A method for obtaining ocean sound velocity profile based on self-organizing competitive neural network

Abstract

The application discloses a method for obtaining ocean sound velocity profile based on self-organizing competitive neural network, and relates to the field of ocean sound velocity profile. First, a sample profile of historical sound velocity profile is expressed as an average value plus a form of orthogonal empirical function vector multiplied by corresponding profile coefficients An; then, historical sea surface parameters Xn are processed; An in S1 and the historical sea surface parameters Xn in S2 are combined to form a sample training vector; a self-organizing competitive neural network algorithm is used to train the sample set, and an artificial neuron topological structure is formed; then, real-time sea surface parameters are obtained, and the real-time sea surface parameters are input into the neuron topological structure to obtain An corresponding to the real-time sea surface parameters; finally, the sample profile is expressed as the form of An by combining the historical average sound velocity profile, the EOF vector and the real-time EOF coefficient, and a real-time ocean sound velocity profile is obtained. The application can improve the precision of the obtained sound velocity profile by exploring the relationship between the sound velocity profile and the sea surface remote sensing parameters in the sample data through the artificial neural network.

Description

Technical Field

[0001] This invention relates to the field of ocean sound velocity profiling technology, and more specifically, to a method for obtaining ocean sound velocity profiles based on a self-organizing competitive neural network. Background Technology

[0002] Sound velocity profiles are the most fundamental acoustic parameters of the ocean, providing essential marine environmental information for underwater acoustic applications such as underwater target identification, marine environmental monitoring, and underwater communication. Because the ocean is a constantly changing and complex system, ocean sound velocity profiles exhibit strong temporal and spatial variability. Therefore, accurate, large-area, real-time acquisition of ocean sound velocity profiles is a pressing technical problem that needs to be solved.

[0003] The most direct method for obtaining sound velocity profiles is through on-site measurement using equipment such as sound velocity meters, but this is costly and difficult to perform over large areas. Ocean satellite remote sensing can meet the needs of large-scale and real-time ocean measurements, but its measured parameters are limited to the water surface. In recent years, with the continuous development of observation technologies such as Argo buoys and remote sensing, ocean profile and surface data samples have been continuously accumulated. By exploring the relationship between ocean profiles and surface data, deriving underwater sound velocity profiles from sea surface remote sensing data has become the most important means of obtaining ocean sound velocity profiles over large areas in real time, and has become the main development direction of related technologies. In existing technologies, the method of using an orthogonal empirical function (EOF) plus linear fitting can accurately invert water profile parameters, namely the single empirical orthogonal function regression (sEOF-r) method. The US Navy has adopted this method as part of its modular marine environmental data prediction. In global ocean sound velocity profile reconstruction, the sEOF-r method has shown a certain level of accuracy, but it has also shown significant errors in some sea areas, and it still cannot meet the accuracy requirements of many underwater sonar applications for sound velocity profile information.

[0004] To address this, the present invention proposes a method for obtaining ocean sound velocity profiles based on a self-organizing competitive neural network. This method utilizes an artificial neural network to explore the relationship between the sound velocity profile contained in the sample data and the remote sensing parameters of the sea surface, thereby improving the accuracy of the sound velocity profile acquisition method and solving the problems existing in the prior art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for obtaining ocean sound velocity profiles based on a self-organizing competitive neural network. This invention uses a self-organizing competitive neural network to replace the traditional linear regression inversion framework. By utilizing the topological structure of neurons, it can overcome the constraints of linear relationships in traditional regression methods and use a nonlinear inversion framework to express nonlinear relationships. By exploring the relationship between the sound velocity profile and sea surface remote sensing parameters contained in the sample data through artificial neural networks, the accuracy of the obtained sound velocity profile can be improved.

[0006] The above-mentioned technical objective of this invention is achieved through the following technical solution: a method for obtaining ocean sound velocity profiles based on a self-organizing competitive neural network, specifically including the following steps:

[0007] S1. Perform EOF processing on the historical sound velocity profiles and represent the sample profile of each historical sound velocity profile as the average value plus the orthogonal empirical function vector multiplied by the corresponding profile coefficient An of each order.

[0008] S2. Process the historical sea surface parameter Xn, and the amount of the selected historical sea surface parameter is not limited;

[0009] S3. Using days as the time unit, the profile coefficient An from step S1 and the historical sea surface parameter Xn from step S2 are combined into a sample training vector. Data from the same day are combined into a training vector [A1, A2, ..., An, X1, X2, ..., Xn].

[0010] S4. The training sample set is trained using a self-organizing competitive neural network algorithm to form an artificial neuron topology structure in a machine learning manner. The number of neurons is set to the number of samples according to the needs of obtaining the sound speed profile, that is, each training sample is a separate class.

[0011] S5. Based on the location of the sea area where the sound velocity profile needs to be obtained, obtain the corresponding real-time sea surface parameters;

[0012] S6. Based on the known real-time sea surface parameters, input them into the neuron topology structure described in step S4, determine the corresponding neurons, and obtain the EOF profile coefficients An corresponding to the real-time sea surface parameters through the corresponding neurons.

[0013] S7. Combining the historical average sound velocity profile of the sea area, the EOF vector, and the EOF profile coefficients An of each order corresponding to the real-time sea surface parameters obtained in step S6, each sample profile is represented as the average value plus the orthogonal empirical function vector multiplied by the corresponding coefficients An of each order, thus obtaining the real-time ocean sound velocity profile.

[0014] Furthermore, the historical sound speed profile described in step S1 is obtained by calculating Argo buoy data.

[0015] Furthermore, the historical sea surface parameters mentioned in step S2 include sea surface height anomalies, temperature anomalies, wind speed anomalies, wave height anomalies, heat flux anomalies, and month.

[0016] In summary, the present invention has the following beneficial effects:

[0017] 1. There are complex nonlinear relationships among various ocean parameters. This invention uses a self-organizing competitive neural network to replace the traditional linear regression inversion framework. By utilizing the topological structure of neurons, the limitations of linear relationships in traditional regression methods can be overcome. The nonlinear inversion framework is used to express nonlinear relationships, and its accuracy is significantly higher than that of the linear inversion framework in the prior art, thus improving the accuracy of obtaining sound velocity profiles.

[0018] 2. The number of neurons in the neuron topology of the self-organizing competitive neural network used in this invention can be consistent with the number of samples, which can ensure that the relationship between various parameters is fully subdivided; at the same time, compared with the traditional linear regression method, the number of neurons in the neuron topology of this invention can ensure that the individual features of the data are discovered, which can improve the accuracy of obtaining the sound velocity profile.

[0019] 3. The ocean is a complex multi-parameter system, and its sound velocity profile is constrained by various factors, including waves and the month. In this invention, a self-organizing competitive neural network is used to replace the traditional linear regression inversion framework. Multiple parameters such as wind speed, heat flux, and month can be introduced. As the solution conditions are increased and refined, it is beneficial to explore the influence of multiple factors on the sound velocity profile and improve the inversion accuracy. Attached Figure Description

[0020] Figure 1 This is a flowchart from an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0022] Example: A method for obtaining ocean sound velocity profiles based on a self-organizing competitive neural network, such as... Figure 1 As shown, the specific steps include:

[0023] S1. Perform EOF processing on the historical sound speed profiles, and represent the sample profile of each historical sound speed profile as the average value plus the orthogonal empirical function vector multiplied by the corresponding profile coefficient An.

[0024] S2. Process the historical sea surface parameter Xn, and the number of historical sea surface parameters selected is unlimited.

[0025] S3. Using days as the time unit, the profile coefficient An from step S1 and the historical sea surface parameter Xn from step S2 are combined into a sample training vector. Data from the same day are combined into a training vector [A1, A2, ..., An, X1, X2, ..., Xn].

[0026] S4. The training sample set is trained using a self-organizing competitive neural network algorithm to form an artificial neuron topology in a machine learning manner. The number of neurons is set to the number of samples according to the needs of obtaining the sound speed profile, that is, each training sample is a separate class.

[0027] S5. Based on the location of the sea area where the sound velocity profile needs to be obtained, obtain the corresponding real-time sea surface parameters.

[0028] S6. Based on the known real-time sea surface parameters, input them into the neuron topology in step S4, determine the corresponding neurons, and obtain the EOF profile coefficients An corresponding to the real-time sea surface parameters through the corresponding neurons.

[0029] S7. Combining the historical average sound velocity profile of the sea area, the EOF vector, and the EOF profile coefficients An of each order corresponding to the real-time sea surface parameters obtained in step S6, each sample profile is represented as the average value plus the orthogonal empirical function vector multiplied by the corresponding coefficients An of each order, thus obtaining the real-time ocean sound velocity profile.

[0030] In step S1, the historical sound speed profile is obtained by calculating Argo buoy data.

[0031] The historical sea surface parameters in step S2 include sea surface height anomalies, temperature anomalies, wind speed anomalies, wave height anomalies, heat flux anomalies, and month anomalies, etc.

[0032] In this embodiment, there are no restrictions on the parameters that can be selected; any information that the user considers to have an impact on the underwater profile and that can be obtained is acceptable.

[0033] The ocean sound velocity profile acquisition method based on a self-organizing competitive neural network in this embodiment has the following advantages:

[0034] 1. There are complex nonlinear relationships among various ocean parameters. This invention uses a self-organizing competitive neural network to replace the traditional linear regression inversion framework. By utilizing the topological structure of neurons, the limitations of linear relationships in traditional regression methods can be overcome. The nonlinear inversion framework is used to express nonlinear relationships, and its accuracy is significantly higher than that of the linear inversion framework in the prior art, thereby improving the accuracy of obtaining sound velocity profiles.

[0035] 2. The number of neurons in the neuron topology of the self-organizing competitive neural network used in this invention can be consistent with the number of samples, ensuring that the relationships between various parameters are sufficiently subdivided. Furthermore, compared to traditional linear regression methods, the number of neurons in the neuron topology of this invention ensures that individual characteristics of the data are discovered, thereby improving the accuracy of obtaining sound velocity profiles.

[0036] 3. The ocean is a complex multi-parameter system, and its sound velocity profile is influenced by various factors, including waves and the month. In this invention, by employing a self-organizing competitive neural network to replace the traditional linear regression inversion framework, multiple parameters such as wind speed, heat flux, and month can be introduced. As the solution conditions increase and become more refined, it is beneficial to uncover the influence of various factors on the sound velocity profile and improve the inversion accuracy.

[0037] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A method for obtaining ocean sound velocity profiles based on a self-organizing competitive neural network, characterized by: Specifically, the following steps are included: S1. Perform EOF processing on the historical sound velocity profiles and represent the sample profile of each historical sound velocity profile as the average value plus the orthogonal empirical function vector multiplied by the corresponding profile coefficient An of each order. S2. Process the historical sea surface parameters Xn, and the amount of the selected historical sea surface parameters is unlimited; wherein the historical sea surface parameters include sea surface height anomalies, temperature anomalies, wind speed anomalies, wave height anomalies, heat flux anomalies, and month; S3. Using days as the time unit, the profile coefficient An from step S1 and the historical sea surface parameter Xn from step S2 are combined into a sample training vector. Data from the same day are combined into a training vector [A1, A2, ..., An, X1, X2, ..., Xn]. S4. The training sample set is trained using a self-organizing competitive neural network algorithm to form an artificial neuron topology structure in a machine learning manner. The number of neurons is set to the number of samples according to the needs of obtaining the sound speed profile, that is, each training sample is a separate class. S5. Based on the location of the sea area where the sound velocity profile needs to be obtained, obtain the corresponding real-time sea surface parameters; S6. Based on the known real-time sea surface parameters, input them into the neuron topology structure described in step S4, determine the corresponding neurons, and obtain the EOF profile coefficients An corresponding to the real-time sea surface parameters through the corresponding neurons. S7. Combining the historical average sound velocity profile of the sea area, the EOF vector, and the EOF profile coefficients An of each order corresponding to the real-time sea surface parameters obtained in step S6, each sample profile is represented as the average value plus the orthogonal empirical function vector multiplied by the corresponding coefficients An of each order, thus obtaining the real-time ocean sound velocity profile.

2. The method for obtaining ocean sound velocity profiles based on a self-organizing competitive neural network according to claim 1, characterized in that: The historical sound speed profile mentioned in step S1 is obtained by calculating Argo buoy data.

Images