Seawater sound velocity profile prediction method and device based on BNNConvLSTM

Abstract

The present invention provides a method and device for predicting seawater sound velocity profile based on BNNConvLSTM. The method comprises obtaining raw data within a preset time period of a sea area to be measured, and calculating the seawater sound velocity profile data of the sea area to be measured based on the raw data; applying a preset ConvLSTM model based on the seawater sound velocity profile data to obtain sound velocity profile prediction data of the sea area to be measured; and applying a preset BNNConvLSTM model based on the seawater sound velocity profile data to obtain a prediction interval of the sound velocity profile prediction data of the sea area to be measured within a specified confidence interval. The present invention makes full use of the spatiotemporal characteristics of the seawater sound velocity profile, designs a ConvLSTM model and a BNNConvLSTM model to respectively calculate the predicted value of the seawater sound velocity to be measured and the prediction interval within a specified confidence interval, thereby achieving a measurement of the uncertainty in the prediction caused by the variability of the marine environment.

Description

Technical Field

[0001] The present invention relates to the technical field of seawater sound velocity profile prediction, and in particular to a seawater sound velocity profile prediction method and device based on BNNConvLSTM. Background Art

[0002] The sound velocity profile of seawater reflects how the sound velocity at a specific location changes with depth, providing a visual indication of the changes in sound velocity at that location. This profile is primarily obtained through direct and indirect measurement. Direct measurement uses the propagation time or phase of a sound wave at a fixed depth to directly determine the propagation velocity of the sound wave. However, due to the vastness of the ocean and the variability of the marine environment, direct measurement consumes significant manpower and material resources, increasing measurement costs. Indirect measurement utilizes the relationship between temperature, salinity, depth, and seawater sound velocity. The temperature, salinity, and depth measured by hydrographic instruments are then calculated using empirical formulas for sound velocity to create the sound velocity profile. With the advancement of satellite remote sensing technology, seawater temperature, salinity, and depth data have become increasingly accessible. However, both direct and indirect measurements only measure the current sound velocity value and cannot predict future values.

[0003] In the current era of building a strong maritime nation, predicting ocean sound velocity profiles is crucial for exploring the marine environment, developing marine resources, developing the marine economy, strengthening maritime capabilities, and safeguarding national maritime rights and interests. For example, sound propagation in the ocean is influenced by the sound velocity profile, which in turn impacts sonar performance evaluation and prediction. Given that neither direct nor indirect sound velocity measurements can accurately predict future seawater sound velocity profiles, and thus accurate sonar performance predictions, a robust sound velocity profile prediction method is urgently needed. Furthermore, due to the variability of the ocean environment, in addition to achieving accurate predictions of ocean sound velocity, measuring the uncertainty in the predictions is also crucial. Therefore, it is necessary to design a seawater sound velocity profile prediction method based on the spatiotemporal characteristics of seawater sound velocity that offers high accuracy and can account for the uncertainty caused by this variability. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a seawater sound velocity profile prediction method and device based on BNNConvLSTM with high prediction accuracy and the ability to measure prediction uncertainty, so as to describe the interval uncertainty of the prediction point.

[0005] To achieve the above objectives, according to one aspect of the present invention, a method for predicting seawater sound velocity profile based on BNNConvLSTM is provided, comprising:

[0006] Acquire raw data within a preset time period of the sea area to be measured, and calculate seawater sound velocity profile data of the sea area to be measured based on the raw data;

[0007] Applying a preset ConvLSTM model according to the seawater sound velocity profile data to obtain sound velocity profile prediction data of the sea area to be measured;

[0008] According to the seawater sound velocity profile data, the preset BNNConvLSTM model is applied to obtain the prediction interval of the sound velocity profile prediction data of the sea area to be tested under the specified confidence interval. When applying the BNNConvLSTM model, the number of repeated experiments of the BNNConvLSTM model is set to obtain the mean y and standard deviation σ of the output data of multiple tests, and the prediction interval of the sound velocity profile prediction data of the sea area to be tested under the specified confidence interval is calculated, that is, Where n is the number of samples.

[0009] In one embodiment, the obtaining of raw data within a preset time period of the sea area to be measured and calculating the seawater sound velocity profile data of the sea area to be measured based on the raw data includes: obtaining a raw data set of seawater within a preset time period, the raw data set including temperature data, salinity data and depth data; performing high-resolution data processing of the raw data set in the depth dimension, and applying a preset sound velocity empirical formula to calculate the seawater sound velocity profile data of the sea area to be measured, the sound velocity empirical formula being:

[0010] C=1449.2+4.6T-0.055T 2 +0.00029T 3 +(1.34-0.01T)(S-35)+0.016Z

[0011] Where C is the speed of sound in m / s, T is the temperature in °C, S is the salinity in ‰, and Z is the depth in m.

[0012] In one embodiment, before applying a preset ConvLSTM model according to the seawater sound speed profile data to obtain the sound speed profile prediction data of the sea area to be measured, it includes: training the ConvLSTM model and the BNNConvLSTM model, including: obtaining the original data of seawater in a preset historical period, and calculating the seawater sound speed profile data in the preset historical period, and constructing an input data set; dividing the seawater sound speed profile data into a training set and a test set, and the training set and the test set respectively include a number of sound speed-depth samples; applying the training set and the test set to train the ConvLSTM model to obtain the trained ConvLSTM model; constructing the BNNConvLSTM model based on the trained ConvLSTM model and training it.

[0013] In one embodiment, the constructing of the input data set includes: normalizing the seawater sound speed profile data to eliminate dimensionality effects; constructing the seawater sound speed profile data into a five-dimensional tensor of (samples, time, rows, cols, channels) to form an input data set, wherein samples represents the number of sound speed-depth samples of the input seawater sound speed profile data, time represents the prediction time step, rows is set to represent the depth features in the sound speed-depth samples, cols is set to represent the sound speed data in the sound speed-depth samples, cols is set to 1 column, and channels is set to 1.

[0014] In one embodiment, the ConvLSTM model includes two ConvLSTM layers, and each ConvLSTM layer is respectively connected to a Batch Normalization layer and a Dropout layer, wherein the number of convolution kernels of the two ConvLSTM layers is 50, the convolution kernel size of the first ConvLSTM layer is 6×6, the convolution kernel size of the second ConvLSTM layer is 4×4, the attenuation rate of the Batch Normalization layer is set to 0.99, and the probability of the Dropout layer is set to 0.2; the optimization algorithm of the ConvLSTM model is set to Adam, and the activation function of the ConvLSTM model is set to Tanh function.

[0015] In one embodiment, the applying the training set and the test set to train the ConvLSTM model to obtain the trained ConvLSTM model includes: training the ConvLSTM model according to the training set and the test set, wherein the batch size is set to 32, the epoch maximum value is set to 800, and early stopping is set, and the training is stopped when the loss no longer decreases for 10 consecutive times or when the maximum value set by the epoch is reached, thereby obtaining the trained ConvLSTM model.

[0016] In one embodiment, the BNNConvLSTM model is constructed and trained based on the trained ConvLSTM model, including: based on the trained ConvLSTM model, setting the Dropout layer to be effective during testing to construct a BNNConvLSTM model; applying the training set and the test set to train the BNNConvLSTM model to obtain the trained BNNConvLSTM model.

[0017] Based on the same inventive concept, the present invention also provides a seawater sound velocity profile prediction device based on BNNConvLSTM, comprising:

[0018] The raw data processing module is used to obtain the raw data of the sea area to be measured within a preset time period, and calculate the seawater sound velocity profile data of the sea area to be measured based on the raw data;

[0019] A sound velocity profile prediction value module is used to apply a preset ConvLSTM model to obtain the sound velocity profile prediction data of the sea area to be measured according to the seawater sound velocity profile data;

[0020] The sound speed profile prediction interval module is used to apply a preset BNNConvLSTM model according to the seawater sound speed profile data to obtain the prediction interval of the sound speed profile prediction data of the sea area to be measured under a specified confidence interval.

[0021] Based on the same inventive concept, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and runnable on the processor. When the processor executes the computer program, it implements any of the above-mentioned methods for predicting seawater sound speed profiles based on BNNConvLSTM.

[0022] Based on the same inventive concept, the present invention also provides a computer-readable storage medium having computer instructions stored thereon, wherein the computer instructions are used to enable a computer to execute the seawater sound velocity profile prediction method based on BNNConvLSTM as described in any of the above.

[0023] The seawater sound velocity profile prediction method and device based on BNNConvLSTM provided by the present invention can respectively obtain the predicted value of the seawater sound velocity to be measured and the prediction interval under the specified confidence interval according to the trained ConvLSTM model and the trained BNNConvLSTM model. On the basis of not changing the original network model, the ConvLSTM model is optimized by using a random deactivation method to better measure the uncertainty in the result, fully utilizing the spatiotemporal characteristics of the sound velocity profile, giving the predicted interval while obtaining a high-precision definite prediction result, and measuring the uncertainty in the prediction caused by the variability of the marine environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0025] Figure 1 This is a flow chart of an embodiment of a method for predicting seawater sound velocity profile based on BNNConvLSTM of the present invention;

[0026] Figure 2 Schematic diagram of the network structure of the ConvLSTM model of the present invention;

[0027] Figure 3 The graph shows the true and predicted values of the speed of sound at (165.5°E, 0.5°N), where the black dotted line represents the true value and the black solid line represents the predicted value;

[0028] Figure 4 The graph shows the true and predicted values of the sound speed at (170.5°E, 6.5°N), where the black dotted line represents the true value and the black solid line represents the predicted value;

[0029] Figure 5 The graph shows the true and predicted values of the speed of sound at (175.5°E, 0.5°N), where the black dotted line represents the true value and the black solid line represents the predicted value;

[0030] Figure 6 The figure shows the true value and predicted interval of the sound speed at (165.5°E, 0.5°N), where the black dashed line represents the true value and the gray shaded area represents the predicted interval.

[0031] Figure 7 The figure shows the true value and predicted interval of the sound speed at (170.5°E, 6.5°N), where the black dotted line represents the true value and the gray shaded area represents the predicted interval.

[0032] Figure 8 The figure shows the true value and predicted interval of the sound speed at (175.5°E, 0.5°N), where the black dotted line represents the true value and the gray shaded area represents the predicted interval.

[0033] Figure 9 This is a structural schematic diagram of an embodiment of a seawater sound velocity profile prediction device based on BNNConvLSTM of the present invention;

[0034] Figure 10 FIG. 1 is a schematic structural diagram of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0036] like Figure 1As shown, the embodiment of the present invention provides a seawater sound velocity profile prediction method based on BNNConvLSTM, which specifically includes the following steps:

[0037] S1: Obtain the original data of the sea area to be measured within a preset time period, and calculate the seawater sound velocity profile data of the sea area to be measured based on the original data.

[0038] S2: Apply the preset ConvLSTM model based on the seawater sound speed profile data to obtain the sound speed profile prediction data of the sea area to be measured.

[0039] S3: Apply the preset BNNConvLSTM model based on the seawater sound speed profile data to obtain the prediction interval of the sound speed profile prediction data of the measured sea area under the specified confidence interval.

[0040] In the present embodiment, the specific method for calculating the corresponding seawater sound velocity profile data based on the original data of seawater is as follows: first, the original data set of seawater in a preset time period is obtained, and the original data set includes temperature data, salinity data and depth data. Specifically, the original data set of seawater in a preset historical period is obtained from public channels. In the present embodiment, the original data set of seawater is a data set of yearly and monthly types obtained from the China Argo Real-time Data Center (http: / / www.argo.org.cn / ). The time span of the original data set obtained is during the period of 2005-2020, and the data format is NETCDF format (Network Common Data Format). Then, the data set in NETCDF format is subjected to high-resolution data processing in the depth dimension, and the data sample points originally distributed at unequal intervals in the depth are processed so that the originally sparse temperature and salinity data in the data set are high-resolution in the depth dimension.

[0041] Substitute the depth-temperature-salinity data in the data set that has been processed for high resolution in the depth dimension into the empirical formula for sound velocity one by one to obtain the corresponding sound velocity values at different depths, and then obtain the seawater sound velocity profile data. The preset empirical formula for sound velocity is as follows:

[0042] C=1449.2+4.6T-0.055T 2 +0.00029T 3 +(1.34-0.01T)(S-35)+0.016Z

[0043] Where C is the speed of sound in m / s, T is the temperature in °C, S is the salinity in ‰, and Z is the depth in m.

[0044] In this embodiment, before applying the model for prediction based on the seawater sound velocity profile data, the ConvLSTM model and the BNNConvLSTM model need to be trained first, including: obtaining the original data of seawater in a preset historical period, and calculating the seawater sound velocity profile data in the preset historical period to construct an input data set; dividing the seawater sound velocity profile data into a training set and a test set, and the training set and the test set respectively include a number of sound speed-depth samples; using the training set and the test set to train the ConvLSTM model to obtain a trained ConvLSTM model; constructing a BNNConvLSTM model based on the trained ConvLSTM model and training it.

[0045] Specifically, first set the input data set, network structure and related parameters of the ConvLSTM model, and then train the ConvLSTM model. Compared with the traditional LSTM model that obtains the time characteristics of data, ConvLSTM can obtain more comprehensive data features, including time and space characteristics. Figure 2 As shown in the figure, ConvLSTM is similar to LSTM in that it uses the output of the previous layer as the input of the next layer. The difference is that ConvLSTM adds a convolution operation on the basis of LSTM, replacing the original point multiplication operation with a convolution operation, and simultaneously obtains temporal relationships and spatial features.

[0046] f t =σ(W f *[h t-1 ,x t ]+b f )

[0047] i t =σ(W i *[h t-1 ,x t ]+b i )

[0048]

[0049]

[0050] o t =σ(W o *[h t-1 ,x t ]+b o )

[0051] h t =o t ×tanh(c t )

[0052] Among them, * represents the convolution operation, f trepresents the output forgetting matrix, σ represents the σ activation function, W f represents the weight of the forget gate, h t-1 Indicates the state output of the previous moment, x t represents the input at the current moment, b f Represents the bias of the forget gate, i t Indicates the updated value, W i Indicates the input gate i t The weight of b i Indicates the input gate i t Partial bias, represents the candidate state value, tanh(·) represents the activation function, W c Indicates the input gate The weight of b c Indicates the input gate i t Partial bias, c t Indicates the current state value, c t-1 Indicates the state at the previous moment, ht indicates the state output at the current moment, o t represents the output matrix, W o Indicates the output gate o t The weight of b o Indicates the output gate o t Bias.

[0053] In this embodiment, the parameters and network structure of the ConvLSTM model are set through multiple experiments. The calculated seawater sound speed profile data within a preset historical period is first normalized to eliminate the dimensional influence, and then divided into a training set and a test set according to the proportion. The sound speed-depth sample data in the training set is used as the input data for training, and the sound speed-depth sample data in the test set is used to test the seawater sound speed profile prediction data of the sea area to be tested.

[0054] In the construction of the input data set, according to the data input format requirements of ConvLSTM, the input data set of the ConvLSTM model is constructed as a five-dimensional tensor of (samples, time, rows, cols, channels), where samples represents the number of input sound speed-depth samples (the number of samples); time represents the prediction time step; combined with the depth characteristics of the seawater sound speed profile data, a two-dimensional matrix containing depth information is formed, rows is set to represent the depth characteristics in the sound speed-depth sample, the specific value is determined according to the depth information, such as 0-500m, the depth interval is 1m, rows is set to 501 rows, cols is set to represent the sound speed data in the sound speed-depth sample, cols is set to 1 column, and channels is set to 1. In this embodiment, the time span characteristics and depth characteristics of the sound speed data are combined with the rows and cols data features of ConvLSTM to form a two-dimensional matrix with 501 rows and 1 column, where rows represents the depth characteristics from 0-500m, cols represents the sound speed data, which is 1 column, and channels is set to 1. To determine the specific time step, experiments are conducted at different time steps. In this embodiment, RMSE (Root Mean Square Error) is selected as the evaluation indicator, and experiments are conducted on the selected step range to determine the specific time step. The time span of the original data set used in this embodiment is from January 2005 to December 2020, a total of 192 months, and the minimum unit is month. The final time step is determined to be 7 months.

[0055] In this embodiment, the number of ConvLSTM layers is set to two layers, and the Batch Normalization layer and the Dropout layer are connected after each ConvLSTM layer, and the attenuation rate of the Batch Normalization layer and the probability of the Dropout layer are set. Regarding the selection of experimental parameters, the Adam optimizer is selected as the optimization algorithm and the Tanh function is selected as the activation function. On the basis of setting the ConvLSTM layer to two layers, the number of convolution kernels of different layers is set, and the appropriate number and size of convolution kernels are finally obtained through experiments. After multi-layer processing, the output layer performs denormalization on the data, and finally obtains the output sound speed prediction value. The output of the ConvLSTM model is output in the form of seawater depth profile data. The structure and required parameters of the ConvLSTM model are shown in Table 1.

[0056] Table 1 ConvLSTM structure

[0057]

[0058] In this embodiment, the number of ConvLSTM layers was selected from 1, 2, 3, or 4. Experimental comparison ultimately led to the choice of two ConvLSTM layers. A Batch Normalization layer and a Dropout layer were connected after each ConvLSTM layer. The decay rate of the Batch Normalization layer was set to 0.99, and the probability of the Dropout layer was set to 0.2. In the ConvLSTM model, the Dropout layer was active during training but not during testing. This effectively mitigates overfitting during training. Regarding experimental parameter selection, Adam was chosen as the optimization algorithm, and the Tanh function as the activation function. The number of convolution kernels was selected from 10, 30, or 50. Based on the two-layer ConvLSTM layer, the number of convolution kernels in each layer was adjusted, ultimately achieving a final value of 50 kernels per layer. The convolution kernel size was selected from 2×2, 4×4, or 6×6. After setting the ConvLSTM layer to two and the number of convolution kernels to 50, the kernel size of each layer was adjusted to obtain the desired kernel size. In this embodiment, the convolution kernel size of the first layer ConvLSTM is 6×6, and the convolution kernel size of the second layer ConvLSTM is 4×4.

[0059] In this embodiment, in the division of training set and test set, the data from 2005 to 2017 is used as the training set, and the data from the last three years of 2018 to 2020 is used as the test set. The batch size is set to 32, the epoch maximum value is 800, and early stopping is set. When the loss no longer decreases for 10 consecutive times or reaches the highest value set by epoch, the training is stopped. By training the ConvLSTM model, the optimal network parameters and network structure of the convolutional long short-term memory neural network (ConvLSTM) based on the sound speed value are determined, and the target data set is input to obtain the predicted value of the ConvLSTM for the sound speed under the optimal network structure and parameters. The predicted value is a definite value.

[0060] The trained ConvLSTM model is used to calculate the predicted data of the seawater sound speed profile of the sea area to be measured. The output of the ConvLSTM model is in the form of seawater sound speed profile data.

[0061] In this embodiment, based on the trained ConvLSTM model, the Dropout layer is set to be active during testing, that is, a BNNConvLSTM model is constructed. The specific method is to rerun the code according to the principle of the MC Dropout algorithm, so that the Dropout layer is activated during the test phase. At the same time, based on the trained BNNConvLSTM model, the number of repeated experiments of the BNNConvLSTM model is set to obtain the mean y and standard deviation σ of the output data of multiple tests, and calculate the prediction interval under the specified confidence interval, that is, Where n is the number of samples. For example, if we set the number of experiments to 200, we will obtain 200 results. Based on these 200 results, we can obtain the mean and standard deviation of the 200 results. In the BNNConvLSTM model, the Dropout layer is used during both training and testing.

[0062] In this embodiment, based on the trained BNNConvLSTM model, the number of repeated experiments is set to 200, the mean and standard deviation of the sound speed are obtained, and the upper and lower bounds of the sound speed under the 95% confidence interval are obtained by calculation.

[0063] The trained BNNConvLSTM model is used to calculate the prediction interval of the seawater sound velocity profile prediction data of the measured sea area under the specified confidence interval. The output of the BNNConvLSTM model is displayed in the form of a graph, and the prediction interval is shown in the graph as follows Figure 6-8 The shaded area shown.

[0064] In this embodiment, the interval is evaluated mainly by the two indicators of reliability and clarity. Specifically, the interval coverage and the average interval width are used to measure the reliability and clarity respectively.

[0065] PT Coverage Probability (PICP) represents the percentage of true values falling within the prediction interval and is used to measure the reliability of the prediction interval. Its calculation expression is:

[0066]

[0067] n is the number of samples used for testing:

[0068]

[0069] y i is the true sound speed value of the i-th, lower i and upper i are the lower and upper bounds of the prediction interval, respectively.

[0070] Mean PI Width (MPIW) represents the difference between the upper and lower bounds of the prediction interval. From a mathematical point of view, it is the average value of the interval width, which is a measure of the width of the prediction interval. It is calculated as:

[0071]

[0072] Based on the high-resolution temperature and salinity data, the corresponding sound velocity values were directly obtained using the empirical sound velocity formula. The selected locations were (165.5°E, 0.5°N), (170.5°E, 6.5°N), and (175.5°E, 0.5°N). The time period was from January 2005 to December 2020, a total of 192 months. The structure and parameters of the resulting ConvLSTM model are shown in Table 2.

[0073] Table 2 ConvLSTM neural network parameters

[0074]

[0075] Finally, the true values of the sound speed in the 0-400m depth range at (165.5°E, 0.5°N), (170.5°E, 6.5°N), and (175.5°E, 0.5°N) from January to June 2020 are displayed together with the determined predicted values (i.e., the predicted values obtained by the ConvLSTM model). Figures 3 to 5 As shown in the figure, the black dotted line represents the true value, and the black solid line represents the predicted value.

[0076] After obtaining the optimal ConvLSTM neural network model, the upper and lower bounds of the sound speed range prediction under the 95% confidence interval are obtained according to the set number of repeated experiments. Figures 6 to 8 As shown, the black dashed line represents the true value and the gray shaded area represents the prediction interval.

[0077] Table 3 Evaluation index values of prediction intervals under two methods

[0078]

[0079] The predicted intervals are measured using two metrics: interval coverage and average interval width. Table 3 shows that the interval coverage at the three selected locations exceeds 90%, and the average interval width is within 10 meters. This indicates that the predicted intervals at a 95% confidence interval obtained by the BNNConvLSTM model in this embodiment are highly reliable and clear.

[0080] The seawater sound velocity profile prediction method based on BNNConvLSTM provided by the present invention can obtain the predicted value of the seawater sound velocity to be measured and the prediction interval under the specified confidence interval according to the trained ConvLSTM model and the trained BNNConvLSTM model, respectively. On the basis of not changing the original network model, the ConvLSTM model is optimized by using a random deactivation method to better measure the uncertainty in the result, fully utilizing the spatiotemporal characteristics of the sound velocity profile, giving the prediction interval while obtaining a high-precision definite prediction result, and measuring the uncertainty in the prediction caused by the variability of the marine environment.

[0081] like Figure 9 As shown, based on the same inventive concept and corresponding to the above-mentioned embodiment method, an embodiment of the present invention further provides a seawater sound velocity profile prediction device based on BNNConvLSTM, comprising:

[0082] The raw data processing module 10 is used to obtain the raw data of the sea area to be measured within a preset time period, and calculate the seawater sound velocity profile data of the sea area to be measured based on the raw data;

[0083] The sound velocity profile prediction value module 20 is used to obtain the sound velocity profile prediction data of the sea area to be measured based on the seawater sound velocity profile data using a preset ConvLSTM model;

[0084] The sound speed profile prediction interval module 30 is used to apply a preset BNNConvLSTM model according to the seawater sound speed profile data to obtain the prediction interval of the sound speed profile prediction data of the sea area to be measured under a specified confidence interval.

[0085] The apparatus of the above embodiment is used to implement the corresponding method in the above embodiment and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0086] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, an embodiment of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, the seawater sound speed profile prediction method based on BNNConvLSTM described in the above-mentioned embodiment is implemented.

[0087] Figure 10 A more specific hardware schematic diagram of an electronic device provided in this embodiment is shown. The device may include: a processor 100, a memory 200, an input / output interface 300, a communication interface 400, and a bus 500. The processor 100, the memory 200, the input / output interface 300, the communication interface 400, and the bus 500 are communicatively connected with each other within the device.

[0088] The processor 100 can be implemented using a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided by the embodiments of the present invention.

[0089] The memory 200 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage devices, dynamic storage devices, etc. The memory 200 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present invention are implemented through software or firmware, the relevant program codes are stored in the memory 200 and are called and executed by the processor 100.

[0090] The input / output interface 300 is used to connect input / output modules to implement information input and output. The input / output modules can be configured as components within the device (not shown in the figure) or can be externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and output devices may include a display, speaker, vibrator, indicator light, etc.

[0091] The communication interface 400 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via a wired method (such as USB, network cable, etc.) or a wireless method (such as mobile network, WiFi, Bluetooth, etc.).

[0092] The bus 500 comprises a path for transmitting information between the various components of the device (eg, the processor 100 , the memory 200 , the input / output interface 300 , and the communication interface 400 ).

[0093] It should be noted that although the above device only shows the processor 100, the memory 200, the input / output interface 300, the communication interface 400, and the bus 500, in a specific implementation, the device may also include other components necessary for normal operation. In addition, it will be understood by those skilled in the art that the above device may only include the components necessary to implement the embodiments of this specification, and does not necessarily include all the components shown in the figures.

[0094] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, one embodiment of the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the seawater sound speed profile prediction method based on BNNConvLSTM as described in the above-mentioned embodiment.

[0095] The computer-readable storage media of this embodiment includes permanent and non-permanent, removable and non-removable media that can be used to store information by any method or technology; the information can be computer-readable instructions, data structures, program modules or other data. Examples of computer-readable storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computer device.

[0096] The computer instructions stored in the computer storage medium of the above embodiment are used to enable the computer to execute the seawater sound velocity profile prediction method based on BNNConvLSTM as described in the above embodiment, and have the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0097] 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 the present invention is limited to these examples. Within the scope of the present invention, 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 in different aspects of the embodiments of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0098] The embodiments of the present invention are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of the present invention. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention should be included in the scope of protection of the present invention.

Claims

1. A seawater sound velocity profile prediction method based on BNNConvLSTM, characterized in that: include: Obtaining raw data within a preset time period of the sea area to be measured, and calculating the seawater sound velocity profile data of the sea area to be measured based on the raw data, including: obtaining a raw data set of seawater within a preset time period, performing high-resolution data processing of the raw data set in a depth dimension, and applying a preset sound velocity empirical formula to calculate the seawater sound velocity profile data of the sea area to be measured, wherein the sound velocity empirical formula is: Where C is the speed of sound in m / s, T is the temperature in °C, S is the salinity in ‰, and Z is the depth in m. Applying a preset ConvLSTM model according to the seawater sound velocity profile data to obtain sound velocity profile prediction data of the sea area to be measured; Before applying a preset ConvLSTM model to obtain the sound speed profile prediction data of the sea area to be measured based on the seawater sound speed profile data, the method includes: training the ConvLSTM model and the BNNConvLSTM model, including: obtaining original data of seawater in a preset historical period, and calculating the seawater sound speed profile data in the preset historical period to construct an input data set; dividing the seawater sound speed profile data into a training set and a test set, wherein the training set and the test set respectively include a plurality of sound speed-depth samples; applying the training set and the test set to train the ConvLSTM model to obtain the trained ConvLSTM model; constructing the BNNConvLSTM model based on the trained ConvLSTM model and training it; The constructing of the input data set includes: constructing the seawater sound velocity profile data into a five-dimensional tensor of (samples, time, rows, cols, channels) to form an input data set; The step of constructing the BNNConvLSTM model based on the trained ConvLSTM model and training the model includes: setting the Dropout layer to function during testing based on the trained ConvLSTM model to construct a BNNConvLSTM model; and training the BNNConvLSTM model using the training set and the test set to obtain the trained BNNConvLSTM model. The preset BNNConvLSTM model is applied according to the seawater sound speed profile data to obtain the prediction interval of the sound speed profile prediction data of the sea area to be measured under the specified confidence interval.

2. The seawater sound velocity profile prediction method based on BNNConvLSTM according to claim 1, characterized in that: The original data set includes temperature data, salinity data and depth data.

3. The seawater sound velocity profile prediction method based on BNNConvLSTM according to claim 1 or 2, characterized in that The constructing of the input data set further includes: performing normalization processing on the seawater sound velocity profile data to eliminate dimension effects; Among them, samples represents the number of sound speed-depth samples of the input seawater sound speed profile data, time represents the prediction time step, rows is set to represent the depth features in the sound speed-depth samples, cols is set to represent the sound speed data in the sound speed-depth samples, cols is set to 1 column, and channels is set to 1.

4. The seawater sound velocity profile prediction method based on BNNConvLSTM according to claim 1 or 2, characterized in that: The ConvLSTM model includes two ConvLSTM layers, and each ConvLSTM layer is connected to a BatchNormalization layer and a Dropout layer, respectively. The number of convolution kernels in both ConvLSTM layers is 50, the convolution kernel size of the first ConvLSTM layer is 6×6, and the convolution kernel size of the second ConvLSTM layer is 4×4. The attenuation rate of the BatchNormalization layer is set to 0.99, and the probability of the Dropout layer is set to 0.

2. The optimization algorithm of the ConvLSTM model is set to Adam, and the activation function of the ConvLSTM model is set to Tanh function.

5. The seawater sound velocity profile prediction method based on BNNConvLSTM according to claim 1 or 2, characterized in that: The applying the training set and the test set to train the ConvLSTM model to obtain the trained ConvLSTM model includes: The ConvLSTM model is trained according to the training set and the test set, wherein the batch size is set to 32, the epoch maximum is set to 800, and early stopping is set. Training is stopped when the loss no longer decreases for 10 consecutive times or when the maximum value set by the epoch is reached, thereby obtaining the trained ConvLSTM model.

6. A seawater sound velocity profile prediction device based on BNNConvLSTM, characterized in that: include: The raw data processing module is used to obtain raw data within a preset time period of the sea area to be measured, and calculate the seawater sound velocity profile data of the sea area to be measured based on the raw data, including: obtaining a raw data set of seawater within a preset time period, performing high-resolution data processing on the raw data set in the depth dimension, and applying a preset sound velocity empirical formula to calculate the seawater sound velocity profile data of the sea area to be measured, wherein the sound velocity empirical formula is: Where C is the speed of sound in m / s, T is the temperature in °C, S is the salinity in ‰, and Z is the depth in m. A sound speed profile prediction value module is used to apply a preset ConvLSTM model to obtain the sound speed profile prediction data of the sea area to be measured according to the seawater sound speed profile data; before applying the preset ConvLSTM model to obtain the sound speed profile prediction data of the sea area to be measured according to the seawater sound speed profile data, it includes: training the ConvLSTM model and the BNNConvLSTM model, including: obtaining the original data of seawater in a preset historical period, and calculating the seawater sound speed profile data in the preset historical period to construct an input data set; dividing the seawater sound speed profile data into a training set and a test set, the training set and the test set respectively including a number of sound speed-depth samples; applying the training set and the test set to train the ConvLSTM model to obtain the trained ConvLSTM model; constructing the BNNConvLSTM model based on the trained ConvLSTM model and training it; The constructing of the input data set includes: constructing the seawater sound velocity profile data into a five-dimensional tensor of (samples, time, rows, cols, channels) to form an input data set; The step of constructing the BNNConvLSTM model based on the trained ConvLSTM model and training the model includes: setting the Dropout layer to function during testing based on the trained ConvLSTM model to construct a BNNConvLSTM model; and training the BNNConvLSTM model using the training set and the test set to obtain the trained BNNConvLSTM model. The sound speed profile prediction interval module is used to apply a preset BNNConvLSTM model according to the seawater sound speed profile data to obtain the prediction interval of the sound speed profile prediction data of the sea area to be measured under a specified confidence interval.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the seawater sound velocity profile prediction method based on BNNConvLSTM as described in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute the seawater sound speed profile prediction method based on BNNConvLSTM as described in any one of claims 1 to 5.

Images