the time series format through a filter with a resolution of 2*2, and the change characteristics between temperature and salinity between all two adjacent moments are extracted.

Optionally, the sequentially outputting the temperature values at the next moment of each group of two adjacent moments, including:

The output information is controlled by the output gate of the LSTM cyclic unit in the long short-term memory LSTM cyclic neural network architecture, and the output information is directed to the temperature value at the next moment of each group of two adjacent moments.

For the convenience of understanding, the prediction method of this embodiment is described in detail:

It should be noted that convolutional neural networks are widely used to capture spatial relationships, such as extracting features between pixels in an image. Unlike fully-connected layers, convolutional units do not consider the entire input vector, but operate on the input vector sequentially using the same weights and fixed-size windows or convolutions. Using convolutions with different weights, different features can be extracted.

In the prediction method disclosed in the embodiment of the present invention, the original one-dimensional time series temperature and salinity data are converted into two-dimensional temperature and salinity time series grid data. Through such transformation, we can use the convolution operation to extract the adjacent temperature and salinity changes on the time axis. As shown in Figure 2, it shows the extraction process of temperature-salt variation features from two-dimensional temperature-salt grid data through convolution kernel. Among them, taking time i and i+1 as an example, the convolution operation can obtain the feature maps of temperature and salinity changes at these two times. Using convolutions with different weights, different feature maps can be obtained. Various feature maps will provide more input information for the operations of subsequent feature recurrent layers.

It should be further stated that the LSTM recurrent neural network is a powerful neural network structure, which can make a targeted approximation of the sequence and learn meaningful features. LSTM recurrent neural network is a widely used recurrent neural network. As shown in Figure 3, in the LSTM cell unit, the input gate controls whether the input value can be accumulated into the state (i.e., the memory cell) to finally get the new value. The state unit can loop itself linearly, and the forget gate controls its weights. The output of the cell is controlled by the output gate, which can be controlled to close.

Specifically, the entire calculation of the LSTM network can be defined by the following series of equations:

i _t =σ(W ⁱ H+ ^bi )

f _t =σ(W ^f H+b ^f )

O _t =σ(W ^o H+ ^bo )

c _t =tanh(W ^c H+b ^c )

m _t =f _t ⊙m _t-1 +i _t ⊙c _t

h _t =tanh(o _t ⊙m _t )

Among them, i _t , f _t , o _t and c _t are the input gate, forgetting gate, output gate and the new state of the memory cell, respectively. σ is a sigmoid function, and W ⁱ , W ^f , W ^o and W ^c are weight matrices. b _i , b _f , b _c and b _c are the corresponding bias terms. ⊙ represents the product of the corresponding elements of the vector. m _t is the final state of the memory cell, h _t is the final output of the memory cell. H is the concatenation of the new input x _t and the hidden vector h _t-1 of the previous moment.

In the embodiment of the present invention, the change characteristics of temperature and salinity extracted by the convolution layer are used as the input of the LSTM cycle unit, and the output of the cycle unit is the predicted value of the temperature at the next moment.

Finally, it should be noted that with regard to the LSTM cyclic neural network architecture convolved with temperature and salt data, the neural network architecture for seawater temperature prediction based on the LSTM cyclic neural network based on temperature and salt data disclosed in the embodiment of the present invention is mainly divided into four parts, such as Figure 4 shows the input layer, convolutional layer, recurrent layer and output layer, respectively. In the input layer, we convert the two time series of temperature and salinity into the form of a two-dimensional array, where the temperature and salinity at the same time are arranged in the same column, and the time series of the same variable are arranged in the same row. In the convolution layer, a stride 1 convolution operation is performed on the two-dimensional data using a 2*2 filter. Through convolution, the relationship between temperature and salinity at two adjacent moments can be extracted. Using k filters for convolution, k variable features can be extracted. In the recurrent layer, each LSTM recurrent unit accepts the k features passed by the convolution of the previous layer as the input of the current moment. The loop unit transmits state information to the next moment and controls the output information through the output gate. We point the output information at each moment to the temperature value at the next moment. When the circulation unit receives the characteristic values of the temperature-salt relationship at time t-1 and t, the output information is the prediction of the temperature value at time t+1. We define the loss function as the mean square error (MSE) between these output information and the real value at the next moment, and through backpropagation to minimize the loss value to achieve the effect of model optimization.

Using the trained model, a new set of temperature and salinity time series is used as input, and the output of the model is the predicted value of the temperature at the next moment at the last moment of the time series.

The seawater temperature prediction method disclosed in this embodiment receives temperature and salinity data in a time series format; the temperature and salinity data is the temperature and salinity data of seawater at any position of longitude, latitude and depth; and the temperature and salinity data are input into the seawater temperature prediction In the model, a seawater temperature prediction result is obtained; wherein, the seawater temperature prediction result is a temperature prediction value at the next moment of the last moment of the time series. It can be seen that the present invention uses two variables of seawater temperature and salinity as reference quantities for predicting seawater temperature. Compared with the method of using a single sea surface temperature as the reference quantity for predicting seawater temperature, more accurate seawater temperature prediction results can be obtained, and the present invention can obtain more accurate seawater temperature prediction results. The invention is not only applicable to the sea surface, but also applicable to the prediction of seawater temperature at any position of longitude, latitude and depth, and has a wider application range.

Based on the seawater temperature prediction method disclosed in the above embodiments of the present invention, FIG. 5 specifically discloses a seawater temperature prediction device applying the seawater temperature prediction method.

As shown in FIG. 5 , another embodiment of the present invention discloses a device for predicting seawater temperature, and the device includes:

A receiving unit 501, configured to receive temperature and salinity data in a time series format; the temperature and salinity data is temperature and salinity data of seawater at any position of longitude, latitude and depth;

A prediction unit 502, configured to input the temperature and salinity data into a seawater temperature prediction model to obtain a seawater temperature prediction result;

Wherein, the seawater temperature prediction result is the temperature prediction value at the next moment of the last moment of the time series.

Optionally, the prediction device further includes:

The extraction unit is used to perform a convolution operation on the historical temperature and salinity data in the time series format, and extract the variation characteristics between the temperature and salinity between all two adjacent moments;

The training unit is used for taking all the changing features as input in a chronological order, to perform long-short-term memory LSTM cyclic neural network training, and sequentially outputting the temperature values at the next moment of each group of two adjacent moments.

Optionally, the temperature and salinity data in the time series format is in the form of a two-dimensional array, the temperature and salinity at the same time are arranged in the same column, and the time series of the same variable are arranged in the same row, finally forming the temperature and salinity format in the time series format. data.

Optionally, the extraction unit is configured to perform a convolution operation with a step size of 1 on the historical temperature and salinity data in the time series format through a 2*2 resolution filter, and extract the temperature and salinity at all two adjacent moments. characteristics of change between.

Optionally, the training unit is used to control the output information through the output gate of the LSTM cyclic unit in the long short-term memory LSTM cyclic neural network architecture, and direct the output information to the temperature at the next moment of each group of two adjacent moments. value.

For the specific working process of the receiving unit 501 and the predicting unit 502 in the seawater temperature prediction device disclosed in the above embodiments of the present invention, reference may be made to the corresponding content in the seawater temperature prediction method disclosed in the above embodiments of the present invention, which will not be repeated here. .

The seawater temperature prediction device disclosed in this embodiment receives temperature and salinity data in a time series format; the temperature and salinity data is the temperature and salinity data of seawater at any position of longitude, latitude and depth; and the temperature and salinity data are input into the seawater temperature prediction In the model, a seawater temperature prediction result is obtained; wherein, the seawater temperature prediction result is a temperature prediction value at the next moment of the last moment of the time series. It can be seen that the present invention uses two variables of seawater temperature and salinity as reference quantities for predicting seawater temperature. Compared with the method of using a single sea surface temperature as the reference quantity for predicting seawater temperature, more accurate seawater temperature prediction results can be obtained, and the present invention can obtain more accurate seawater temperature prediction results. The invention is not only applicable to the sea surface, but also applicable to the prediction of seawater temperature at any position of longitude, latitude and depth, and has a wider application range.

It should also be noted that the terms "comprising", "comprising" or any other variation thereof are intended to encompass a non-exclusive inclusion such that a process, method, article or device comprising a series of elements includes not only those elements, but also Other elements not expressly listed, or which are inherent to such a process, method, article of manufacture, or apparatus are also included. Without further limitation, an element qualified by the phrase "comprising a..." does not preclude the presence of additional identical elements in the process, method, article of manufacture or apparatus that includes the element.

It will be appreciated by those skilled in the art that the embodiments of the present application may be provided as a method, a system or a computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The above are merely examples of the present application, and are not intended to limit the present application. Various modifications and variations of this application are possible for those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included within the scope of the claims of this application.

Claims (10)

1. A method for predicting seawater temperature, comprising:

receiving thermohaline data in a time series format; the temperature and salinity data is the temperature and salinity data of the seawater at any longitude and latitude depth position;

inputting the temperature and salt data into a seawater temperature prediction model to obtain a seawater temperature prediction result;

and the seawater temperature prediction result is a temperature prediction value at the next moment of the last moment of the time series.

2. The prediction method of claim 1, wherein the training process of the seawater temperature prediction model comprises:

performing convolution operation on the historical temperature and salinity data in the time series format, and extracting the change characteristics between the temperature and the salinity at all two adjacent moments;

and taking all the change characteristics as input according to a time sequence, carrying out long-short term memory (LSTM) recurrent neural network training, and sequentially outputting the temperature value of the next moment of each group of adjacent two moments.

3. The prediction method according to claim 2, wherein the warm salt data in time series format is in two-dimensional array form, the temperature and salinity at the same time are arranged in the same column, the time series of the same variable is arranged in the same row, and finally the warm salt data in time series format is formed.

4. The prediction method according to claim 3, wherein the convolution operation is performed on the historical temperature and salinity data in a time series format to extract the change characteristics between the temperature and the salinity at all two adjacent time instants, and comprises the following steps:

and (3) performing convolution operation with the step size of 1 on the historical temperature and salinity data in the time series format through a filter with the resolution of 2 x 2, and extracting the change characteristics between the temperature and the salinity at all the two adjacent moments.

5. The prediction method according to claim 2, wherein the sequentially outputting the temperature value at the next time of each set of two adjacent times comprises:

and controlling output information through an output gate of an LSTM circulation unit in the long-short term memory LSTM circulation neural network architecture, and pointing the output information to a temperature value at the next moment of each group of two adjacent moments.

6. A prediction device of seawater temperature, comprising:

the receiving unit is used for receiving the temperature and salinity data in a time series format; the temperature and salinity data is the temperature and salinity data of the seawater at any longitude and latitude depth position;

the prediction unit is used for inputting the temperature and salinity data into a seawater temperature prediction model to obtain a seawater temperature prediction result;

and the seawater temperature prediction result is a temperature prediction value at the next moment of the last moment of the time series.

7. The prediction apparatus according to claim 6, further comprising:

the extraction unit is used for performing convolution operation on the historical temperature and salinity data in the time series format and extracting the change characteristics between the temperature and the salinity at two adjacent moments;

and the training unit is used for performing long-short term memory (LSTM) cyclic neural network training by taking all the change characteristics as input according to a time sequence, and sequentially outputting the temperature value of the next moment of each group of two adjacent moments.

8. The prediction device of claim 7, wherein the time series format of the warm salt data is in a two-dimensional array form, the temperature and the salinity at the same time are arranged in the same column, the time series of the same variable is arranged in the same row, and finally the time series format of the warm salt data is formed.

9. The prediction device according to claim 8, wherein the extraction unit is configured to extract the variation between temperature and salinity at two adjacent time instants by performing a convolution operation with a step size of 1 on the historical temperature and salinity data in a time series format through a 2 x 2 resolution filter.

10. The prediction apparatus as claimed in claim 7, wherein the training unit is configured to control the output information through an output gate of an LSTM cyclic neural network architecture, and point the output information to a temperature value at a next time of each set of two adjacent times.

Images