WO2024065888A1 - 一种基于定深数据的全海深温度剖面反演方法及系统 - Google Patents

一种基于定深数据的全海深温度剖面反演方法及系统 Download PDF

Info

Publication number
WO2024065888A1
WO2024065888A1 PCT/CN2022/125403 CN2022125403W WO2024065888A1 WO 2024065888 A1 WO2024065888 A1 WO 2024065888A1 CN 2022125403 W CN2022125403 W CN 2022125403W WO 2024065888 A1 WO2024065888 A1 WO 2024065888A1
Authority
WO
WIPO (PCT)
Prior art keywords
depth
temperature
temperature profile
eof
sea
Prior art date
Application number
PCT/CN2022/125403
Other languages
English (en)
French (fr)
Inventor
秦继兴
李倩倩
严娴
顾怡鸣
王海斌
王梦圆
吴双林
Original Assignee
中国科学院声学研究所
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中国科学院声学研究所 filed Critical 中国科学院声学研究所
Publication of WO2024065888A1 publication Critical patent/WO2024065888A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Definitions

  • the invention belongs to the fields of hydroacoustic engineering, ocean engineering and sonar technology, and specifically relates to a full-sea-depth temperature profile inversion method and system based on fixed-depth data.
  • seawater temperature is the factor that has the greatest impact on the speed of sound, so it is crucial to obtain the full-sea depth temperature profile in near real time.
  • Satellite remote sensing can obtain sea surface temperature data for a large area or even the whole world in real time, with high spatial resolution.
  • the regression statistical analysis method can construct an empirical regression model of sea surface temperature anomaly, sea surface height anomaly and full-sea depth temperature profile, and then invert the three-dimensional ocean temperature field, but the accuracy is low near the thermocline.
  • the artificial neural network method can estimate the sound speed profile using sea surface parameters. However, the information obtained from remote sensing satellite data only stays on the surface or near the surface of the ocean.
  • AUVs autonomous underwater vehicles
  • AUVs can be equipped with temperature sensors, and after the buoyancy is basically balanced, the temperature data at a specified depth can be obtained by adjusting the pitch and roll angles.
  • the purpose of the present invention is to overcome the defect that the three-dimensional ocean temperature field obtained by inversion in the prior art has low accuracy near the thermocline.
  • the present invention proposes a full-sea-depth temperature profile inversion method based on fixed-depth data, the method comprising:
  • Step S1 Obtain ocean temperatures at two fixed depths in real time
  • Step S2 input the acquired temperature value into a pre-established and trained neural network model to obtain a p-order EOF coefficient
  • Step S3 Extract the average temperature profile and the first p-order EOF basis functions from the historical hydrological data, and calculate the full-sea depth temperature profile using the p-order EOF coefficients.
  • the neural network model adopts BP neural network, and the neural network training is completed through the historical hydrological data of temperature at a fixed depth and the first p-order EOF coefficients.
  • the training process of BP neural network is divided into two stages: forward propagation and back propagation;
  • the first is forward propagation, that is, inputting two fixed-depth temperature data, passing through the input layer, and reaching the output layer after being processed layer by layer through the hidden layer to obtain the network calculation result;
  • Back propagation is to transfer the loss error back to the input layer through the hidden layer, and evenly distribute the error to all units in each layer, and correct the connection weights of each unit according to the gradient descent method;
  • connection weights are continuously adjusted until the loss error is within the threshold range, completing the model training.
  • step S1 is specifically as follows: according to the depths corresponding to the two extreme points of the p-th order EOF basis function extracted from the historical hydrological data, two AUVs are deployed at the two fixed depths in the designated sea area, and the AUVs are equipped with temperature sensors. After the buoyancy is basically balanced, the yaw angle and pitch angle are adjusted to obtain the temperature data at the two fixed depths in real time.
  • step S3 is specifically as follows:
  • the sampling values of the temperature profile at M time points in the historical water level data are obtained.
  • Each temperature profile has N values at depth after stratification.
  • the M temperature profile samples are expressed in the form of a matrix:
  • t M (z N ) represents the value of the temperature profile at the Nth depth at the Mth time point.
  • the average temperature of the M temperature profiles in each layer is calculated to obtain the average temperature profile
  • T transposing the vector; subtract the average temperature profile from the temperature matrix T Get the temperature perturbation:
  • V [v 1 , ...v N ] ⁇ R N ⁇ N is the eigenvector of the matrix ⁇ T ⁇ T T ⁇ R N ⁇ N , that is, the empirical orthogonal function to be extracted;
  • U [u 1 , ...u M ] ⁇ R M ⁇ M is the eigenvector of the matrix ⁇ T T ⁇ T ⁇ R M ⁇ M ,
  • ⁇ diag([ ⁇ 1 ... ⁇ N ]) ⁇ R M ⁇ N represents the eigenvalue matrix;
  • the eigenvalue corresponding to each eigenvector represents the weight of this eigenvector;
  • the cumulative variance contribution rate of the first m-order mode functions is expressed as:
  • ⁇ k represents the kth eigenvalue
  • the temperature profile is reconstructed using the first p-order EOF basis functions with a contribution rate Em greater than 95%, and the full-sea depth temperature profile matrix is obtained:
  • the temperature value at the i-th depth in the temperature profile is:
  • ⁇ p represents the p-th order EOF coefficient
  • vp represents the p-th order EOF basis function
  • the p is preferably 2.
  • the present invention also provides a full-sea-depth temperature profile inversion system based on fixed-depth temperature data, the system comprising:
  • Temperature acquisition module used to obtain ocean temperature at two fixed depths in real time
  • the EOF coefficient calculation module inputs the acquired temperature value into the pre-established and trained neural network model to obtain the p-order EOF coefficient;
  • the full-sea-depth temperature inversion module is used to extract the average temperature profile and the first p-order EOF basis functions from historical hydrological data, and calculate the full-sea-depth temperature profile using the p-order EOF coefficients.
  • system further comprises:
  • the neural network training module is used to complete the training of the neural network through the historical hydrological data of the temperature at a fixed depth and the first p-order EOF coefficients.
  • the neural network adopts BP neural network, and the training process is divided into two stages: forward propagation and back propagation;
  • the first is forward propagation, that is, inputting two fixed-depth temperature data, passing through the input layer, and reaching the output layer after being processed layer by layer through the hidden layer to obtain the network calculation result;
  • Back propagation is to transfer the loss error back to the input layer through the hidden layer, and evenly distribute the error to all units in each layer, and correct the connection weights of each unit according to the gradient descent method;
  • connection weights are continuously adjusted until the loss error is within the threshold range, completing the model training.
  • the present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the above-described methods when executing the computer program.
  • the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes any of the methods described above.
  • the method of the present invention only needs to use the temperature sensor data on the AUV to obtain limited 1-2 fixed-depth temperature data, and can invert the full-sea-depth temperature profile in real time.
  • this method can realize mobile observation, and has high autonomy, strong load capacity, good maneuverability, and high intelligence. It meets the accuracy requirements of measurement data while meeting the economic requirements.
  • FIG1 shows a flowchart of the neural network model training and inversion method for inverting full-sea-depth temperature profiles based on fixed-depth data
  • Figure 2 shows a schematic diagram of the error of using water temperature data at different depths to invert the full-sea-depth temperature profile
  • Figure 3 shows the cumulative variance contribution rate graph of the first 15 EOFs
  • Figure 4 shows the relationship between the first two order EOF coefficients and the 24°C isotherm and temperature gradient;
  • Figure 4(a) shows the normalized amplitude of the first order EOF coefficient and the 24°C isotherm;
  • Figure 4(b) shows the normalized amplitude of the second order EOF coefficient and the temperature gradient;
  • Figure 5 shows the scatter plot and fitted straight line of the first two order EOF coefficients and the 24°C isotherm and temperature gradient
  • Figure 5(a) shows the scatter plot and fitted straight line of the first order EOF coefficient and the 24°C isotherm
  • Figure 5(b) shows the scatter plot and fitted straight line of the second order EOF coefficient and temperature gradient
  • Figure 6 shows a schematic diagram of the error in inverting the full-sea-depth temperature profile using the water temperature information at depths z 2 and z 3 ;
  • FIG7 shows a schematic diagram of the fixed coordinate system and the moving coordinate system of the AUV
  • Figure 8 shows the topological structure of the BP neural network
  • FIG9 shows a flow chart of the method for inverting the full-sea-depth temperature profile based on fixed-depth data.
  • the present invention uses AUV survey data to extract the temperatures at 1-2 fixed depths, and based on the back propagation neural network, establishes a model for inverting the full-sea depth temperature profile using fixed-depth temperature data, which can solve the problem that traditional observation platforms cannot obtain temperature data in real time and quickly and are costly.
  • the present invention proposes a full-sea-depth temperature profile inversion method based on fixed-depth data, which is particularly suitable for situations where internal waves exist.
  • the first two EOF modes can reconstruct any temperature profile more accurately, and the information at the extreme point depths of the first two EOF basis functions can reflect the thermocline information to the greatest extent.
  • a method for inverting the full-sea-depth temperature profile using limited fixed-depth temperature data is proposed.
  • a mathematical model between the fixed-depth temperature data and the first two-order EOF coefficients is established through a BP neural network, and then the average temperature profile extracted from historical hydrological data and the EOF basis function are combined to achieve the inversion of the temperature profile.
  • the flow chart is shown in Figure 1.
  • Step 1 Analyze the impact of fixed-depth temperature data selection on the temperature profile inversion results.
  • the measured temperature at each depth in the historical hydrological data is used to replace the water temperature value at any depth measured by underwater autonomous submersibles such as AUV, and the effect of inverting the temperature profile of the water temperature at different depths is analyzed.
  • the present invention uses the measurement data of the temperature chain, and selects one by one from shallow to deep, at equal intervals (1 meter). That is, the temperature at a fixed depth (from 10m to 87m, with an interval of 1m) is used to train the BP neural network.
  • the curve with * in Figure 2 plots the normalized average root mean square error of the temperature profile inverted by the water temperature at each depth
  • the solid curve plots the normalized first-order EOF basis function
  • the dotted curve plots the normalized second-order EOF basis function absolute value
  • the horizontal axis scales corresponding to the three dotted lines represent the depths of z1 , z2 , and z3 , respectively, where z1 is the depth corresponding to the extreme point of the first-order EOF basis function, and z2 and z3 are the depths corresponding to the two extreme points of the second-order EOF basis function, respectively.
  • the figure shows that the inversion accuracy changes with depth in a roughly "smiley face" shape, that is, at depths of z 2 or less, the inversion error decreases significantly with water depth; at depths of z 3 or more, the inversion error increases significantly with water depth; and within the depth range of z 2 -z 3 , the inversion error is small and basically remains unchanged.
  • the temperature change at a discrete depth can reflect the overall shape of the full-sea depth temperature profile to a certain extent.
  • the water temperature at different depths contains different amounts of information, among which the water temperature in the depth range of z 2 -z 3 contains the most information.
  • Step 2 Perform EOF decomposition on historical hydrological data and analyze the physical meaning of the first two-order EOF basis functions and EOF coefficients.
  • Empirical orthogonal functions are eigenvectors extracted from a certain number of sample data. Studies have shown that the first few orthogonal functions can effectively reconstruct the temperature profile.
  • the temperature profile is sampled at M time points. Each temperature profile has N values at depth after stratification.
  • the M temperature profile samples are represented in the form of a matrix:
  • ti ( zj ) represents the value of the temperature profile at the jth depth at the ith sampling point.
  • the average temperature of the M temperature profiles in each layer is calculated to obtain the average temperature profile
  • T represents the transposition of the vector. Subtract the average temperature profile from the temperature matrix T Get the temperature perturbation:
  • X U ⁇ ⁇ RM ⁇ N is the coefficient matrix.
  • the eigenvalue corresponding to each eigenvector represents the weight of this eigenvector.
  • the cumulative variance contribution rate of the first m-order mode functions can be expressed as:
  • the contribution rate of the first-order mode is the largest, and the higher the order, the smaller the contribution rate of the modal function.
  • the temperature profile is reconstructed using the first p-order EOF basis functions with a contribution rate Em greater than 95%.
  • the reconstructed temperature profile matrix is expressed as:
  • the historical hydrological data were decomposed by EOF, and the cumulative variance contribution rate of the first 15-order EOF basis functions was calculated. As shown in Figure 3, it was found that the cumulative variance contribution rate of the first two-order EOF modes in the South China Sea can reach more than 95%.
  • the temperature profile is reconstructed using the first two EOF basis functions, and the i-th temperature profile can be approximately expressed as:
  • ⁇ i represents the i-th order EOF coefficient
  • vi represents the i-th order EOF basis function
  • t mean represents the average ocean temperature at the i-th sampling time of the historical water level data.
  • the first-order EOF coefficient represents the vertical displacement of the thermocline. The larger ⁇ 1 is, the shallower the thermocline is.
  • the second-order EOF coefficient represents the change in temperature gradient. The larger ⁇ 2 is, the more drastic the change in the thermocline is. This is verified below.
  • Figure 4(a) plots the normalized 24°C isotherm and the normalized first-order EOF coefficient of the temperature profile in the training set. It can be seen from the figure that the change trend of the isotherm and the first-order projection coefficient is highly correlated, with a correlation coefficient of about 0.98.
  • Figure 5(a) plots the scatter plot and fitting line of the 24°C isotherm corresponding to the depth and the first-order EOF coefficient.
  • z 24°C is the depth corresponding to the 24°C isotherm.
  • the first-order EOF coefficient is similar to the changing trend of the seawater layer temperature, so it can also be interpreted that the first-order EOF coefficient can approximately reflect the changing trend of the seawater layer.
  • the temperature gradient of the training set temperature profile in this depth range is calculated.
  • Figure 4(b) plots the normalized temperature gradient and the normalized second-order EOF coefficient. It can be seen from the figure that the temperature gradient is highly correlated with the changing trend of the second-order projection coefficient, and the correlation coefficient reaches 0.90.
  • Figure 5(b) plots the scatter plot and fitting line of the temperature gradient and the second-order EOF coefficient:
  • thermocline represent the upper and lower boundaries of the thermocline, respectively. Therefore, the water temperature data within the depth range of z 2 -z 3 can reflect the thermocline information to the greatest extent, thus achieving higher accuracy when inverting the temperature profile. This also explains the conclusion drawn in step 1.
  • the above analysis provides a reference for the selection of the measurement depth of fixed depth data.
  • Step 3 Use the water temperature information at depths z 2 and z 3 to replace the fixed-depth temperature data obtained by the AUV to invert the full-sea-depth temperature profile.
  • the inversion accuracy of the temperature profile in the training set will increase accordingly.
  • the results in Figure 2 show that the inversion accuracy depends on the measurement depth of the fixed-depth data, while the temperature profile mainly depends on the characteristics of the thermocline.
  • Step 2 proves that z 2 and z 3 represent the depths of the upper and lower boundaries of the thermocline, respectively.
  • the water temperatures at these two depths can basically determine the basic structure of the temperature profile.
  • the water temperature information at these two depths is used instead of the fixed-depth temperature data obtained by the AUV to invert the full-sea depth temperature profile.
  • the inversion results are shown in Figure 6.
  • the root mean square error of the temperature profile inversion is basically below 0.2°C, and the mean root mean square error is 0.1137.
  • Step 4 Combined with the analysis of historical hydrological data, two AUVs were used to measure the fixed-depth temperature data of the upper and lower boundaries of the thermocline.
  • depth navigation is a common navigation state.
  • Two AUVs need to be equipped with temperature sensors to work together and obtain temperature data during depth navigation in a designated sea area. After the buoyancy is basically balanced, when the AUV has a certain sway speed, depth navigation is achieved by adjusting the yaw angle and pitch angle of the AUV.
  • a fixed coordinate system needs to be established to describe the change in the position of each AUV.
  • the fixed coordinate system E- ⁇ is selected with the earth as the reference system.
  • the origin E of the coordinate system can be taken at any point on the ground or sea surface.
  • the E ⁇ axis remains horizontal, and the main heading of the AUV is often taken as the positive direction of E ⁇ ; the E ⁇ and E ⁇ axes are perpendicular to each other and in the horizontal plane. The direction can be selected at will.
  • the E ⁇ axis is perpendicular to the E ⁇ coordinate plane, and its positive direction points to the center of the earth.
  • a motion coordinate system needs to be established. The motion coordinate system will shift with the movement of the AUV. In the motion coordinate system, the Ox axis is taken on the straight line formed by the position points of the two measuring AUVs.
  • the Oy axis is perpendicular to the Ox axis, and the Oz axis is in the longitudinal midplane, pointing to the bottom of the ship and perpendicular to the waterline, as shown in Figure 7.
  • the relevant experimental data extracted in the fixed coordinate system more accurately and intuitively reflects the motion process of the underwater AUV.
  • Step 5 Using the two fixed-depth temperature data obtained by AUV measurement, the full-sea-depth temperature profile is inverted in real time.
  • a network model between the temperature at depths z2 and z3 and the first two-order EOF coefficients was established using the BP neural network based on historical hydrological data, as shown in Figure 8. After the AUV measured the water temperatures at these two depths, they were input into the trained network model, and combined with the historical average temperature profile and the first two-order EOF basis functions, the full-sea depth temperature profile was obtained by real-time inversion.
  • the training process of BP neural network is divided into two stages: forward propagation and back propagation.
  • the first is forward propagation, that is, the input data (2 fixed-depth temperature data) is passed through the input layer, processed layer by layer through the hidden layer, and reaches the output layer, and the network calculation result is obtained.
  • the loss function loss
  • the error between the network calculation result and the actual output can be calculated, where y out is the network output and y is the actual output.
  • y out is the network output and y is the actual output.
  • the loss error exceeds the set threshold, it enters the back propagation stage.
  • Back propagation is to pass the loss error back to the input layer through the hidden layer, and evenly distribute the error to all units in each layer, and correct the connection weights of each unit according to the gradient descent method.
  • the forward and backward propagation cycles are carried out so that the connection weights can be continuously adjusted until the loss error is within the threshold range, indicating that the model training is completed.
  • the present invention proposes a full-sea-depth temperature profile inversion method based on fixed-depth data (as shown in FIG9 ), the method comprising:
  • Step S1 Obtain ocean temperatures at two fixed depths in real time
  • two AUVs were deployed at the two fixed depths in the designated sea area.
  • the AUVs were equipped with temperature sensors. After the buoyancy was basically balanced, the yaw angle and pitch angle were adjusted to obtain the temperature data at the two fixed depths in real time.
  • Step S2 input the acquired temperature value into a pre-established and trained neural network model to obtain a p-order EOF coefficient
  • Step S3 Extract the average temperature profile and the first p-order EOF basis functions from the historical hydrological data, and use the p-order EOF coefficient to calculate the full-sea depth temperature profile:
  • the sample values of the temperature profile at M time points in the historical water level data are obtained.
  • Each temperature profile has N values at depth after stratification.
  • the M temperature profile samples are expressed in the form of a matrix:
  • t M (z N ) represents the value of the temperature profile at the Nth depth at the Mth time point.
  • the average temperature of the M temperature profiles in each layer is calculated to obtain the average temperature profile
  • T transposing the vector; subtract the average temperature profile from the temperature matrix T Get the temperature perturbation:
  • V [v 1 , ⁇ v N ] ⁇ RN ⁇ N is the eigenvector of the matrix ⁇ T ⁇ T T ⁇ RN ⁇ N , that is, the empirical orthogonal function to be extracted;
  • the cumulative variance contribution rate of the first m-order mode functions is expressed as:
  • ⁇ k represents the kth eigenvalue
  • the temperature profile is reconstructed using the first p-order EOF basis functions whose contribution rate Em is greater than the set value, and the full-sea-depth temperature profile matrix is obtained:
  • the temperature value at the i-th depth in the temperature profile is:
  • ⁇ p represents the p-th order EOF coefficient
  • vp represents the p-th order EOF basis function
  • It represents the average temperature of the ith depth in the historical water level data.
  • the preferred value of p is 2.
  • the neural network model adopts BP neural network, and the training of the neural network is completed through the historical hydrological data of the temperature at a fixed depth and the first p-order EOF coefficient.
  • the present invention also provides a full-sea-depth temperature profile inversion system based on fixed-depth temperature data, the system comprising:
  • Temperature acquisition module used to obtain ocean temperature at two fixed depths in real time
  • the EOF coefficient calculation module inputs the acquired temperature value into the pre-established and trained neural network model to obtain the p-order EOF coefficient;
  • the full-sea-depth temperature inversion module is used to extract the average temperature profile and the first p-order EOF basis functions from historical hydrological data, and calculate the full-sea-depth temperature profile using the p-order EOF coefficients.
  • the system further comprises:
  • the neural network training module is used to complete the training of the neural network through the historical hydrological data of the temperature at a fixed depth and the first p-order EOF coefficients.
  • the present invention may also provide a computer device, comprising: at least one processor, a memory, at least one network interface and a user interface.
  • the various components in the device are coupled together through a bus system. It is understood that the bus system is used to achieve connection and communication between these components.
  • the bus system also includes a power bus, a control bus and a status signal bus.
  • the user interface may include a display, a keyboard, or a pointing device (e.g., a mouse, a track ball, a touch pad, or a touch screen).
  • a pointing device e.g., a mouse, a track ball, a touch pad, or a touch screen.
  • the memory in the disclosed embodiments of the present application can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories.
  • the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory.
  • the volatile memory can be a random access memory (RAM), which is used as an external cache.
  • RAM Static RAM
  • DRAM Dynamic RAM
  • SDRAM Synchronous DRAM
  • DDRSDRAM Double Data Rate SDRAM
  • ESDRAM Enhanced SDRAM
  • SLDRAM Synchlink DRAM
  • DRRAM Direct Rambus RAM
  • the memory stores the following elements, executable modules or data structures, or a subset thereof, or an extended set thereof: an operating system and applications.
  • the operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process hardware-based tasks.
  • the application includes various application programs, such as a media player (Media Player), a browser (Browser), etc., which are used to implement various application services.
  • the program for implementing the method of the embodiment of the present disclosure can be included in the application.
  • the processor may also call a program or instruction stored in the memory, specifically, a program or instruction stored in an application program, to:
  • the above method can be applied to a processor or implemented by a processor.
  • the processor may be an integrated circuit chip with signal processing capabilities.
  • each step of the above method can be completed by an integrated logic circuit of hardware in the processor or an instruction in the form of software.
  • the above processor can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.
  • DSP Digital Signal Processor
  • ASIC Application Specific Integrated Circuit
  • FPGA Field Programmable Gate Array
  • the above disclosed methods, steps and logic block diagrams can be implemented or executed.
  • the general processor can be a microprocessor or the processor can also be any conventional processor, etc.
  • the steps of the above disclosed method can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor.
  • the software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc.
  • the storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.
  • the embodiments described in the present invention can be implemented by hardware, software, firmware, middleware, microcode or a combination thereof.
  • the processing unit can be implemented in one or more application specific integrated circuits (Application Specific Integrated Circuits, ASIC), digital signal processors (Digital Signal Processing, DSP), digital signal processing devices (DSP Device, DSPD), programmable logic devices (Programmable Logic Device, PLD), field programmable gate arrays (Field-Programmable Gate Array, FPGA), general processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in the present application or a combination thereof.
  • ASIC Application Specific Integrated Circuits
  • DSP Digital Signal Processing
  • DSP Device digital signal processing devices
  • PLD programmable logic devices
  • FPGA field programmable gate array
  • general processors controllers, microcontrollers, microprocessors, other electronic units for performing the functions described in the present application or a combination thereof.
  • the technology of the present invention can be implemented by executing the functional modules (such as procedures, functions, etc.) of the present invention.
  • the software code can be stored in a memory and executed by a processor.
  • the memory can be implemented in the processor or outside the processor.
  • the present invention may also provide a non-volatile storage medium for storing a computer program.
  • a computer program When the computer program is executed by a processor, each step in the above method embodiment can be implemented.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)

Abstract

一种基于定深数据的全海深温度剖面反演方法及系统，该方法包括：实时获取两个固定深度处的海洋温度；将获取的温度值输入预先建立和训练好的神经网络模型，得到p阶EOF系数；从历史水文数据中提取出平均温度剖面与前p阶EOF基函数，利用p阶EOF系数计算全海深温度剖面。该方法无需布放深度垂直温度链，系统复杂度低，易于布放和操作，可在较大区域内应用；利用AUV获取深海温度，具有较好的机动性，可根据任务需要进行布放，实现大区域内三维温度场建模。

Description

一种基于定深数据的全海深温度剖面反演方法及系统
相关申请
本申请要求名称为“一种基于定深数据的全海深温度剖面反演方法及系统”、于2022年9月29日提交的中国专利申请号为202211200508.7的优先权，在此通过引用包括该件申请。
技术领域
本发明属于水声工程、海洋工程、声呐技术领域，具体涉及一种基于定深数据的全海深温度剖面反演方法及系统。
背景技术
海水中最重要的声学参数是声波传播速度，海水声速剖面的垂直分布是影响水下声场特性的主要因素之一，它决定了声波在海洋中折射和传播路径。海水温度是影响声速最大的因素，因此，近实时地获取全海深温度剖面至关重要。
一般而言，温度数据往往要通过CTD、XBT或锚碇温度链等进行现场采样，但是该方法费时费力且无法同步获取大面积的数据，因此科学家更倾向于采用反演法。卫星遥感手段可以实时获取大面积乃至全球的海表温度资料，具有较高的空间分辨率。回归统计分析的方法可以构建海面温度异常、海面高度异常与全海深温度剖面的经验回归模型，进而反演得到三维海洋温度场，但在温跃层附近精度较低。使用人工神经网络的方法利用海表面参数可以估计声速剖面，然而通过遥感卫星资料得到的信息仅仅停留在海洋表层或者近表层，利用表层信息反演得到的也仅是温度剖面的平均稳态场，反演精度较低，参见参考文献[1](“Estimation of Sound Speed Profiles Using Artificial Neural Networks”，2006年发表在《IEEE Geoscience&Remote Sensing Letters》第3期，起始页码为467)。海洋声层析也是一种常见的反演方法，可以利用声学方法来监测海洋的中尺度过程，然而发射-接收器的高功耗一直是主动声层析的制约因素。研究人员利用历史的水文资料结合有限深度范围(20m-40m)直接测量的温度重构了全海深范围内的声速剖面，然而并未研究对于重构出全海深声速剖面时需要用到实测数据的最小深度范围等问题，参见参考文献[2](“利用 有限深度声速数据重构全海深声速剖面”，2008年发表在《声学技术》第5期，起始页码为106)。
随着科技水平的不断发展，水下自治机器人(Autonomous Underwater Vehicle,AUV)越来越多地被应用于海洋特征观测，AUV的快速发展使得直接测量任一深度处的水文参数成为可能。AUV上可搭载温度传感器，在浮力基本平衡后可以通过调节首摇角和纵摇角来获取指定深度处的温度数据。
发明内容
本发明的目的在于克服现有技术反演得到的三维海洋温度场在温跃层附近精度较低的缺陷。
为了实现上述目的，本发明提出了一种基于定深数据的全海深温度剖面反演方法，所述方法包括：
步骤S1：实时获取两个固定深度处的海洋温度；
步骤S2：将获取的温度值输入预先建立和训练好的神经网络模型，得到p阶EOF系数；
步骤S3：从历史水文数据中提取出平均温度剖面与前p阶EOF基函数，利用p阶EOF系数计算全海深温度剖面。
作为上述方法的一种改进，所述神经网络模型采用BP神经网络，通过固定深度处的温度与前p阶EOF系数的历史水文数据完成神经网络的训练。
BP神经网络的训练过程分为正向传播和反向传播两个阶段；
首先是正向传播，即输入2个定深温度数据，经输入层传入，通过隐含层逐层处理后到达输出层，得到网络计算结果；根据损失函数loss＝|y out-y|计算出网络计算结果与实际输出之间的误差，其中y out为网络输出，y为实际输出；若损失误差超过设定阈值，则进入到反向传播阶段；
反向传播是将损失误差通过隐含层反向传递输入层，并将误差平均分配到各层的所有单元，根据梯度下降法修正各个单元的连接权值；
正向、反向传播循环进行，使得连接权值不断调整，直至损失误差在阈值范围内，完成模型训练。
作为上述方法的一种改进，所述步骤S1具体为：依据从历史水文数据中提取的第p阶EOF基函数两个极值点处对应的深度，在指定海域这两个固定深度处布设两个AUV，AUV上搭载温度传感器，在浮力基本平衡后通过调节首摇角和纵摇角来 实时获取两个固定深度处的温度数据。
作为上述方法的一种改进，所述步骤S3具体为：
获得历史水位数据中温度剖面在M个时间点的采样值，每个温度剖面经过层化处理在深度上具有N个值，将M个温度剖面样本表示成矩阵的形式：
Figure PCTCN2022125403-appb-000001
其中，t M(z N)表示温度剖面在第M个时间点第N个深度上的取值，计算M个温度剖面在每一层的平均温度，得到平均温度剖面
Figure PCTCN2022125403-appb-000002
Figure PCTCN2022125403-appb-000003
其中，[] T表示对向量进行转置；将温度矩阵T减去平均温度剖面
Figure PCTCN2022125403-appb-000004
得到温度扰动：
Figure PCTCN2022125403-appb-000005
将ΔT进行奇异值分解，得到：
ΔT T＝U∑V T
其中，V＝[v 1,…v N]∈R N×N是矩阵ΔTΔT T∈R N×N的特征向量，即为要提取的经验正交函数；U＝[u 1,…u M]∈R M×M是矩阵ΔT TΔT∈R M×M的特征向量，∑＝diag([λ 1…λ N])∈R M×N表示特征值矩阵；每一个特征向量对应的特征值表示此特征向量的权重；前m阶模态函数的累积方差贡献率表示为：
Figure PCTCN2022125403-appb-000006
其中，
Figure PCTCN2022125403-appb-000007
λ k表示第k个特征值；
利用贡献率E m大于95％的前p阶EOF基函数重构温度剖面，得到全海深温度剖面矩阵：
Figure PCTCN2022125403-appb-000008
其中，[α 1M α 2M…α pM] T表示系数矩阵U∑＝[α 1…α N]∈R M×N的前p阶；
Figure PCTCN2022125403-appb-000009
表示矩阵V＝[v 1,…,v N]∈R N×N的前p阶；
其中，温度剖面中第i个深度温度值为：
Figure PCTCN2022125403-appb-000010
其中，i∈N；α p表示第p阶EOF系数，v p表示第p阶EOF基函数，
Figure PCTCN2022125403-appb-000011
表示历史水位数据第i个深度平均温度。
作为上述方法的一种改进，所述p优选2。
本发明还提供一种基于定深温度数据的全海深温度剖面反演系统，所述系统包括：
温度采集模块，用于实时获取两个固定深度处的海洋温度；
计算EOF系数模块，将获取的温度值输入预先建立和训练好的神经网络模型，得到p阶EOF系数；
反演全海深温度模块，用于从历史水文数据中提取出平均温度剖面与前p阶EOF基函数，利用p阶EOF系数计算全海深温度剖面。
作为上述系统的一种改进，所述系统还包括：
神经网络训练模块，用于通过固定深度处的温度与前p阶EOF系数的历史水文数据完成神经网络的训练。
神经网络采用BP神经网络，训练过程分为正向传播和反向传播两个阶段；
首先是正向传播，即输入2个定深温度数据，经输入层传入，通过隐含层逐层处理后到达输出层，得到网络计算结果；根据损失函数loss＝|y out-y|计算出网络计算结果与实际输出之间的误差，其中y out为网络输出，y为实际输出；若损失误差超过设定阈值，则进入到反向传播阶段；
反向传播是将损失误差通过隐含层反向传递输入层，并将误差平均分配到各层的所有单元，根据梯度下降法修正各个单元的连接权值；
正向、反向传播循环进行，使得连接权值不断调整，直至损失误差在阈值范围内，完成模型训练。
本发明还提供一种计算机设备，包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的计算机程序，所述处理器执行所述计算机程序时实现如上述任一项所述的方法。
本发明还提供一种计算机可读存储介质，所述计算机可读存储介质存储有计算机程序，所述计算机程序当被处理器执行时使所述处理器执行如上述任一项所述的方法。
与现有技术相比，本发明的优势在于：
本发明的方法仅需要利用AUV搭载温度传感器数据获取有限的1-2个定深温度数据，便可以实时反演全海深温度剖面。与定点观测、巡航观测以及锚碇浮标阵观测等传统观测方式相比，该方法可以实现移动观测，且自主性高、负载能力强、机动性好、智能化程度高，在满足经济性的同时也符合测量数据的精度要求。
附图说明
图1所示为基于定深数据反演全海深温度剖面方法的神经网络模型训练及反演流程图；
图2所示为不同深度的水温数据用于反演全海深温度剖面的误差示意图；
图3所示为前15阶EOF的累积方差贡献率图；
图4所示为前两阶EOF系数与24℃等温线及温度梯度的关系图；图4(a)表示第一阶EOF系数与24℃等温线的归一化幅值；图4(b)表示第二阶EOF系数与温度梯度的归一化幅值；
图5所示为前两阶EOF系数与24℃等温线及温度梯度的散点图及拟合直线图；图5(a)表示第一阶EOF系数与24℃等温线的散点图及拟合直线；图5(b)表示第二阶EOF系数与温度梯度的散点图及拟合直线；
图6所示为以z 2、z 3深度处的水温信息反演全海深温度剖面的误差示意图；
图7所示为AUV固定坐标系与运动坐标系示意图；
图8所示为BP神经网络拓扑结构；
图9所示为基于定深数据反演全海深温度剖面方法的流程图。
具体实施方式
本发明利用AUV调查数据提取1-2个固定深度处的温度，基于反向传播神经网络，建立一个利用定深温度数据反演全海深温度剖面的模型，能够解决传统观测平台无法实时、快速获取温度数据且成本昂贵的问题。
下面结合附图对本发明的技术方案进行详细的说明。
本发明提出一种基于定深数据的全海深温度剖面反演方法，尤其适用于内波存在的情形中。已有研究表明在内波环境下，前两阶EOF模态就可以较为精确地重构任意温度剖面，并且前两阶EOF基函数极值点深度处的信息能够最大程度地反映温跃层信息。在此基础上提出一种利用有限的定深温度数据反演全海深温度剖面的方法。首先通过BP神经网络建立定深温度数据与前两阶EOF系数之间的数学模型，然后结合从历史水文资料中提取的平均温度剖面和EOF基函数实现温度剖面的反演，流程图如图1所示。
步骤1：分析定深温度数据选取对温度剖面反演结果的影响。
利用历史水文数据中各深度处的实测温度代替AUV等水下自主潜器测量的任一深度处的水温值，分析不同深度的水温反演温度剖面的效果。为了确定该固定深度的最优选取标准，本发明利用温度链的测量数据，从浅至深，等间隔(1米)逐一选取。即利用固定深度(从10m至87m，以1m为间隔)处的温度训练BP神经网络。图2带*曲线绘制了以各个深度水温反演温度剖面的归一化平均均方根误差，实线曲线绘制了归一化的第一阶EOF基函数，带点曲线绘制了归一化的第二阶EOF基函数绝对值，三条虚线对应的横轴刻度分别表示z 1、z 2、z 3深度，其中z 1为第一阶EOF基函数极值点处对应的深度，z 2和z 3分别为第二阶EOF基函数两个极值点处对应的深度。图中可见反演精度随深度变化大致呈现“笑脸型”，即在z 2以浅，反演误差随水深大幅度下降；在z 3以深，反演误差随水深大幅度上升；而在z 2-z 3深度范围内，反演误差偏小且基本保持不变。由图可知，一个离散深度处的温度变化可以在一定程度上反映全海深温度剖面的整体形态，然而不同深度处的水温所含的信息量不同，其中z 2-z 3深度范围的水温所含信息量最多。
步骤2：对历史水文数据进行EOF分解，分析前两阶EOF基函数及EOF系数的物理意义。
如果直接对海水温度剖面进行估计，则待估计的变量太多，运算量将十分大，对算法的性能要求会很高，因此采用经验正交函数来表示海水温度剖面。 经验正交函数是从一定数量的样本数据中提取出的特征向量，研究表明，前几阶正交函数就能有效地重构温度剖面。对温度剖面在M个时间点采样，每个温度剖面经过层化处理在深度上具有N个值，将M个温度剖面样本表示成矩阵的形式：
Figure PCTCN2022125403-appb-000012
其中，t i(z j)表示的是温度剖面在第i个采样点第j个深度上的取值，计算M个温度剖面在每一层的平均温度，得到平均温度剖面
Figure PCTCN2022125403-appb-000013
Figure PCTCN2022125403-appb-000014
其中符号T表示对向量进行转置。将温度矩阵T减去平均温度剖面
Figure PCTCN2022125403-appb-000015
得到温度扰动：
Figure PCTCN2022125403-appb-000016
将ΔT进行奇异值分解，可得到：
ΔT T＝U∑V T     (4)
其中，其中，V＝[v 1,…v N]∈R N×N是矩阵A＝ΔTΔT T∈R N×N的特征向量，即为要提取的经验正交函数。U＝[u 1,…u M]∈R M×M是矩阵B＝ΔT TΔT∈R M×M的特征向量且∑＝diag([λ 1…λ N])∈R M×N为特征值矩阵。X＝U∑∈R M×N是系数矩阵。
Figure PCTCN2022125403-appb-000017
且Ω＝[λ 1,…λ N]为矩阵A的特征值。每一个特征向量对应的特征值表示此特征向量的权重，特征值越小，其对应的特征向量(经验正交函数)包含更少的信息。前m阶模态函数的累积方差贡献率可表示为：
Figure PCTCN2022125403-appb-000018
其中，第一阶模态的贡献率最大，且阶数越高，模态函数的贡献率越小。
利用贡献率E m大于95％的前p阶EOF基函数重构温度剖面，重构的温度剖面矩阵表示为：
Figure PCTCN2022125403-appb-000019
其中，[α 1M α 2M…α pM] T是系数矩阵U∑＝[α 1…α N]∈R M×N的前p阶；
Figure PCTCN2022125403-appb-000020
表示矩阵V＝[v 1,…,v N]∈R N×N的前p阶。
对历史水文数据进行EOF分解，计算前15阶EOF基函数的累积方差贡献率，如图3所示，发现在南海海域前两阶EOF模态的累积方差贡献率就可达到95％以上。
因此利用前两阶EOF基函数重构温度剖面，则第i条温度剖面可近似表示为：
Figure PCTCN2022125403-appb-000021
其中，α i表示第i阶EOF系数，v i表示第i阶EOF基函数，t mean表示历史水位数据第i个采样时间平均海洋温度。
结合步骤1得出的结论，对前两阶EOF基函数及EOF系数的物理意义进行分析。已有研究给出了EOF系数的物理解释，第一阶EOF系数表示温跃层的垂直位移，α 1越大，温跃层越浅。而第二阶EOF系数则代表温度梯度的变化，且α 2越大，温跃层变化越剧烈，以下对此进行了验证。由图2可知，第一阶EOF基函数的极值点对应的深度为z 1＝55m，此深度处，平均温度为
Figure PCTCN2022125403-appb-000022
图4(a)绘制了训练集中温度剖面的归一化24℃等温线与归一化第一阶EOF系数，由图可知，等温线与第一阶投影系数的变化趋势高度相关，相关系数约为0.98。图5(a)绘制了24℃等温线对应深度与第一阶EOF系数的散点图及拟合直线
Figure PCTCN2022125403-appb-000023
z 24℃为24℃等温线对应的深度。第一阶EOF系数与海水水层温度的变化趋势相近，因此还可以解释为，第一阶EOF系数可以近似反映出海水水层的变化趋势。由图2可知，第二阶EOF基函数的两个极值点对应的深度分别为z 2＝49m,z 3＝62m，计算训练集温度剖面在该深度范围中的温度梯度。图4(b)绘制了归一化温度梯度与归一化第二阶EOF系数，由图可知，温度梯度与第二阶投影系数的变化趋势高度相关，相关系数达到0.90，图5(b)绘制了温度梯度 与第二阶EOF系数的散点图及拟合直线：
Figure PCTCN2022125403-appb-000024
以上结论与已有研究一致，z 2和z 3分别表示温跃层的上下界，因此z 2-z 3深度范围内的水温数据能够最大程度地反映温跃层信息，从而反演温度剖面时精度更高。这也解释了步骤1得出的结论。以上分析为定深数据的测量深度选取提供了参考依据。
步骤3：以z 2、z 3深度处的水温信息代替AUV获取的定深温度数据反演全海深温度剖面。
可想而知，随着输入层信息的增加，训练集中温度剖面的反演精度会随之提升，图2结果表明，反演精度取决于定深数据的测量深度，而温度剖面主要取决于温跃层的特征，步骤2证明了z 2、z 3分别表示温跃层上下界的深度，这两个深度处的水温基本可以确定温度剖面的基本结构，以这两个深度处的水温信息代替AUV获取的定深温度数据反演全海深温度剖面，反演结果如图6所示，温度剖面反演的均方根误差基本在0.2℃以下，均方根误差的均值为0.1137
℃。
步骤4：结合对历史水文数据进行的分析，利用两台AUV测得温跃层上下界的定深温度数据。
对于自治水下机器人，定深航行是一种常见的航行状态。两台AUV需搭载温度传感器协同作业，在指定海域定深航行获取温度数据。定深航行在浮力基本平衡后，当AUV具有一定纵荡速度时，通过调节AUV的首摇角和纵摇角来实现。在多AUV系统中，需要建立固定坐标系来描述每个AUV位置的变化。一般以大地作为参考系选择固定坐标系E-ξηζ，坐标系原点E可取地面或海面上任何一点，Eξ轴保持水平，常以AUV的主航向为Eξ的正向；Eη和Eξ轴互相垂直且在水平面内，方向可以任选，Eζ轴垂直于Eζη坐标面，其正向指向地心。为方便描述两个运动AUV之间的相互距离关系，还需要建立运动坐标系，运动坐标系会随AUV运动而发生位移。在运动坐标系中，把Ox轴取在两个测量AUV的位置点所构成的直线上。Oy轴与Ox轴垂直，Oz轴在纵中平面内，指向船底方向，与水线面垂直，如图7所示。为了清晰地观察水下AUV的真实位置和航行规律，需要将运动坐标系转换为固定坐标系，在固定坐标系下提取的相关实验数据更加准确直观地反映了水下AUV的运动过程。假设运动坐标系的原 点在固定坐标系下的坐标值为(O x,O y,O z)，坐标轴旋转角度为
Figure PCTCN2022125403-appb-000025
AUV在运动坐标系下的坐标为(X 1,Y 1,Z 1)，在固定坐标系下的坐标为(X 2,Y 2,Z 2)，则运动坐标系中的任意一点在固定坐标系中可表示为：
Figure PCTCN2022125403-appb-000026
步骤5：通过AUV测量得到的两个定深温度数据，实时反演全海深温度剖面。
以历史水文数据通过BP神经网络建立z 2、z 3深度处的温度与前两阶EOF系数之间的网络模型，如图8所示，AUV实测获取这两个深度的水温后，输入训练好的网络模型中，并结合历史平均温度剖面及前两阶EOF基函数，实时反演得到全海深温度剖面。
BP神经网络的训练过程分为正向传播和反向传播两个阶段。首先是正向传播，即输入数据(2个定深温度数据)经输入层传入，通过隐含层逐层处理后到达输出层，并得到网络计算结果。根据损失函数loss＝|y out-y|可以计算出网络计算结果与实际输出之间的误差，其中y out为网络输出，y为实际输出。若损失误差超过设定阈值，则进入到反向传播阶段。反向传播是将损失误差通过隐含层反向传递输入层，并将误差平均分配到各层的所有单元，根据梯度下降法修正各个单元的连接权值。正向、反向传播循环进行，使得连接权值能够不断调整，直至损失误差在阈值范围内，表明模型训练完成。
根据以上试验过程，本发明提出一种基于定深数据的全海深温度剖面反演方法(如图9所示)，所述方法包括：
步骤S1：实时获取两个固定深度处的海洋温度；
依据从历史水文数据中提取的第p阶EOF基函数两个极值点处对应的深度，在指定海域这两个固定深度处布设两个AUV，AUV上搭载温度传感器，在浮力基本平衡后通过调节首摇角和纵摇角来实时获取两个固定深度处的温度数据。
步骤S2：将获取的温度值输入预先建立和训练好的神经网络模型，得到p阶EOF系数；
步骤S3：从历史水文数据中提取出平均温度剖面与前p阶EOF基函数，利用p阶EOF系数计算全海深温度剖面：
获得历史水位数据中温度剖面在M个时间点的采样值，每个温度剖面经过层化 处理在深度上具有N个值，将M个温度剖面样本表示成矩阵的形式：
Figure PCTCN2022125403-appb-000027
其中，t M(z N)表示温度剖面在第M个时间点第N个深度上的取值，计算M个温度剖面在每一层的平均温度，得到平均温度剖面
Figure PCTCN2022125403-appb-000028
Figure PCTCN2022125403-appb-000029
其中，[] T表示对向量进行转置；将温度矩阵T减去平均温度剖面
Figure PCTCN2022125403-appb-000030
得到温度扰动：
Figure PCTCN2022125403-appb-000031
将ΔT进行奇异值分解，得到：
ΔT T＝U∑V T
其中，V＝[v 1,Δv N]∈R N×N是矩阵ΔTΔT T∈R N×N的特征向量，即为要提取的经验正交函数；U＝[u 1,Δu M]∈R M×M是矩阵ΔT TΔT∈R M×M的特征向量，∑＝diag(λ 1…λ N])∈R M×N表示特征值矩阵；每一个特征向量对应的特征值表示此特征向量的权重；前m阶模态函数的累积方差贡献率表示为：
Figure PCTCN2022125403-appb-000032
其中，
Figure PCTCN2022125403-appb-000033
λ k表示第k个特征值；
利用贡献率E m大于设定值的前p阶EOF基函数重构温度剖面，得到全海深温度剖面矩阵：
Figure PCTCN2022125403-appb-000034
其中，[α 1M α 2M…α pM] T表示系数矩阵U∑＝[α 1…α N]∈R M×N的前p 阶；
Figure PCTCN2022125403-appb-000035
表示矩阵V＝[v 1,…,v N]∈R N×N的前p阶；
其中，温度剖面中第i个深度温度值为：
Figure PCTCN2022125403-appb-000036
其中，i∈N；α p表示第p阶EOF系数，v p表示第p阶EOF基函数，
Figure PCTCN2022125403-appb-000037
表示历史水位数据第i个深度平均温度。p优选2。
所述神经网络模型采用BP神经网络，通过固定深度处的温度与前p阶EOF系数的历史水文数据完成神经网络的训练。
本发明还提供一种基于定深温度数据的全海深温度剖面反演系统，所述系统包括：
温度采集模块，用于实时获取两个固定深度处的海洋温度；
计算EOF系数模块，将获取的温度值输入预先建立和训练好的神经网络模型，得到p阶EOF系数；
反演全海深温度模块，用于从历史水文数据中提取出平均温度剖面与前p阶EOF基函数，利用p阶EOF系数计算全海深温度剖面。
所述系统还包括：
神经网络训练模块，用于通过固定深度处的温度与前p阶EOF系数的历史水文数据完成神经网络的训练。
本发明还可提供的一种计算机设备，包括：至少一个处理器、存储器、至少一个网络接口和用户接口。该设备中的各个组件通过总线系统耦合在一起。可理解，总线系统用于实现这些组件之间的连接通信。总线系统除包括数据总线之外，还包括电源总线、控制总线和状态信号总线。
其中，用户接口可以包括显示器、键盘或者点击设备(例如，鼠标，轨迹球(track ball)、触感板或者触摸屏等。
可以理解，本申请公开实施例中的存储器可以是易失性存储器或非易失性存储器，或可包括易失性和非易失性存储器两者。其中，非易失性存储器可以是只读存储器(Read-Only Memory，ROM)、可编程只读存储器(Programmable ROM，PROM)、可擦除可编程只读存储器(Erasable PROM，EPROM)、电可擦除可编程只读存储器(Electrically EPROM，EEPROM)或闪存。易失性存储器可以是随机存取 存储器(Random Access Memory，RAM)，其用作外部高速缓存。通过示例性但不是限制性说明，许多形式的RAM可用，例如静态随机存取存储器(Static RAM，SRAM)、动态随机存取存储器(Dynamic RAM，DRAM)、同步动态随机存取存储器(Synchronous DRAM，SDRAM)、双倍数据速率同步动态随机存取存储器(Double Data Rate SDRAM，DDRSDRAM)、增强型同步动态随机存取存储器(Enhanced SDRAM，ESDRAM)、同步连接动态随机存取存储器(Synchlink DRAM，SLDRAM)和直接内存总线随机存取存储器(Direct Rambus RAM，DRRAM)。本文描述的存储器旨在包括但不限于这些和任意其它适合类型的存储器。
在一些实施方式中，存储器存储了如下的元素，可执行模块或者数据结构，或者他们的子集，或者他们的扩展集：操作系统和应用程序。
其中，操作系统，包含各种系统程序，例如框架层、核心库层、驱动层等，用于实现各种基础业务以及处理基于硬件的任务。应用程序，包含各种应用程序，例如媒体播放器(Media Player)、浏览器(Browser)等，用于实现各种应用业务。实现本公开实施例方法的程序可以包含在应用程序中。
在本上述的实施例中，还可通过调用存储器存储的程序或指令，具体的，可以是应用程序中存储的程序或指令，处理器用于：
执行上述方法的步骤。
上述方法可以应用于处理器中，或者由处理器实现。处理器可能是一种集成电路芯片，具有信号的处理能力。在实现过程中，上述方法的各步骤可以通过处理器中的硬件的集成逻辑电路或者软件形式的指令完成。上述的处理器可以是通用处理器、数字信号处理器(Digital Signal Processor，DSP)、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列(Field Programmable Gate Array，FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。可以实现或者执行上述公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。结合上述公开的方法的步骤可以直接体现为硬件译码处理器执行完成，或者用译码处理器中的硬件及软件模块组合执行完成。软件模块可以位于随机存储器，闪存、只读存储器，可编程只读存储器或者电可擦写可编程存储器、寄存器等本领域成熟的存储介质中。该存储介质位于存储器，处理器读取存储器中的信息，结合其硬件完成上述方法的步骤。
可以理解的是，本发明描述的这些实施例可以用硬件、软件、固件、中间件、 微码或其组合来实现。对于硬件实现，处理单元可以实现在一个或多个专用集成电路(Application Specific Integrated Circuits，ASIC)、数字信号处理器(Digital Signal Processing，DSP)、数字信号处理设备(DSP Device，DSPD)、可编程逻辑设备(Programmable Logic Device，PLD)、现场可编程门阵列(Field-Programmable Gate Array，FPGA)、通用处理器、控制器、微控制器、微处理器、用于执行本申请所述功能的其它电子单元或其组合中。
对于软件实现，可通过执行本发明的功能模块(例如过程、函数等)来实现本发明技术。软件代码可存储在存储器中并通过处理器执行。存储器可以在处理器中或在处理器外部实现。
本发明还可提供一种非易失性存储介质，用于存储计算机程序。当该计算机程序被处理器执行时可以实现上述方法实施例中的各个步骤。
最后所应说明的是，以上实施例仅用以说明本发明的技术方案而非限制。尽管参照实施例对本发明进行了详细说明，本领域的普通技术人员应当理解，对本发明的技术方案进行修改或者等同替换，都不脱离本发明技术方案的精神和范围，其均应涵盖在本发明的权利要求范围当中。

Claims (9)

  1. 一种基于定深数据的全海深温度剖面反演方法，所述方法包括：
    步骤S1：实时获取两个固定深度处的海洋温度；
    步骤S2：将获取的温度值输入预先建立和训练好的神经网络模型，得到p阶EOF系数；
    步骤S3：从历史水文数据中提取出平均温度剖面与前p阶EOF基函数，利用p阶EOF系数计算全海深温度剖面。
  2. 根据权利要求1所述的基于定深温度数据的全海深温度剖面反演方法，其特征在于，所述神经网络模型采用BP神经网络，通过固定深度处的温度与前p阶EOF系数的历史水文数据完成神经网络的训练；
    BP神经网络的训练过程分为正向传播和反向传播两个阶段；
    首先是正向传播，即输入2个定深温度数据，经输入层传入，通过隐含层逐层处理后到达输出层，得到网络计算结果；根据损失函数loss＝|y out-y|计算出网络计算结果与实际输出之间的误差，其中y out为网络输出，y为实际输出；若损失误差超过设定阈值，则进入到反向传播阶段；
    反向传播是将损失误差通过隐含层反向传递输入层，并将误差平均分配到各层的所有单元，根据梯度下降法修正各个单元的连接权值；
    正向、反向传播循环进行，使得连接权值不断调整，直至损失误差在阈值范围内，完成模型训练。
  3. 根据权利要求1所述的基于定深温度数据的全海深温度剖面反演方法，其特征在于，所述步骤S1具体为：依据从历史水文数据中提取的第p阶EOF基函数两个极值点处对应的深度，在指定海域这两个固定深度处布设两个AUV，AUV上搭载温度传感器，在浮力基本平衡后通过调节首摇角和纵摇角来实时获取两个固定深度处的温度数据。
  4. 根据权利要求1所述的基于定深温度数据的全海深温度剖面反演方法，其特征在于，所述步骤S3具体为：
    获得历史水位数据中温度剖面在M个时间点的采样值，每个温度剖面经过层化处理在深度上具有N个值，将M个温度剖面样本表示成矩阵的形式：
    Figure PCTCN2022125403-appb-100001
    其中，t M(z N)表示温度剖面在第M个时间点第N个深度上的取值，计算M个温度剖面在每一层的平均温度，得到平均温度剖面
    Figure PCTCN2022125403-appb-100002
    Figure PCTCN2022125403-appb-100003
    其中，[] T表示对向量进行转置；将温度矩阵T减去平均温度剖面
    Figure PCTCN2022125403-appb-100004
    得到温度扰动：
    Figure PCTCN2022125403-appb-100005
    将ΔT进行奇异值分解，得到：
    ΔT T＝UΣV T
    其中，V＝[v 1,…v N]∈R N×N是矩阵ΔTΔT T∈R N×N的特征向量，即为要提取的经验正交函数；U＝[u 1,…u M]∈R M×M是矩阵ΔT TΔT∈R M×M的特征向量，Σ＝diag([λ 1 … λ N])∈R M×N表示特征值矩阵；每一个特征向量对应的特征值表示此特征向量的权重；前m阶模态函数的累积方差贡献率表示为：
    Figure PCTCN2022125403-appb-100006
    其中，
    Figure PCTCN2022125403-appb-100007
    λ k表示第k个特征值；
    利用贡献率E m大于设定值的前p阶EOF基函数重构温度剖面，得到全海深温度剖面矩阵：
    Figure PCTCN2022125403-appb-100008
    其中，[α 1M α 2M … α pM] T表示系数矩阵UΣ＝[α 1 … α N]∈R M×N的前p 阶；
    Figure PCTCN2022125403-appb-100009
    表示矩阵V＝[v 1,…,v N]∈R N×N的前p阶；
    其中，温度剖面中第i个深度温度值为：
    Figure PCTCN2022125403-appb-100010
    其中，i∈N；α p表示第p阶EOF系数，v p表示第p阶EOF基函数，
    Figure PCTCN2022125403-appb-100011
    表示历史水位数据第i个深度平均温度。
  5. 根据权利要求4所述的基于定深温度数据的全海深温度剖面反演方法，其特征在于，所述p优选2。
  6. 一种基于定深温度数据的全海深温度剖面反演系统，所述系统包括：
    温度采集模块，用于实时获取两个固定深度处的海洋温度；
    计算EOF系数模块，将获取的温度值输入预先建立和训练好的神经网络模型，得到p阶EOF系数；
    反演全海深温度模块，用于从历史水文数据中提取出平均温度剖面与前p阶EOF基函数，利用p阶EOF系数计算全海深温度剖面。
  7. 根据权利要求6所述的基于定深温度数据的全海深温度剖面反演系统，其特征在于，所述系统还包括：
    神经网络训练模块，用于通过固定深度处的温度与前p阶EOF系数的历史水文数据完成神经网络的训练；
    神经网络采用BP神经网络，训练过程分为正向传播和反向传播两个阶段；
    首先是正向传播，即输入2个定深温度数据，经输入层传入，通过隐含层逐层处理后到达输出层，得到网络计算结果；根据损失函数loss＝|y out-y|计算出网络计算结果与实际输出之间的误差，其中y out为网络输出，y为实际输出；若损失误差超过设定阈值，则进入到反向传播阶段；
    反向传播是将损失误差通过隐含层反向传递输入层，并将误差平均分配到各层的所有单元，根据梯度下降法修正各个单元的连接权值；
    正向、反向传播循环进行，使得连接权值不断调整，直至损失误差在阈值范围内，完成模型训练。
  8. 一种计算机设备，包括存储器、处理器及存储在所述存储器上并可在所述 处理器上运行的计算机程序，其特征在于，所述处理器执行所述计算机程序时实现如权利要求1至5中任一项所述的方法。
  9. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质存储有计算机程序，所述计算机程序当被处理器执行时使所述处理器执行如权利要求1至5任一项所述的方法。
PCT/CN2022/125403 2022-09-29 2022-10-14 一种基于定深数据的全海深温度剖面反演方法及系统 WO2024065888A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211200508.7A CN115409160B (zh) 2022-09-29 2022-09-29 一种基于定深数据的全海深温度剖面反演方法及系统
CN202211200508.7 2022-09-29

Publications (1)

Publication Number Publication Date
WO2024065888A1 true WO2024065888A1 (zh) 2024-04-04

Family

ID=84167175

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/125403 WO2024065888A1 (zh) 2022-09-29 2022-10-14 一种基于定深数据的全海深温度剖面反演方法及系统

Country Status (2)

Country Link
CN (1) CN115409160B (zh)
WO (1) WO2024065888A1 (zh)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2000298069A (ja) * 1999-04-14 2000-10-24 Oki Electric Ind Co Ltd 海洋音響トモグラフィデータ処理・表示装置
CN108981957A (zh) * 2018-05-31 2018-12-11 西北工业大学 基于自组织神经网络及经验正交函数的水下温度场重构方法
CN111259943A (zh) * 2020-01-10 2020-06-09 天津大学 基于机器学习的温跃层预测方法
CN111307266A (zh) * 2020-02-21 2020-06-19 山东大学 声速获取方法及基于该方法的全球海洋声速场构建方法
CN112598113A (zh) * 2020-12-15 2021-04-02 广东海洋大学 一种基于自组织竞争型神经网络的海洋声速剖面获取方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110057365B (zh) * 2019-05-05 2022-06-21 哈尔滨工程大学 一种大潜深auv下潜定位方法
CN112596412A (zh) * 2020-12-11 2021-04-02 中国科学院沈阳自动化研究所 一种多auv仿真平台
CN114818232A (zh) * 2021-01-19 2022-07-29 中国科学院声学研究所 一种海水空间温度剖面的重构方法及系统
CN113191087B (zh) * 2021-02-02 2022-09-27 中国人民解放军海军大连舰艇学院 联合历史剖面声速与实测表层声速的航渡式测深数据剖面声速改正方法
CN113218493B (zh) * 2021-04-08 2023-12-08 中国人民解放军国防科技大学 一种基于经验正交函数法的声速剖面反演方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2000298069A (ja) * 1999-04-14 2000-10-24 Oki Electric Ind Co Ltd 海洋音響トモグラフィデータ処理・表示装置
CN108981957A (zh) * 2018-05-31 2018-12-11 西北工业大学 基于自组织神经网络及经验正交函数的水下温度场重构方法
CN111259943A (zh) * 2020-01-10 2020-06-09 天津大学 基于机器学习的温跃层预测方法
CN111307266A (zh) * 2020-02-21 2020-06-19 山东大学 声速获取方法及基于该方法的全球海洋声速场构建方法
CN112598113A (zh) * 2020-12-15 2021-04-02 广东海洋大学 一种基于自组织竞争型神经网络的海洋声速剖面获取方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
HAO, RIXU ET AL.: "Sea Temperature Forecasting Method Based on EOF-NAR Neural Network Hybrid Model", PROCEEDINGS OF THE 2019 OCEAN ACADEMIC (INTERNATIONAL) BIENNIAL CONFERENCE OF THE OCEAN SOCIETY OF CHINA, 19 August 2022 (2022-08-19) *
LI HONGLIN, LI QIANQIAN; ZHOU KANGYING; YAN XIAN; CAO SHOULIAN; MA ZHICHUAN: "Indian Ocean Sound Velocity Profile Analysis of CTD and Argo Data in Winter", vol. 40, no. 5, 30 September 2021 (2021-09-30), pages 738 - 744, XP093151508, ISSN: 1000-310X, DOI: 10.11684/j.issn.1000-310X.2021.05.012 *

Also Published As

Publication number Publication date
CN115409160B (zh) 2023-04-28
CN115409160A (zh) 2022-11-29

Similar Documents

Publication Publication Date Title
CA3067573A1 (en) Target tracking systems and methods for uuv
CN102980579B (zh) 一种自主水下航行器自主导航定位方法
CN107589749B (zh) 水下机器人自主定位与节点地图构建方法
CN113191087B (zh) 联合历史剖面声速与实测表层声速的航渡式测深数据剖面声速改正方法
JP2017533514A (ja) 順序再帰的フィルタリング３次元変分に基づく実測海洋環境データ同化方法
CN112598113A (zh) 一种基于自组织竞争型神经网络的海洋声速剖面获取方法
Mittendorf et al. Sea state identification using machine learning—A comparative study based on in-service data from a container vessel
CN109579850A (zh) 基于对水速度辅助惯导的深水智能导航方法
CN112710304B (zh) 一种基于自适应滤波的水下自主航行器导航方法
Xue et al. Online identification of a ship maneuvering model using a fast noisy input Gaussian process
WO2024065888A1 (zh) 一种基于定深数据的全海深温度剖面反演方法及系统
CN116805028B (zh) 一种基于浮体运动响应的波面反演方法及系统
Tang et al. Domain decomposition and matching for time-domain analysis of motions of ships advancing in head sea
Zhou et al. Nonparametric modeling of ship maneuvering motions in calm water and regular waves based on R-LSTM hybrid method
CN116202492A (zh) 一种基于高低频分离的随船波浪测量方法
CN112085779B (zh) 一种波浪参数估算方法及装置
CN115469314A (zh) 一种均匀圆环阵稳健水下目标方位跟踪方法及系统
CN110767322B (zh) 一种基于响应面模型的海洋浮式平台热点应力推算方法
RU138640U1 (ru) Информационно-управляющий комплекс для автоматизации судовождения и динамического позиционирования судна
Yao et al. Extreme motion predictions of semi-submersible platform based on deep learning method and prototype monitoring data
King et al. Computational issues on observability and optimal sensor locations
Xu et al. Prediction of Total Force and Moment of Ship Based on Improved BP Neural Network
Ouimet et al. Collective estimation of ocean nonlinear internal waves using robotic underwater drifters
CN117572874B (zh) 一种自回正无人船及控制方法
CN103336525A (zh) 随机系统高权值便捷ukf滤波方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22960491

Country of ref document: EP

Kind code of ref document: A1