Search

CN-121980933-A - Submarine reflection loss high-precision estimation method based on PINN noise field continuation

CN121980933ACN 121980933 ACN121980933 ACN 121980933ACN-121980933-A

Abstract

The invention relates to a submarine reflection loss high-precision estimation method based on PINN noise field prolongation, which comprises the following steps of 1, setting an N-element vertical receiving array, wherein the depth of each array element is Frequency point received by it Noise field data of (2) is Step 2, dividing M distance grids according to the frequency and the calculation domain of the noise field data to be processed respectively And step 3, designing a physical information neural network PINN with the noise field data prolongation, and training the network by taking the collected noise field data as input to obtain a trained NoisePINN network. The invention effectively breaks through the limitation of spatial sampling of the vertical receiving array, is equivalent to expanding the aperture of the vertical array, and can obtain the submarine reflection loss result with higher precision and resolution.

Inventors

  • ZHENG GUANGYING
  • ZHANG SHUAISHUAI
  • Wei Xuanjie
  • ZHAO WENJUN
  • GUO XIAOWEI
  • YANG XIAOHONG
  • HU ZHEJIAN

Assignees

  • 中国船舶集团有限公司第七一五研究所

Dates

Publication Date
20260505
Application Date
20260119

Claims (10)

  1. 1. A submarine reflection loss high-precision estimation method based on PINN noise field continuation is characterized by comprising the following steps, Step 1, giving Element vertical receiving array, each array element depth is Frequency point received by it Noise field data of (2) is ; Step2, dividing according to the frequency and calculation domain of the processed noise field data A plurality of distance grids, respectively ; Step 3, designing a physical information neural network PINN with noise field data prolongation, taking the collected noise field data as input, and training the network to obtain a trained NoisePINN network; Step 4, resetting Individual array elements, wherein The array elements are virtual array elements for data prolongation, and the array element depth is Taking the real part and the imaginary part of the environment noise envelope function after the extension are predicted based on the trained NoisePINN network as input; step 5, calculating the cross spectrum density matrix of the extended noise field based on the extended environmental noise envelope function data ; Step 6, calculating the submarine reflection loss based on the extended ambient noise field cross spectral density function , In the middle of Represents glancing angle of sound ray incident on the seabed, superscript Representing the sum of the transposed conjugate, steering vector, In the middle of The wave number representing the propagation of sound, Representing imaginary units.
  2. 2. The method for estimating a high accuracy of a reflection loss of a sea floor based on PINN noise floor prolongation according to claim 1, wherein in the construction of the network of step 3 NoisePINN, a fully connected neural network is used as an approximator, comprising 1 input layer, 5 hidden layers and 1 output layer, and a sine activation function is used, and a NoisePINN network receives the reflection loss of the sea floor As input, output envelope function The real part of (2) And imaginary part Superscript And Representing the real and imaginary parts, respectively.
  3. 3. The method for estimating the high-precision of the reflection loss of the sea floor based on PINN noise floor extension according to claim 2, wherein the loss function of the NoisePINN network comprises 3 parts, namely partial differential equation loss, sea surface boundary condition loss and data fitting loss.
  4. 4. The method for estimating a high accuracy of a reflection loss of a sea floor based on PINN noise floor extension according to claim 3, wherein the partial differential equation loss Representing an envelope function of network predictions The partial differential equation as follows is satisfied, In the middle of The differential operator is represented by a differential operator, Representing the refractive index of the light and, For the reference sound speed, For the reference wave number, Representing the frequency; In the formula, Representing the total number of partial differential equation configuration points, Is the first The partial differential equation configures points at which the control partial differential equation is forcibly executed.
  5. 5. The method for estimating a high accuracy of a marine reflection loss based on PINN noise floor extension of claim 4, wherein the sea surface boundary condition loss is a loss of Representing an envelope function of network predictions The sea surface pressure relief boundary condition is satisfied, 。
  6. 6. The method for estimating a high accuracy of a marine reflection loss based on PINN noise floor extension as set forth in claim 5, wherein the data fitting loss is used to represent an ambient noise data fitting The calculation process is based on the envelope function of network prediction Prediction of In the middle of Representing a second type of hanker function; Noise field cross spectral density is determined by The distance grid is combined to predict and obtain, The data fit loss is expressed as the error of the predicted and measured noise field cross spectral density functions, 。
  7. 7. The method for estimating the high accuracy of the reflection loss of the sea floor based on PINN noise floor extension according to claim 6, wherein the composite loss function of the NoisePINN network is as follows: Wherein, the 、 、 The weights of the individual penalty terms are separately.
  8. 8. The method for estimating the reflection loss of the seabed with high precision based on PINN noise floor extension as claimed in claim 7, wherein the super parameters of the NoisePINN network are as follows: 、 、 Training point The number is set to 10000, the network architecture comprises 5 hidden layers, 160 neurons in each layer, and ADAM54 optimizers are used for optimizing To optimize the loss function.
  9. 9. The method for estimating a high accuracy of a marine reflection loss based on PINN noise floor extension according to claim 1, wherein in the data extension of step 4, the method comprises the steps of Input to trained NoisePINN network predictive extended envelope function The real part of (2) And imaginary part 。
  10. 10. The method for estimating a high accuracy of a marine reflection loss based on PINN th prolongation of a noise field according to claim 1, wherein in the construction of the cross spectral density matrix of the noise field based on the prolongation of step 5, the method is based on a post-prolongation environmental noise envelope function Calculating the extended noise field cross spectrum density matrix ; 。

Description

Submarine reflection loss high-precision estimation method based on PINN noise field continuation Technical Field The invention belongs to the technical field of underwater sound engineering, and particularly relates to a submarine reflection loss high-precision estimation method based on PINN noise field continuation. Background The submarine reflection loss is a key parameter for describing the acoustic characteristics of a submarine medium, directly influences the propagation attenuation law of sound waves in sea water, and has important application in the fields of marine acoustic detection, underwater communication, submarine geological exploration, marine environment monitoring and the like. The accurate acquisition of the submarine reflection loss has important significance for optimizing the design of a sonar system, improving the detection precision of underwater targets and inverting the property of submarine sediments. The conventional submarine reflection loss estimation method is mainly divided into two types, namely active detection and passive detection. The active detection method is used for inverting the reflection loss by transmitting acoustic wave signals with specific frequencies, collecting the reflected echoes by using a receiving array and analyzing the energy attenuation characteristics of the reflected echoes. Although the method is high in precision, the method depends on an active sound source, is easy to expose a detection position, is easy to be influenced by multi-path interference and ocean noise in a complex ocean environment, has the problems of high power consumption, complex operation and the like, and is difficult to meet the long-term and hidden ocean environment monitoring requirement. The passive detection method takes a noise field (such as wind wave noise, biological noise, shipping noise and the like) naturally existing in the marine environment as a signal source, inverts the submarine reflection loss by analyzing the spatial distribution characteristics of the noise field, has the advantages of strong concealment, no need of actively transmitting signals, long-term continuous operation and the like, and becomes a research hot spot in recent years. Among them, the passive detection technology based on the vertical receiving array is widely used for reflection loss estimation because it can directly acquire the vertical spatial distribution information of the noise field. However, the existing passive detection method based on the vertical receiving array is limited by the number of array elements and the arrangement depth range, and has the problem of insufficient space sampling density, namely, on one hand, the number of the array elements of the vertical receiving array in actual deployment is limited, noise field information in the full water depth range is difficult to cover, and on the other hand, the space between the array elements is often larger due to the limitation of equipment cost and marine environment, so that the characterization precision of the space change of the noise field is insufficient. The limitations cause larger errors in the ratio of the upper beam energy to the lower beam energy calculated by methods such as beam forming and the like, thereby affecting the estimation precision and resolution of the submarine reflection loss and being difficult to meet the requirement of high-precision marine environment inversion. In order to solve the problem of insufficient spatial sampling of the vertical receiving array, researchers put forward a data continuation method based on interpolation or statistical model, but the statistical characteristics of the data are not considered, the physical rule of marine acoustic propagation is not considered, non-physical errors are easy to generate in a complex sound field environment, the continuation data is distorted, and the estimation performance of reflection loss cannot be effectively improved. Therefore, how to combine the acoustic propagation physical law to realize the high-precision prolongation of the noise field on the basis of limited measurement data, thereby improving the estimation precision of the reflection loss of the sea bottom, and becoming the key technical problem to be solved urgently in the current marine acoustic field. Disclosure of Invention The invention aims to solve the technical problem of providing a submarine reflection loss high-precision estimation method based on Physical Information Neural Network (PINN) noise field prolongation, which effectively breaks through the limitation of vertical receiving array space sampling, is equivalent to expanding the vertical array aperture, and can obtain submarine reflection loss results with higher precision and resolution. The technical proposal of the invention is to provide a submarine reflection loss high-precision estimation method based on PINN noise field continuation, which comprises the following steps, Step 1, givingElement