Gradient amplification and condensation method and equipment based on sound propagation characteristics

Abstract

The invention discloses a gradient amplification and condensation method and equipment based on sound propagation characteristics. The method comprises the following steps: acquiring three-dimensional physical ocean data of a research area, and calculating a sound velocity field of the research area; dividing the sound velocity field into a shallow sea profile and a deep sea profile according to the effective depth of each sound velocity profile in the sound velocity field, and performing normalization and interpolation processing on multiple groups of sound velocity profiles in the shallow sea profile and the deep sea profile; calculating a sound velocity gradient matrix according to the plurality of groups of sound velocity profiles after normalization and interpolation processing, and performing pre-classification and transverse scaling on the plurality of groups of sound velocity profiles after normalization and interpolation processing according to positive and negative gradients of the sound velocity profiles; and clustering the multiple groups of sound velocity profiles after pre-classification and transverse scaling. The method is suitable for shallow sea and deep sea areas at the same time, the clustering characteristics are balanced by scaling up or scaling down the positive and negative gradients according to the proportion, the sound velocity profile clustering result can be effectively improved, the clustering result has the obvious acoustic propagation characteristic, and the research on the underwater acoustic environment effect can be conveniently carried out.

Description

Technical Field

[0001] This invention relates to the field of ocean sound velocity profile clustering technology, specifically to a gradient scaling clustering method and device based on sound propagation characteristics. Background Technology

[0002] Ocean sound velocity profile clustering uses an unsupervised method to classify sound velocity profiles with similar structures into one class, thereby dividing the research sea area into different classification sea areas with different characteristics. By studying the underwater acoustic environment characteristics of each classification sea area, it is beneficial to quickly and accurately determine the sound propagation law in a specific sea area during underwater acoustic warfare.

[0003] However, using conventional methods to cluster sound velocity profiles is insufficient for sound field analysis. Influenced by different latitude and longitude sea areas, directly using sound velocity profiles for clustering focuses more on the numerical magnitude of the sound velocity profiles than on their shape and structure, failing to effectively distinguish between sea areas favorable and unfavorable for sound propagation. Using sound velocity gradient profiles for clustering is affected by the significant positive and negative values ​​and extreme positive and negative values ​​of the gradient in the study sea area, easily causing the cluster centers to become unbalanced in terms of gradient changes, and even resulting in the loss of regions with significant acoustic significance. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a gradient scaling clustering method and device based on sound propagation characteristics.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a gradient scaling clustering method based on sound propagation characteristics, comprising: Three-dimensional physical oceanographic data of the study area are collected, and sound velocity data at each point in the study area are calculated based on the three-dimensional physical oceanographic data to obtain the sound velocity field of the study area. The sound velocity field is divided into shallow sea profile and deep sea profile according to the effective depth of each sound velocity profile in the sound velocity field, and the multiple sets of sound velocity profiles in the shallow sea profile and deep sea profile are normalized and interpolated respectively. The sound velocity gradient matrix is ​​calculated based on multiple sets of sound velocity profiles after normalization and interpolation. The multiple sets of sound velocity profiles after normalization and interpolation are pre-classified and horizontally scaled according to the positive and negative gradients of the sound velocity profiles. Clustering was performed on multiple sets of sound velocity profiles after pre-classification and horizontal scaling, and the clustering results were summarized and organized.

[0006] Furthermore, the three-dimensional physical ocean data includes temperature and salinity data at different latitudes, longitudes, and depths within the study area, and the sound speed data is calculated as follows: ; in, For the sound speed data at any point in the study area, This represents the salinity of the seawater at that point. The seawater temperature at that point. The depth of the seawater at that point.

[0007] Furthermore, the normalization method is as follows: ; in, Scaling factor This represents the normalized standard depth. The normalized standard depth of the deep-sea profile is greater than that of the shallow-sea profile. The maximum effective depth for each profile. and These are the normalized depth and speed of sound, respectively. Represents the speed of sound at the surface.

[0008] Furthermore, the Akima interpolation method is used for interpolation processing, and the interpolation depth interval of the deep-sea profile is greater than that of the shallow-sea profile.

[0009] Furthermore, the method for calculating the sound velocity gradient at different depths of a single sound velocity profile in the sound velocity gradient matrix is ​​as follows: ; in, Let be the sound speed gradient at the i-th point in the sound speed profile. The symbol is for partial differentials. Let be the speed of sound at the i-th point after interpolation. Let be the depth of the i-th point after interpolation.

[0010] Furthermore, the method for pre-classifying multiple sets of sound velocity profiles after normalization and interpolation according to the number of sign changes of the positive and negative gradients of the sound velocity profile is as follows: (1) Type I: positive gradient, negative gradient; (2) Second type: positive and negative gradients, negative and positive gradients; (3) Third type: positive and negative positive gradients, negative positive and negative gradients; (4) Fourth type: positive-negative-positive-negative gradient, negative-positive-negative-positive gradient; (5) Fifth category: positive-negative-positive-negative-positive gradient, negative-positive-negative-positive-negative gradient; Among them, for sound speed profiles containing more than five different positive and negative gradient sign changes, even numbers are assigned to the fourth category, and odd numbers are assigned to the fifth category.

[0011] Furthermore, the lateral scaling method is as follows: ; in, Let be the gradient vector of the i-th profile after horizontal scaling. This is the horizontal scaling factor of the gradient. Let be the gradient vector of the i-th profile before horizontal scaling. , and These are the sets of vectors representing the positive and negative gradients of the profile, respectively. To find the function with the maximum value, To find the minimum value of the function.

[0012] Furthermore, before clustering the multiple sets of sound velocity profiles after pre-classification and horizontal scaling, the gradient vectors of the horizontally scaled profiles are first standardized using Z-score, as follows: ; in, This is the standardized profile gradient vector. The mean of all sample data. is the standard deviation of all sample data.

[0013] Furthermore, the clustering results are summarized and organized, including summarizing and visualizing the category information of shallow sea and deep sea profiles, and using evaluation indicators to evaluate the clustering results, including the Calinski-Harabasz index, the Davidson-Boldt index, and the silhouette coefficient.

[0014] In a second aspect, the present invention provides a sound velocity profile clustering device based on gradient pre-classification scaling, comprising a storage medium and a processor, wherein the storage medium stores a computer program, which, when executed by the processor, is used to implement the above-described method.

[0015] Beneficial effects: (1) This invention proposes a sound velocity profile clustering method that focuses more on the propagation of beneficial or non-beneficial sound. This method is applicable to both shallow and deep sea areas. It effectively extracts feature information such as surface sound channels and deep sea sound channels through positive and negative gradient pre-classification and performs further unsupervised clustering. (2) In order to avoid the imbalance of clustering results caused by the significant positive and negative gradient values ​​and extreme positive and negative values ​​of the background of the sea area under study, this invention proposes a horizontal scaling method for positive and negative gradient profiles. By scaling up or down the positive and negative gradients according to the proportion, the clustering characteristics can be balanced, which can effectively improve the sound velocity profile clustering results. Moreover, the clustering results have obvious acoustic propagation characteristics, which is convenient for carrying out research on underwater acoustic environment effects. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the gradient scaling clustering method based on sound propagation characteristics. Figure 2 These are comparison images of shallow sea profiles before and after normalization; Figure 3 This is a diagram showing the effect of clustering the second type of positive and negative gradients after horizontal scaling of the gradient profile. Detailed Implementation

[0017] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0018] like Figure 1 As shown, this embodiment of the invention provides a gradient scaling clustering method based on sound propagation characteristics, including: Three-dimensional physical oceanographic data of the study area were collected, and the sound velocity data at each point in the study area was calculated based on the three-dimensional physical oceanographic data to obtain the sound velocity field of the study area. The aforementioned three-dimensional physical oceanographic data includes temperature and salinity data at different latitudes, longitudes, and depths within the study area. The sound velocity data was calculated as follows: ; in, For the sound speed data at any point in the study area, This represents the salinity of the seawater at that point. The seawater temperature at that point. The depth of the seawater at that point is given by the formula above. The data calculated using the formula is for a single point. The set of single-point data at the same latitude and longitude is the sound speed profile, and the set of all sound speed profiles is the sound speed field.

[0019] The sound velocity field is divided into shallow-sea and deep-sea profiles based on the effective depth of each sound velocity profile. Multiple sound velocity profiles within each profile are then normalized and interpolated. The preferred effective depth is 200 meters. The normalization method is as follows: ; in, Scaling factor This represents the normalized standard depth. The normalized standard depth of the deep-sea profile is greater than that of the shallow-sea profile. Preferably, the normalized standard depth of the deep-sea profile is set to 2000, and the normalized standard depth of the shallow-sea profile is set to 200. The maximum effective depth for each profile. and These are the normalized depth and speed of sound, respectively. This represents the surface sound speed. The normalized sound speed profile has the same sound speed gradient as the original sound speed profile. See also Figure 2 , Figure 2 The illustration shows the comparison before and after the shallow sea profile was normalized.

[0020] The Akima interpolation method can be used for interpolation, with the interpolation depth interval for deep-sea profiles being greater than that for shallow-sea profiles. Preferably, the interpolation depth interval for deep-sea profiles is set to 20 meters, and the interpolation depth interval for shallow-sea profiles is set to 10 meters.

[0021] The sound velocity gradient matrix is ​​calculated based on multiple sets of normalized and interpolated sound velocity profiles. The normalized and interpolated sound velocity profiles are then pre-classified and laterally scaled according to their positive and negative gradients. The method for calculating the sound velocity gradient at different depths for a single sound velocity profile in the above sound velocity gradient matrix is ​​as follows: ; in, Let be the sound speed gradient at the i-th point in the sound speed profile. The symbol is for partial differentials. Let be the speed of sound at the i-th point after interpolation. Let be the depth of the i-th point after interpolation.

[0022] The method for pre-classifying multiple sets of sound velocity profiles after normalization and interpolation based on the number of sign changes of the positive and negative gradients of the sound velocity profile is as follows: (1) Type I: positive gradient, negative gradient; (2) Second type: positive and negative gradients, negative and positive gradients; (3) Third type: positive and negative positive gradients, negative positive and negative gradients; (4) Fourth type: positive-negative-positive-negative gradient, negative-positive-negative-positive gradient; (5) Fifth category: positive-negative-positive-negative-positive gradient, negative-positive-negative-positive-negative gradient; Among them, for sound speed profiles containing more than five different positive and negative gradient sign changes, even numbers are assigned to the fourth category, and odd numbers are assigned to the fifth category.

[0023] The above-mentioned lateral scaling enlarges or reduces the positive and negative gradient portions of a single profile to the [-1,0] and [0,1] intervals respectively according to a certain proportion. The lateral scaling method is as follows: ; in, Let be the gradient vector of the i-th profile after horizontal scaling. This is the horizontal scaling factor of the gradient. Let be the gradient vector of the i-th profile before lateral scaling, and let be the set of profile gradient vectors. , and These are the sets of vectors representing the positive and negative gradients of the profile, respectively. To find the function with the maximum value, To find the minimum value of the function.

[0024] Clustering was performed on multiple sets of sound velocity profiles after pre-classification and horizontal scaling, and the clustering results were summarized and organized. Before clustering the multiple sets of sound velocity profiles after pre-classification and horizontal scaling, the gradient vectors of the horizontally scaled profiles were first standardized using Z-score, as follows: ; in, This is the standardized profile gradient vector. The mean of all sample data. Let be the standard deviation of all sample data. Then, the K-means algorithm is used for clustering, and the average value of each cluster is taken as the gradient profile of the classification center.

[0025] Methods using direct clustering based on sound velocity profiles focus more on the numerical magnitude of the sound velocity profiles than on their shape and structure, failing to effectively distinguish between areas favorable and unfavorable for sound propagation. Methods using direct clustering based on sound velocity gradient profiles mostly concentrate on areas with negative gradients, indicating that clustering is significantly influenced by the positive and negative values ​​and extreme values ​​of gradients in the background of the studied sea area, neglecting many positive gradient profiles that are beneficial for sound propagation. Using positive and negative gradient pre-classification significantly improves upon the problems caused by direct clustering based on sound velocity profiles, and the clustering results are subject to forced constraints with clear physical significance. Due to the significant influence of negative gradients in the background of the studied sea area, unsupervised clusterers still underestimate the contribution of positive gradient channels. This problem is significantly improved after using a lateral scaling strategy; see [link to relevant documentation] for details. Figure 3 , Figure 3 This demonstrates the effect of using a K-Means clusterer to cluster the "positive-negative" gradient profiles in the second major category after horizontal scaling. It should be noted that the final classifier algorithm is not limited to using the K-Means method; hierarchical clustering (AGNES), fuzzy C-value clustering, and even deep learning clustering can also replace the K-Means method.

[0026] The clustering results were summarized and organized, including the aggregation and visualization of the category information of shallow and deep sea profiles, and the evaluation results were evaluated using evaluation metrics, including the Calinski-Harabasz index (CHI), the Davidson-Bourdin index (DBI), and the silhouette coefficient (SCI).

[0027] CHI is used to measure the relationship between inter-class dispersion and intra-class dispersion. A higher value indicates a better clustering effect. The specific calculation method is as follows: ; in, For the sample size, For the number of taxonomic clusters, and These are the intra-class covariance and the inter-class covariance, respectively, used to measure the degree of cohesion within and between classes.

[0028] In Data Similarity Index (DBI), the ratio of the sum of the intra-cluster diameters to the inter-cluster distance is called the similarity between classification clusters. The closer the similarity between two categories, the worse the classification effect. Therefore, a larger DBI indicates a worse clustering effect, and a smaller DBI indicates a better clustering effect. The specific calculation method is as follows: ; in, For the number of taxonomic clusters, Represents the sum of the inner diameters of the two clusters. Represents inter-cluster distance; SCI is the average silhouette coefficient of each sample point, used to measure the distance between each sample point within a cluster. A larger SCI indicates that samples within the same cluster are closer together, samples in different clusters are farther apart, and the clustering effect is better. The specific calculation method is as follows: ; in, For the number of taxonomic clusters, For the sample The average distance to other samples in the cluster. For the sample The average distance to the nearest sample in the cluster.

[0029] Based on the above embodiments, those skilled in the art can easily understand that the present invention also provides a gradient scaling clustering device based on sound propagation characteristics, including a storage medium and a processor, wherein the storage medium stores a computer program, which is executed by the processor to implement the above-described method.

[0030] The above description is merely a preferred embodiment of the present invention. It should be noted that for those skilled in the art, other parts not specifically described are existing technology or common knowledge. Several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A gradient scaling clustering method based on sound propagation characteristics, characterized in that, include: Three-dimensional physical oceanographic data of the study area are collected, and sound velocity data at each point in the study area are calculated based on the three-dimensional physical oceanographic data to obtain the sound velocity field of the study area. The sound velocity field is divided into shallow sea profile and deep sea profile according to the effective depth of each sound velocity profile in the sound velocity field, and the multiple sets of sound velocity profiles in the shallow sea profile and deep sea profile are normalized and interpolated respectively. The sound velocity gradient matrix is ​​calculated based on multiple sets of sound velocity profiles after normalization and interpolation. The multiple sets of sound velocity profiles after normalization and interpolation are pre-classified and horizontally scaled according to the positive and negative gradients of the sound velocity profiles. Clustering was performed on multiple sets of sound velocity profiles after pre-classification and horizontal scaling, and the clustering results were summarized and organized.

2. The gradient scaling clustering method based on sound propagation characteristics according to claim 1, characterized in that, The three-dimensional physical oceanographic data includes temperature and salinity data at different latitudes, longitudes, and depths within the study area. The sound speed data is calculated as follows: ； in, For the sound speed data at any point in the study area, This represents the salinity of the seawater at that point. The seawater temperature at that point. The depth of the seawater at that point.

3. The gradient scaling clustering method based on sound propagation characteristics according to claim 1, characterized in that, The normalization method is as follows: ； in, Scaling factor This represents the normalized standard depth. The normalized standard depth of the deep-sea profile is greater than that of the shallow-sea profile. The maximum effective depth for each profile. and These are the normalized depth and speed of sound, respectively. Represents the speed of sound at the surface.

4. The gradient scaling clustering method based on sound propagation characteristics according to claim 1, characterized in that, Interpolation is performed using the Akima interpolation method, and the interpolation depth interval of the deep-sea profile is greater than that of the shallow-sea profile.

5. The gradient scaling clustering method based on sound propagation characteristics according to claim 1, characterized in that, The method for calculating the sound velocity gradient at different depths of a single sound velocity profile in the sound velocity gradient matrix is ​​as follows: ； in, Let be the sound speed gradient at the i-th point in the sound speed profile. The symbol is for partial differentials. Let be the speed of sound at the i-th point after interpolation. Let be the depth of the i-th point after interpolation.

6. The gradient scaling clustering method based on sound propagation characteristics according to claim 1, characterized in that, The method for pre-classifying multiple sets of sound velocity profiles after normalization and interpolation based on the number of sign changes of the positive and negative gradients of the sound velocity profile is as follows: (1) Type I: positive gradient, negative gradient; (2) Second type: positive and negative gradients, negative and positive gradients; (3) Third type: positive and negative positive gradients, negative positive and negative gradients; (4) Fourth type: positive-negative-positive-negative gradient, negative-positive-negative-positive gradient; (5) Fifth category: positive-negative-positive-negative-positive gradient, negative-positive-negative-positive-negative gradient; Among them, for sound speed profiles containing more than five different positive and negative gradient sign changes, even numbers are assigned to the fourth category, and odd numbers are assigned to the fifth category.

7. The gradient scaling clustering method based on sound propagation characteristics according to claim 1, characterized in that, The lateral scaling method is as follows: ； in, Let be the gradient vector of the i-th profile after horizontal scaling. This is the horizontal scaling factor of the gradient. Let be the gradient vector of the i-th profile before horizontal scaling. , and These are the sets of vectors representing the positive and negative gradients of the profile, respectively. To find the function with the maximum value, To find the minimum value of the function.

8. The gradient scaling clustering method based on sound propagation characteristics according to claim 1, characterized in that, Before clustering the multiple sets of sound velocity profiles after pre-classification and horizontal scaling, the gradient vectors of the horizontally scaled profiles are first normalized using Z-score, as follows: ； in, This is the standardized profile gradient vector. The mean of all sample data. is the standard deviation of all sample data.

9. The gradient scaling clustering method based on sound propagation characteristics according to claim 1, characterized in that, The process of summarizing and organizing the clustering results includes compiling and visualizing the category information of shallow sea and deep sea profiles, and evaluating the clustering results using evaluation metrics, including the Calinski-Harabasz index, the Davidson-Boldt index, and the silhouette coefficient.

10. A gradient scaling clustering device based on sound propagation characteristics, comprising a storage medium and a processor, wherein the storage medium stores a computer program, characterized in that, When the computer program is executed by a processor, it is used to implement the method described in any one of claims 1-9.

Images