Sound channel depth calculation method and system combined with ocean multi-physical elements
By combining a multidimensional feature regression machine learning model and a ConvLSTM network with multiple marine physical elements, the problem of low efficiency in traditional acoustic depth measurement is solved, enabling fast and accurate acoustic depth calculation and supporting underwater acoustic positioning and underwater acoustic communication.
Patent Information
- Application Number
- CN202511153932.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-18
- Publication Date
- 2025-12-19
AI Technical Summary
Traditional methods are inefficient and costly when measuring the depth of ocean acoustic channels, and it is difficult to quickly obtain the location of the channels, which affects ocean acoustics and environmental studies.
A multidimensional feature regression machine learning model is used in conjunction with multiple marine physical elements. By collecting sea surface data and historical data in real time, the ConvLSTM network is used for forecasting and calculating the acoustic channel depth.
It enables fast and accurate calculation of acoustic channel depth, improves the reliability and accuracy of the calculation, and supports research on underwater acoustic positioning and underwater acoustic communication.
Smart Images
Figure CN121167660A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of ocean mapping and underwater communication, and particularly relates to a method and system for calculating sound channel depth by combining multiple physical elements of the ocean. BACKGROUND
[0002] Sound waves are the main carrier of ocean mapping and underwater communication. Generally, due to the influence of seawater temperature, salinity and density, the speed of sound waves propagating in the ocean changes accordingly. In the depth direction, the sound speed generally has a minimum value, and the curve formed by all the minimum values is the sound channel. When the sound source is located near the sound channel axis, the sound waves will bend around the sound channel axis due to the refraction law, resulting in the sound energy being concentrated between two water layers with equal sound speed, and it is not easy to dissipate significantly. Therefore, sound waves can propagate hundreds or even thousands of kilometers, which is an important channel for underwater long-distance communication, positioning, navigation, command and control. Therefore, it is important to master and understand the distribution and variation of the underwater sound speed field, and to accurately and quickly obtain the sound channel depth (SCD) position, which plays an important role in underwater environment detection and perception, ocean engineering construction, ocean resource development, and ocean security protection.
[0003] To accurately obtain the SCD, the traditional method is to measure the temperature and salinity profile, and then calculate the sound speed of each water layer in the ocean using the sound speed empirical formula, and then analyze the sound channel position. For example, in the area to be measured, a temperature and salinity sensor (CTD, XTD, etc.) is lowered, and the temperature and salinity elements of each depth layer are measured during the lowering process. This method is flexible to arrange, but the efficiency of each measurement is low. Or use an anchor hydrological observation station to continuously and real-time collect temperature and salinity data near the anchor chain, but it is often fixed at a certain position and not easy to move. Argo (array for real-time geostrophic oceanography) floats can be used to widely obtain hydrological profiles in a large sea area, and the measurement automation degree is high, but the data results are often delayed for about 10 days. No matter which of the above methods, the hydrological elements of each water layer need to be observed, which will take a lot of time, manpower and material resources, and the cost is high and the efficiency is low. Therefore, if a method can be used to calculate the SCD by measuring a few elements at the sea surface, it will greatly reduce the cost and increase the efficiency, and will have a good promoting effect on the research of ocean sound channel and related fields such as ocean acoustics and ocean environment. SUMMARY
[0004] The present application proposes a method and system for calculating sound channel depth by combining multiple physical elements of the ocean to solve the problem of low efficiency in determining the sound channel position by using traditional sound speed profile measurement.
[0005] To achieve the above purpose, the present application adopts the following technical scheme:
[0006] The application discloses a method for calculating the sound channel depth of a marine multi-physical element.
[0007] Step 1: obtaining historical profile water temperature and profile salinity data of a to-be-measured sea area, and then obtaining historical sound velocity profile data of the to-be-measured sea area;
[0008] Step 2: calculating the position of the minimum sound velocity and recording the corresponding sound channel depth by using the sound velocity profile data;
[0009] Step 3: collecting sea surface hydrological data, sea surface meteorological data, marine gravity field data, sea surface dynamic topographic data and seabed topographic data of the to-be-measured sea area in a corresponding period;
[0010] Step 4: taking the data collected in step 3 and corresponding space-time information as features of a training set, taking the sound channel depth in step 2 as a training set label, inputting a multi-dimensional feature regression machine learning model for training, and statistically analyzing the quantitative relationship between the label and the features;
[0011] Step 5: real-time collecting sea surface hydrological data and sea surface meteorological data of the to-be-measured sea area, combining the space-time information of the to-be-measured sea area, marine gravity field data, seabed topographic data and sea surface dynamic topographic data, and inputting the trained multi-dimensional feature regression machine learning model to obtain the required sound channel depth value.
[0012] Further, the sea surface hydrological data includes water temperature, salinity and water flow; the sea surface meteorological data includes air temperature, air pressure and wind speed.
[0013] Further, the water flow and wind speed data include eastward and northward components.
[0014] Further, the marine gravity field data includes marine gravity anomaly, marine vertical deflection, eastward gradient, northward gradient and radial gradient of the marine gravity anomaly.
[0015] Further, step 3 includes: combining historical data and a ConvLSTM network to perform 24-hour step prediction on the sea surface dynamic topography, and obtaining a prediction value of the sea surface dynamic topography on the same day.
[0016] Further, in step 3, the seabed topographic data is obtained in the following manner: taking a to-be-measured point as the center, expanding 1°*1° or a higher range, and statistically analyzing the terrain standard deviation, slope, gradient and kurtosis in the to-be-measured sea area.
[0017] Further, step 3 further includes: uniformly gridding the collected data into 0.25°*0.25° or higher spatial resolution grid data.
[0018] Further, before step 5, a measurement system is further built in the following manner:
[0019] The GNSS receiver, the meteorological sensor and the hydrological sensor are arranged on a water surface platform, and a set of industrial personal computer and data acquisition system loaded with a trained multi-dimensional feature regression machine learning model and marine gravity field data and seabed topography data are carried.
[0020] Further, the meteorological sensor comprises a thermometer, a barometer and an anemometer; and the hydrological sensor comprises a water thermometer, a salinity meter and a water flow meter.
[0021] Another aspect of the present application provides a sound channel depth calculation system combined with marine multi-physical elements, comprising:
[0022] a historical sound velocity profile data calculation module, configured to acquire historical profile water temperature and profile salinity data of a to-be-measured sea area, and further obtain historical sound velocity profile data of the to-be-measured sea area;
[0023] a sound channel depth calculation module, configured to calculate a position of a sound velocity minimum value by using the sound velocity profile data, and record a corresponding sound channel depth;
[0024] a historical marine multi-physical element collection module, configured to collect sea surface hydrological data, sea surface meteorological data, marine gravity field data, sea surface dynamic topography data and seabed topography data of the to-be-measured sea area in a corresponding time period;
[0025] a model training module, configured to input data collected by the historical data collection module and corresponding space-time information as features of a training set, input the sound channel depth in the sound channel depth calculation module as a label of the training set, input a multi-dimensional feature regression machine learning model for training, and statistically analyze a quantitative relationship between the label and the features;
[0026] a real-time sound channel depth derivation module, configured to acquire real-time sea surface hydrological data and sea surface meteorological data of the to-be-measured sea area, combine space-time information of the to-be-measured sea area, marine gravity field data, seabed topography data and sea surface dynamic topography data, input the trained multi-dimensional feature regression machine learning model, and obtain a required sound channel depth value.
[0027] Compared with the prior art, the present application has the beneficial effects of:
[0028] The application calculates the sound channel depth of a to-be-measured point according to multiple types of marine physical elements in a research area. First, the influence of various factors on the sound channel depth is fully considered by using multiple types of marine physical elements, thereby improving the reliability of the calculation. Second, the deep features between data are effectively captured by using ConvLSTM and a machine learning method, thereby further improving the calculation accuracy of the sound channel depth. Thus, the problem of low efficiency of traditional sound velocity profile measurement can be successfully solved, the sound channel depth value below the carrier can be measured in real time according to the position of the carrier, and effective technical support can be provided for the research and application in the fields of underwater acoustic positioning, underwater acoustic communication, target detection and the like. BRIEF DESCRIPTION OF DRAWINGS
[0029] Figure 1 is a flowchart of a sound channel depth calculation method combining multiple marine physical elements provided by an embodiment of the application;
[0030] Figure 2 is a flowchart of another sound channel depth calculation method combining multiple marine physical elements provided by an embodiment of the application;
[0031] Figure 3 is an architecture schematic diagram of a measurement system provided by an embodiment of the application;
[0032] Figure 4 is an architecture schematic diagram of a sound channel depth calculation system combining multiple marine physical elements provided by an embodiment of the application. DETAILED DESCRIPTION
[0033] The application will be further explained in combination with the drawings and specific embodiments:
[0034] As shown in the drawings, Figure 1 a sound channel depth calculation method combining multiple marine physical elements provided by the application comprises:
[0035] S101: historical profile water temperature and profile salinity data of a to-be-measured sea area are acquired, and then historical sound velocity profile data of the to-be-measured sea area are obtained;
[0036] S102: the position of a sound velocity minimum value is calculated by using the sound velocity profile data, and the corresponding sound channel depth is recorded;
[0037] S103: sea surface hydrological data, sea surface meteorological data, marine gravity field data, sea surface dynamic topographic data and seabed topographic data of the to-be-measured sea area in a corresponding period are collected;
[0038] S104: the data collected in S103 are taken as features of a training set together with corresponding spatiotemporal information, the sound channel depth in S102 is taken as a label of the training set, and a multi-dimensional feature regression machine learning model is input for training, the training effect is counted, and the quantitative relationship between the label and the features is listed;
[0039] S105: Real-time collection of sea surface hydrological data and sea surface meteorological data of the to-be-tested sea area, combination of space-time information of the to-be-tested sea area, and ocean gravity field data, seafloor topography data, sea surface dynamic topography data, input into the trained multi-dimensional feature regression machine learning model, and obtaining the required sound channel depth value.
[0040] Further, the sea surface hydrological data includes water temperature, salinity and water flow; the sea surface meteorological data includes air temperature, air pressure and wind speed.
[0041] Further, the water flow and wind speed data contain two components of east and north.
[0042] Further, the ocean gravity field data includes ocean gravity anomaly, ocean vertical deflection, and east gradient, north gradient and radial gradient of ocean gravity anomaly.
[0043] Further, the S103 comprises: combining historical data with a ConvLSTM network to perform 24-hour step prediction of sea surface dynamic topography, and obtaining the prediction value of the sea surface dynamic topography on the same day.
[0044] Further, in the S103, the seafloor topography data is obtained in the following manner: taking the to-be-tested point as the center, expanding 1°x1° or higher range, and counting the terrain standard deviation, slope, gradient and kurtosis in the to-be-tested sea area.
[0045] Further, the S103 further comprises: uniformly gridding the collected data into 0.25°x0.25° or higher spatial resolution grid data.
[0046] Further, before the S105, it further comprises building a measurement system in the following manner:
[0047] A GNSS receiver, a meteorological sensor and a hydrological sensor are provided on a water surface platform, and a set of industrial computer and data acquisition system loaded with trained multi-dimensional feature regression machine learning model and ocean gravity field data, seafloor topography data are carried.
[0048] Further, the meteorological sensor comprises a thermometer, a barometer and a wind speed meter; the hydrological sensor comprises a thermometer, a salinity meter and a water flow meter.
[0049] As shown in Figure 2 Another joint ocean multi-physical element sound channel depth calculation method of the application comprises:
[0050] S201, obtaining historical profile temperature and profile salinity data of the to-be-tested sea area (such as the South China Sea, the range of north latitude 3°58′N to 23°26′N and east longitude 109°38′E to 122°11′E), and calculating historical sound velocity profile data according to the historical profile temperature and profile salinity data.
[0051] The specific formula is:
[0052] C = C (0,35,0) ΔC T + ΔC S + ΔC P + ΔC TSP
[0053] Wherein, C (0,35,0) = 1449.22,
[0054] ΔC T = 4.6233T_5.4585x10 -2 T 2 + 2.822x10 -4 T3_5.07x10 -7 T 4 ,
[0055] ΔC S = 1.391(S_35)_7.8x10 -2 (S_35) 2 ,
[0056] ΔC P = 1.60272x10 -1 P + 3.5216x10 -9 P 3 _3.3603x10 -12 P 4 + 1.0268x10 -5 P 2 ,
[0057] P = 1.11 + 1.02663x10 -1 Z + 2.691x10 -7 Z 2 _4.11x10 -12 Z 3 ,
[0058]
[0059] Wherein C is the sound speed in seawater; C (0,35,0) is the sound speed reference value under reference conditions, i.e. the sound speed when the temperature is 0℃, the salinity is 35 PSU, and the pressure is 0dbar, and the value is 1449.22m / s; ΔC T is the temperature correction term of sound speed; ΔC S is the salinity correction term of sound speed; ΔC P is the pressure correction term of sound speed; ΔC TSPis the correction term of sound speed caused by the joint action of temperature, salinity and pressure; P represents pressure; Z represents water depth; T represents water temperature; S represents salinity.
[0060] S202, using the sound speed profile data obtained in S201, calculating the position of the minimum sound speed and recording the corresponding sound channel depth.
[0061] S203, collecting historical water temperature, salinity, water flow, air temperature, air pressure, wind speed, and marine gravity field, sea surface dynamic topography, seabed topography and other data of the corresponding period. Among them, the water flow and wind speed data need to include east and north components.
[0062] The marine gravity field data need to include marine gravity anomaly, marine vertical deflection, and east gradient, north gradient, and radial gradient of marine gravity anomaly, which can be calculated according to the gravity field model. The specific formula is:
[0063]
[0064] Among them, Δg is the gravity anomaly; γ is the normal gravity; is the normalized earth gravity field potential model coefficient of n order and m degree; is the Legendre function corresponding to the order, which is related to latitude , used to expand the gravity field; ξ, η are the meridian component and the prime vertical component of the vertical deflection respectively; λ, respectively represent longitude and latitude.
[0065] The seabed topography data need to be centered on the point to be measured, and the topography standard deviation, slope, gradient (east and north components), kurtosis and other parameters in the region need to be expanded by 1 ° × 1 ° or higher range.
[0066] For sea surface dynamic topography data, 24-hour step forecast values of the current day need to be obtained in advance according to historical data (past accumulated multi-source spatiotemporal observation or reanalysis data related to marine surface height change). The prediction algorithm can use classic spatiotemporal sequence analysis techniques such as ConvLSTM network.
[0067] Among them, the parameters of ConvLSTM network can be set as follows.
[0068]
[0069]
[0070] The listed data need to be gridded into 0.25 ° × 0.25 ° or higher spatial resolution grid data, and the gridding method can be selected from classic methods such as Kriging.
[0071] S204, input all data in S203 and its corresponding spatiotemporal information as features of the training set, input the sound channel depth in S202 as the label of the training set, input a multi-dimensional feature regression machine learning model such as random forest for training, and count the training effect and list the quantitative relationship between the label and the feature.
[0072] In the random forest model, the hyperparameters can be set as the following values.
[0073]
[0074]
[0075] S205, build a measurement system, as shown in Figure 3 A GNSS receiver (acquire spatiotemporal information), a water temperature meter, a salinity meter, a water flow meter, an air temperature meter, an air pressure meter, a wind speed meter and other sensors are equipped on a water surface platform, and a set of industrial personal computers loaded with the trained random forest model of the application and static data such as marine gravity and seabed topography and a data acquisition system are carried. Specifically, the measurement system is also provided with a solar power supply panel for powering the measurement system.
[0076] S206, place the measurement system built in S205 in the sea area to be measured, so that the measured data of the spatiotemporal information, the water temperature meter, the salinity meter, the water flow meter, the air temperature meter, the air pressure meter, the wind speed meter, and the predicted values of the marine gravity field, the seabed topography, and the sea surface dynamic topography are substituted into the trained random forest model to obtain the required sound channel depth value.
[0077] In summary, the method can quickly calculate the sound channel depth of the point to be measured according to the multiple marine physical elements in the research area. First, by using multiple physical elements, the influence of various factors on the sound channel depth is fully considered, and the reliability of the calculation is improved. Second, using machine learning methods such as ConvLSTM and random forest, deep features between data can be effectively captured, which can further improve the calculation accuracy of the sound channel depth. Moreover, by setting appropriate parameters of the ConvLSTM network and the random forest and the spatiotemporal resolution of the input data, better calculation results can be obtained. Thus, the problem of low efficiency of traditional sound velocity profile measurement can be successfully solved, and the sound channel depth value below the carrier can be measured in real time or even in a walking mode, which provides effective technical support for the research and application in the fields of underwater acoustic positioning, underwater acoustic communication, target detection and other fields.
[0078] Based on the above embodiment, as shown in Figure 4 The application also provides a sound channel depth calculation system combining multiple marine physical elements, comprising:
[0079] A historical sound velocity profile data calculation module is configured to obtain historical profile water temperature and profile salinity data of the sea area to be measured, and further obtain historical sound velocity profile data of the sea area to be measured.
[0080] A sound channel depth calculation module is configured to calculate a position of a sound velocity minimum value by using the sound velocity profile data, and record a corresponding sound channel depth.
[0081] A historical marine multi-physical element collection module is configured to collect sea surface hydrological data, sea surface meteorological data, marine gravity field data, sea surface dynamic topographic data and seabed topographic data of the sea area to be measured in a corresponding period.
[0082] A model training module is configured to input data collected by the historical data collection module and corresponding space-time information as features of a training set, input the sound channel depth in the sound channel depth calculation module as a label of the training set, input a multi-dimensional feature regression machine learning model for training, and count a training effect and list a quantitative relationship between the label and the features.
[0083] A real-time sound channel depth derivation module is configured to input sea surface hydrological data and sea surface meteorological data of the sea area to be measured, space-time information of the sea area to be measured, marine gravity field data, seabed topographic data and sea surface dynamic topographic data into the trained multi-dimensional feature regression machine learning model, and obtain a required sound channel depth value.
[0084] The above only shows the preferred embodiments of the present application, and it should be noted that, for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should also be considered as the protection scope of the present application.
Claims
1. A method for calculating the depth of a sound channel in combination with multiple physical elements of the ocean, characterized in that, The method comprises the following steps: Step 1: Obtain historical profile water temperature and profile salinity data of the sea area to be measured, and then obtain historical sound speed profile data of the sea area to be measured; Step 2: Calculate the position of the minimum sound speed value using the sound speed profile data, and record the corresponding sound channel depth; Step 3: Collect the sea surface hydrological data, sea surface meteorological data, marine gravity field data, sea surface dynamic topographic data and seabed topographic data of the sea area to be measured in the corresponding period; Step 4: Collect the data in step 3, and use the corresponding space-time information as the features of the training set, and use the sound channel depth in step 2 as the label of the training set, and input the multi-dimensional feature regression machine learning model for training, and count the training effect and list the quantitative relationship between the label and the features; Step 5: Real-time acquisition of sea surface hydrological data and sea surface meteorological data of the sea area to be measured, combined with the space-time information of the sea area to be measured, and marine gravity field data, seabed topographic data, sea surface dynamic topographic data, input the trained multi-dimensional feature regression machine learning model, and obtain the required sound channel depth value.
2. The method according to claim 1, wherein, The sea surface hydrological data includes water temperature, salinity and water flow; the sea surface meteorological data includes air temperature, air pressure and wind speed.
3. The method according to claim 2, wherein, The water flow and wind speed data include east and north components.
4. The method according to claim 1, wherein, The marine gravity field data includes marine gravity anomaly, marine vertical deflection, and east gradient, north gradient and radial gradient of marine gravity anomaly.
5. The method according to claim 1, wherein, Step 3 includes: combining historical data with ConvLSTM network to predict 24-hour step of sea surface dynamic topography, and obtaining the prediction value of sea surface dynamic topography on the same day.
6. The method according to claim 1, wherein, In step 3, the seabed topographic data is obtained in the following manner: taking the test point as the center, expanding 1°x1° or higher range, and calculating the terrain standard deviation, slope, gradient and kurtosis in the test sea area.
7. The method according to claim 1, wherein, Step 3 also includes: unifying the collected data into grid data with a spatial resolution of 0.25°x0.25° or higher.
8. The method according to claim 1, wherein, Before step 5, it also includes building a measurement system in the following manner: A water surface platform is equipped with a GNSS receiver, a meteorological sensor and a hydrological sensor, and carries a set of industrial computers and data acquisition systems loaded with trained multi-dimensional feature regression machine learning models and marine gravity field data and seabed topographic data.
9. The method according to claim 1, wherein, The meteorological sensor includes a thermometer, a barometer and a wind speed meter; the hydrological sensor includes a thermometer, a salinity meter and a water flow meter.
10. A system for calculating the depth of a sound channel in combination with multiple physical elements of the ocean, the system comprising: The method comprises the following steps: A historical sound speed profile data calculation module is used to obtain historical profile water temperature and profile salinity data of the sea area to be measured, and then obtain historical sound speed profile data of the sea area to be measured; A sound channel depth calculation module is used to calculate the position of the minimum sound speed value using the sound speed profile data, and record the corresponding sound channel depth; A historical marine multi-physical element collection module is used to collect the sea surface hydrological data, sea surface meteorological data, marine gravity field data, sea surface dynamic topographic data and seabed topographic data of the sea area to be measured in the corresponding period; The model training module is configured to input the data collected by the historical data collection module and corresponding spatiotemporal information as features of a training set, input the sound channel depth in the sound channel depth calculation module as a label of the training set, input the multi-dimensional feature regression machine learning model for training, and count the training effect and list the quantitative relationship between the label and the features. The real-time sound channel depth derivation module is configured to collect real-time sea surface hydrological data and sea surface meteorological data of a to-be-tested sea area, combine spatiotemporal information of the to-be-tested sea area, ocean gravity field data, seabed topography data and sea surface dynamic topography data, input the trained multi-dimensional feature regression machine learning model, and obtain a required sound channel depth value.