Multi-frequency ADCP collaborative observation and high-resolution data fusion method under the influence of deep thermocline

Abstract

This invention relates to the field of underwater acoustic measurement technology, and more particularly to a method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of a deep thermocline. The method includes: acquiring low-frequency velocity data and sound velocity profile data using a low-frequency primary ADCP; when the gradient in the sound velocity profile data exceeds a preset trigger threshold, recording the thermocline depth range and initiating a high-frequency auxiliary ADCP to acquire high-frequency velocity data; calculating the normalized sound velocity gradient layer by layer for the sound velocity profile data, and generating a layer thickness sequence and a regularization coefficient sequence from the normalized sound velocity gradients of each layer. This invention achieves collaborative data acquisition of the full-depth background field and thermocline focused observation by constructing a hardware system including a low-frequency ADCP primary detection subsystem, a high-frequency ADCP auxiliary enhancement subsystem, and a synchronous environmental sensing unit.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic measurement technology, and in particular to a method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines. Background Technology

[0002] An Acoustic Doppler current profiler (ADCP) is an instrument that uses the Doppler effect to measure the velocity and vertical distribution of water flow. It is widely used in marine, hydrological, river, lake, and engineering monitoring fields. The ADCP emits ultrasonic waves into the water, where suspended particles, bubbles, and plankton reflect these waves. If these scattering bodies move with the water flow, the frequency of the returning sound waves shifts due to the Doppler effect. The instrument calculates the flow velocity at different depths by analyzing this frequency shift. A deep thermocline refers to a transitional layer in large bodies of water such as oceans and lakes, located at a relatively deep depth, where the temperature changes rapidly with depth. Under the influence of deep thermoclines (deep layers with rapid temperature changes), multi-frequency ADCPs are significantly affected by changes in the acoustic propagation environment, changes in water stratification dynamics, and differences in multi-frequency channel responses. In deep water environments, traditional shipborne mobile ADCPs face a core technical contradiction due to physical limitations: they cannot simultaneously achieve sufficient detection depth, vertical resolution, and measurement accuracy for thin structures such as thermoclines. Summary of the Invention

[0003] Therefore, it is necessary for the present invention to provide a method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines, so as to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines includes the following steps: Step S1: Use the low-frequency main ADCP to collect low-frequency velocity data and sound velocity profile data; when the gradient in the sound velocity profile data exceeds the preset trigger threshold, record the layer depth range and start the high-frequency auxiliary ADCP to collect high-frequency velocity data. Step S2: Calculate the normalized sound velocity gradient layer by layer for the sound velocity profile data, and generate a layer thickness sequence and a regularization coefficient sequence from the normalized sound velocity gradient of each layer. Step S3: Perform acoustic ray tracking on the low-frequency velocity data and high-frequency velocity data according to the layer thickness sequence, and output the low-frequency corrected flow velocity field, high-frequency corrected flow velocity field and angular deviation sequence; convert the angular deviation sequence into a confidence weight sequence; Step S4: Use the confidence weight sequence to weight the layer-by-layer deviation of the high-frequency corrected velocity field within the depth range to form a high-frequency fitting term; use the layer-by-layer deviation of the low-frequency corrected velocity field within the full water depth range to form a low-frequency constraint term; use the regularization coefficient sequence to weight the velocity difference between adjacent layers within the full water depth to form a regularization term; the velocity distribution corresponding to the minimum value of the sum of the three terms is used as the fused velocity profile; and output the fused analysis error of each depth layer at the minimum value, and summarize them to form a fused confidence field; Step S5: Extract the maximum velocity shear value within the depth range of the fused velocity profile, label the confidence level of each layer with the fused confidence field, and output the velocity profile product.

[0005] This invention fundamentally overcomes the long-standing physical conflict between detection depth and vertical resolution in single-frequency ADCPs by constructing a master-slave collaborative observation architecture consisting of a low-frequency primary ADCP and a high-frequency secondary ADCP. The low-frequency primary ADCP is responsible for continuous scanning of the background field across the entire water depth, while the high-frequency secondary ADCP is activated in real time and focused on the depth range of the strata when the sound velocity gradient exceeds the trigger threshold. The two have a clear division of labor and complement each other, achieving both deep-penetration capability and high-resolution observation of fine thin-layer structures within a single system—a capability that is impossible in traditional single-frequency ADCP systems.

[0006] Regarding data processing accuracy, this invention introduces an adaptive three-dimensional acoustic ray tracking algorithm driven by a layered thickness sequence. Traditional ADCP data processing is based on the assumption of straight-line acoustic ray propagation, which can lead to significant spatial positioning errors and velocity synthesis errors in regions with abrupt changes in sound velocity gradients, such as thermoclines. This invention tracks the actual bending path of the acoustic ray layer by layer based on real-time sound velocity profiles, and independently corrects the actual three-dimensional spatial coordinates and radial velocity synthesis direction for each distance unit of each beam. This achieves dual synchronous correction of spatial position errors and velocity geometry, eliminating measurement deviations introduced by acoustic ray bending at the source.

[0007] Regarding data fusion quality, this invention establishes a cross-step synergistic driving relationship between ray tracking calculation accuracy and fusion inversion constraint strength by synchronously introducing the normalized sound velocity gradient into the layer thickness sequence and the regularization coefficient sequence. At the core of the megalayer, the minimum layer thickness ensures the highest ray tracking resolution, and the minimum synchronous regularization coefficient ensures the fusion algorithm retains the fine structure of the high-frequency corrected velocity field most fully. Both enhancement effects reach their maximum simultaneously at the point of strongest sound velocity gradient, forming a physically self-consistent synergistic gain. Introducing a wavenumber spectrum constraint term further ensures that the vertical energy spectrum of the fused velocity profile conforms to the statistical characteristics of ocean dynamic processes, eliminating non-physical high wavenumber noise that may be introduced due to the spatial coverage of high-frequency velocity data being limited to the megalayer interval.

[0008] Regarding the reliability of the output product, this invention converts the angular deviation sequence into a confidence weight sequence, directly transforming the ray bending amplitude at each depth layer during ray tracking into a confidence level indicator for the fusion result. Furthermore, in constructing the fusion confidence field, a dual amplification process of confidence weight and regularization coefficient is applied to the fusion analysis error in the interlayer interval, enabling the output fusion confidence field to quantitatively reflect the physical facts of large ray bending amplitude, weak fusion constraints, and high uncertainty at the core of the interlayer. The final output velocity profile product simultaneously carries a layer-by-layer confidence level sequence and a confidence label at the shear peak depth, providing users with a quantitative reference for the reliability of the fusion result, significantly improving the usability and reliability of observational data in the analysis of ocean dynamic processes. Attached Figure Description

[0009] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic diagram of the steps involved in a method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of a deep thermocline. Figure 2 This is a schematic diagram of the structure and hierarchical observation of a multi-frequency ADCP collaborative observation system according to one embodiment; Figure 3 This is a schematic diagram of a sound ray bending correction and angular deviation sequence generation embodiment. Detailed Implementation

[0010] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0011] To achieve the above objectives, please refer to Figures 1 to 3 This invention provides a method for collaborative observation and high-resolution data fusion of multi-frequency ADCP under the influence of deep thermoclines, applicable to multi-frequency ADCP including low-frequency primary ADCP and high-frequency secondary ADCP, comprising the following steps: Step S1: Use the low-frequency main ADCP to collect low-frequency velocity data and sound velocity profile data; when the gradient in the sound velocity profile data exceeds the preset trigger threshold, record the layer depth range and start the high-frequency auxiliary ADCP to collect high-frequency velocity data. Step S2: Calculate the normalized sound velocity gradient layer by layer for the sound velocity profile data, and generate a layer thickness sequence and a regularization coefficient sequence from the normalized sound velocity gradient of each layer. Step S3: Perform acoustic ray tracking on the low-frequency velocity data and high-frequency velocity data according to the layer thickness sequence, and output the low-frequency corrected flow velocity field, high-frequency corrected flow velocity field and angular deviation sequence; convert the angular deviation sequence into a confidence weight sequence; Step S4: Use the confidence weight sequence to weight the layer-by-layer deviation of the high-frequency corrected velocity field within the depth range to form a high-frequency fitting term; use the layer-by-layer deviation of the low-frequency corrected velocity field within the full water depth range to form a low-frequency constraint term; use the regularization coefficient sequence to weight the velocity difference between adjacent layers within the full water depth to form a regularization term; the velocity distribution corresponding to the minimum value of the sum of the three terms is used as the fused velocity profile; and output the fused analysis error of each depth layer at the minimum value, and summarize them to form a fused confidence field; Step S5: Extract the maximum velocity shear value within the depth range of the fused velocity profile, label the confidence level of each layer with the fused confidence field, and output the velocity profile product.

[0012] like Figure 2 As shown, the observation system of this invention is installed on a platform above the survey vessel (cross-section) and includes a central controller, a low-frequency main ADCP, a high-frequency auxiliary ADCP, and a sound velocity profiler. The central controller is communicatively connected to both the low-frequency main ADCP and the high-frequency auxiliary ADCP, receiving sound velocity profile data, identifying the depth range of the supercluster, and sending trigger commands to the high-frequency auxiliary ADCP. The low-frequency main ADCP is positioned on one side or bottom of the hull and emits sound pulses at a lower operating frequency across the entire water depth range, generating low-frequency velocity data covering the water surface to the maximum detection depth; its detection beam is represented by a dashed line. The high-frequency auxiliary ADCP is positioned adjacent to the low-frequency main ADCP and emits sound pulses at a higher operating frequency across the supercluster depth range and its upper and lower extensions, generating high-frequency velocity data; its detection beam is represented by a solid line.

[0013] The water body in the figure consists of an upper homogeneous layer, a central mezzanine depth range, and a lower homogeneous layer. The mezzanine depth range is represented by the blue filled area, and its upper and lower boundaries are marked by dashed lines. The low-frequency main ADCP continuously observes the entire water depth and is suitable for obtaining the large-scale structure of the background flow field; the high-frequency auxiliary ADCP focuses on observing only the mezzanine interval and is suitable for obtaining the fine velocity structure of the thin layer inside the mezzanine.

[0014] Furthermore, both the low-frequency primary ADCP and the high-frequency secondary ADCP synchronously receive the 1PPS pulse signal output by the Global Navigation Satellite System, using the rising edge of the second pulse as the time reference to achieve strict time alignment of the data frames from both devices. This ensures the consistency of observation results from different frequency bands in the time dimension during subsequent multi-source data fusion. This structural design achieves coordinated operation between low-frequency full-coverage observation and high-frequency local fine-grained observation, improving the observation resolution and reliability of velocity profiles in deep thermocline environments while controlling energy consumption and computational load.

[0015] Furthermore, step S1 includes the following steps: Step S11: Use the low-frequency main ADCP to transmit acoustic pulses to the full water depth and receive the echo signal to generate low-frequency velocity data. The full water depth acoustic pulse transmission means transmitting acoustic pulses to the water column range from the instrument position to the maximum detection depth of the target water body in one working cycle. In one embodiment, the low-frequency main ADCP operates at a frequency of 150 kHz or 300 kHz. During one working cycle, the low-frequency main ADCP emits acoustic pulses across a water column extending from the instrument's installation location to the maximum detection depth of the target water body, and receives echo signals from each depth layer. Based on the Doppler frequency shift of the echo signals from each depth layer, the low-frequency main ADCP calculates the three-dimensional flow velocity at each layer, generating low-frequency velocity data. This low-frequency velocity data covers the entire water depth range from the water surface to the maximum detection depth, with a vertical resolution of 1 to 8 meters.

[0016] Step S12: Use a sound velocity profiler to synchronously collect sound velocity values ​​at each layer of the full water depth to form sound velocity profile data; In one embodiment, the sound velocity profiler operates synchronously with the low-frequency main ADCP. While the low-frequency main ADCP completes one working cycle, it collects sound velocity values ​​layer by layer across the entire water depth from the water surface to the maximum detection depth, forming sound velocity profile data. The sound velocity profile data is indexed by depth, recording the sound velocity value c(z) corresponding to each sampling depth, in meters per second.

[0017] Step S13: Subtract the sound velocity values ​​of adjacent depth layers in the sound velocity profile data and divide by the corresponding depth interval to obtain the local sound velocity gradient values ​​of each layer, forming a layer-by-layer sound velocity gradient.

[0018] In one embodiment, for each pair of adjacent sampling depth layers in the sound velocity profile data, the difference between the shallow and deep sound velocity values ​​is calculated and divided by the depth interval between the two layers to obtain the local sound velocity gradient value dc / dz between the adjacent layers, in units of per second. This calculation is performed pairwise for all adjacent depth layers in the sound velocity profile data, and the resulting local sound velocity gradient values ​​are arranged by depth to form a layer-by-layer sound velocity gradient. The depth resolution of the layer-by-layer sound velocity gradient is consistent with that of the sound velocity profile data.

[0019] Furthermore, step S1 also includes the following steps: Step S14: Compare the gradient values ​​of each layer of the sound velocity gradient with the preset trigger threshold layer by layer. When the gradient values ​​of each layer in the continuous depth interval are not lower than the trigger threshold and the thickness of the depth interval is not less than 2 meters, the shallowest depth of the depth interval is recorded as the upper boundary of the jump layer and the deepest depth is recorded as the lower boundary of the jump layer, thus forming the jump layer depth range. In one embodiment, the local sound velocity gradient values ​​dc / dz of each depth layer in the layer-by-layer sound velocity gradient are compared layer by layer with a preset trigger threshold, which is set to 0.5s⁻¹. When the local sound velocity gradient values ​​of each layer in a continuous depth interval are not less than 0.5s⁻¹ and the thickness of the interval is not less than 2 meters, the shallowest depth of the interval is recorded as the upper boundary of the mezzanine layer and the deepest depth is recorded as the lower boundary of the mezzanine layer, forming the mezzanine layer depth range.

[0020] Among the above judgment conditions, the restriction that the thickness of the continuous interval is not less than 2 meters is used to exclude gradient abrupt changes caused by single-point noise, ensuring that the identified thermocline depth range corresponds to the real thermocline structure rather than measurement interference. When the local sound velocity gradient values ​​of each layer in the layer-by-layer sound velocity gradient do not exceed the trigger threshold of 0.5s⁻¹, a thermocline depth range is not formed, step S15 is not executed, the high-frequency auxiliary ADCP remains in a low-power standby state, and no high-frequency velocity data is generated.

[0021] Step S15: Send a start command to the high-frequency auxiliary ADCP, configure the observation window to be the interval between 0.5-2 meters above the upper boundary of the mezzanine and 0.5-2 meters below the lower boundary of the mezzanine, and collect high-frequency velocity data in this interval.

[0022] In one embodiment, the central controller sends a hardware trigger signal to the high-frequency auxiliary ADCP based on the depth range of the mesopelagic layer, initiating focused observation of the high-frequency auxiliary ADCP within the corresponding water layer interval. The high-frequency auxiliary ADCP operates at a frequency of 600 kHz or 1.2 MHz, with a vertical resolution of 0.25 to 0.5 meters. The observation window of the high-frequency auxiliary ADCP is configured to extend from 0.5 to 2 meters above the upper boundary of the mesopelagic layer to 0.5 to 2 meters below the lower boundary of the mesopelagic layer. The observation window is extended upwards and downwards by 0.5 to 2 meters based on the depth range of the mesopelagic layer to ensure sufficient coverage of the velocity transition region near the top and bottom boundaries of the mesopelagic layer.

[0023] The high-frequency auxiliary ADCP acquires echo signals from each depth layer within the aforementioned observation window and calculates flow velocities to form high-frequency velocity data. The spatial coverage of the high-frequency velocity data is limited to the depth range of the mezzanine and its upper and lower extensions. Its vertical resolution is significantly higher than that of the low-frequency velocity data. The two data overlap in depth within the mezzanine depth range, forming a dual-source data input for subsequent acoustic ray tracking and fusion processing.

[0024] Both the main ADCP and the high-frequency auxiliary ADCP synchronously receive the 1PPS pulse signal output by the global navigation satellite system, using the rising edge of the pulse signal as the time reference for their respective data frames, ensuring that the low-frequency velocity data and the high-frequency velocity data are strictly aligned in the time dimension.

[0025] Furthermore, step S2 includes the following steps: Step S21: Take the sound velocity value of the shallowest layer in the sound velocity profile data as the surface reference sound velocity; In one embodiment, the sound velocity value corresponding to the shallowest sampling layer in the sound velocity profile data is taken as the surface reference sound velocity. Surface reference sound speed This is used in the subsequent step S22 to normalize the sound velocity gradient of each layer, so that the sound velocity gradient values ​​of different voyages and different sea areas are comparable.

[0026] Step S22: Subtract the sound velocity values ​​of each adjacent depth layer in the sound velocity profile data, divide by the corresponding depth interval, and then divide by the surface reference sound velocity to obtain the normalized sound velocity gradient value of each layer. Take the absolute value and summarize to form the layer-by-layer normalized sound velocity gradient. In one embodiment, for each pair of adjacent sampling depth layers in the sound velocity profile data, the difference between the shallow layer sound velocity value and the deep layer sound velocity value is divided by the depth interval between the two layers, and then divided by the surface reference sound velocity. The normalized sound velocity gradient between adjacent layers is obtained, calculated as follows: ,in The values ​​represent the local sound velocity gradients obtained in step S13. For each pair of adjacent depth layers in the sound velocity profile data, the normalized sound velocity gradient values ​​are calculated and their absolute values ​​are taken. The obtained normalized sound velocity gradient values ​​for each layer are arranged by depth and summarized to form a layer-by-layer normalized sound velocity gradient. The depth resolution of the layer-by-layer normalized sound velocity gradient is consistent with the sound velocity profile data. Within the depth range of the super-layer, the normalized sound velocity gradient values ​​of each layer are significantly higher than those of the uniform layer, and the normalized sound velocity gradient value reaches its maximum at the core of the super-layer.

[0027] Step S23: Determine the layer thickness of each layer based on the normalized sound velocity gradient values ​​of each layer in the layer-by-layer normalized sound velocity gradient, where the normalized sound velocity gradient value is inversely proportional to the layer thickness, thus obtaining the layer thickness sequence. In one embodiment, the normalized sound velocity gradient value g(z) of each depth layer in the layer-by-layer normalized sound velocity gradient is determined according to the inverse relationship between the normalized sound velocity gradient value and the layer thickness. The layer thickness is determined using the following continuous thickness function: in The maximum preset layer thickness is 2 meters; The minimum preset layer thickness is set at 0.2 meters; This is a dynamically updated threshold based on recent layer-by-layer normalized sound velocity gradient statistics. When the normalized sound velocity gradient value g(z) approaches zero, the layer thickness approaches [the threshold value is missing from the original text]. That is, 2 meters; when the normalized sound velocity gradient value Much larger At that time, the layer thickness approaches That is, 0.2 meters. The normalized sound velocity gradient value is the largest at the core of the megalayer, corresponding to the smallest layer thickness, and the calculation resolution of sound ray tracking in step S3 is the highest for this layer; the normalized sound velocity gradient value is close to zero in the uniform layer, corresponding to the largest layer thickness and the least computational cost. After determining the layer thickness layer by layer for all depth layers with normalized sound velocity gradients, the layers are arranged and summarized according to depth to form a layer thickness sequence.

[0028] Step S24: Determine the regularization coefficient for each layer of the same normalized sound velocity gradient value in the layer-by-layer normalized sound velocity gradient. The regularization coefficient is inversely proportional to the normalized sound velocity gradient value between 0 and 1, forming a regularization coefficient sequence.

[0029] In one embodiment, the normalized sound velocity gradient values ​​are the same for each depth layer in the layer-by-layer normalized sound velocity gradient. The regularization coefficients for each layer are determined based on the inverse relationship between the regularization coefficient and the normalized sound velocity gradient value, where the regularization coefficient is greater than 0 and does not exceed 1. The regularization coefficients are determined using the following adaptive adjustment function: in The absolute value of the sound velocity gradient at each layer. To adjust the sensitivity coefficient. When the normalized sound velocity gradient value is too large, If the normalized velocity gradient is too small, the regularization coefficient will be too small, and the constraint on the vertical smoothness of the candidate velocity profile in step S44 will be weakened, allowing the fused velocity profile to retain strong velocity shear details at the core of the mesophase layer; when the normalized sound velocity gradient value is close to zero... With a regularization coefficient close to 1, this layer exhibits the strongest smoothing constraint, effectively suppressing noise fluctuations within the uniform layer.

[0030] After determining the regularization coefficients for each depth layer of the normalized sound velocity gradient, the coefficients are arranged and summarized by depth to form a regularization coefficient sequence. This regularization coefficient sequence perfectly corresponds to the depth distribution of the layer thickness sequence; both are derived from the same normalized sound velocity gradient values ​​in each layer of the normalized sound velocity gradient. Simultaneously determined: the layer thickness and regularization coefficient are smallest at the core of the megalayer, while the layer thickness and regularization coefficient are largest at the uniform layer, reflecting the cross-step synergistic driving relationship between the normalized sound velocity gradient and the accuracy of sound ray tracking calculation and the strength of fusion constraint.

[0031] Furthermore, step S3 includes the following steps: Step S31: Based on the layer thickness sequence, trace the sound path layer by layer for each distance unit of each beam in the low-frequency velocity data, and output the low-frequency corrected velocity field after coordinate correction of each measurement unit. When the gradient values ​​of each layer of the layer-by-layer sound velocity gradient do not exceed the trigger threshold, no layer depth range is generated, the high-frequency auxiliary ADCP remains in standby mode, and no high-frequency velocity data is generated. In one embodiment, the layer thickness of each depth layer in the layer thickness sequence is used as the calculation layer basis for ray tracking. For each distance unit of each beam in the low-frequency velocity data, the ray path is tracked layer by layer starting from the water surface. The sound velocity profile data is divided into several calculation layers according to the layer thickness sequence. Within each calculation layer, the horizontal displacement of the ray within that layer is calculated according to Snell's law based on the sound velocity values ​​at the top and bottom of that layer. and vertical displacement The calculation formula is: in For the first Calculate the sound velocity at the top of each layer. For the first Calculate the sound velocity gradient within the layer. and The entry and exit of the vocal timbre are respectively the first and second parts of the vocal timbre. The grazing angle during calculation. For each calculation layer, the horizontal and vertical displacements are accumulated layer by layer until the total length of the sound ray path equals the nominal slant distance of that distance cell, thus obtaining the actual three-dimensional spatial coordinates of that distance cell.

[0032] The nominal coordinates are replaced with actual three-dimensional spatial coordinates, and the composite direction vector of the radial velocity of each beam is recalculated. The radial velocity of each distance unit in the low-frequency velocity data is then reprojected into three-dimensional velocity components based on the corrected composite direction vector, and the low-frequency corrected velocity field of each measurement unit is output. The low-frequency corrected velocity field covers the entire water depth range from the water surface to the maximum detection depth, and the three-dimensional spatial coordinates and velocity values ​​of each measurement unit have been corrected for acoustic curvature.

[0033] When the gradient values ​​of each layer of the layer-by-layer sound velocity gradient do not exceed the trigger threshold, no layer depth range is generated, the high-frequency auxiliary ADCP remains in a low-power standby state, no high-frequency velocity data is generated, no high-frequency corrected flow velocity field is generated, and no high-frequency fitting term is calculated. The target quantity in step S45 consists only of the sum of the low-frequency constraint term and the regularization term.

[0034] Step S32: When high-frequency velocity data exists, trace the sound path layer by layer for each distance unit of each beam in the high-frequency velocity data using a layered thickness sequence, and output the high-frequency corrected flow velocity field. In one embodiment, when high-frequency velocity data is available, the layer thickness of each depth layer in the layer thickness sequence is used as the calculation basis for ray tracking. For each distance unit of each beam in the high-frequency velocity data, the ray path is tracked layer by layer from the water surface according to the same Snell's law as in step S31. Within each calculation layer, the horizontal and vertical displacements of the ray within that layer are calculated according to the calculation formula described in step S31 based on the sound velocity values ​​at the top and bottom of that layer. These displacements are accumulated layer by layer until the total length of the ray path equals the nominal slant distance of that distance unit, thus obtaining the actual three-dimensional spatial coordinates of that distance unit.

[0035] The composite direction vector of the radial velocity of each beam is recalculated using actual three-dimensional spatial coordinates. The radial velocity of each distance unit in the high-frequency velocity data is then reprojected into three-dimensional velocity components based on the corrected composite direction vector, outputting a high-frequency corrected velocity field. The spatial coverage of the high-frequency corrected velocity field is consistent with that of the high-frequency velocity data, limited to the depth range of the stratum and its upper and lower extension intervals. The three-dimensional spatial coordinates and velocity values ​​of each measurement unit have been corrected for acoustic ray curvature.

[0036] Furthermore, step S3 also includes the following steps: Step S33: Divide the sound velocity profile data into layers of each depth according to the layer thickness sequence and record them as calculation layers. During the sound ray tracking process, calculate the absolute value of the difference between the incident angle of the sound ray entering the calculation layer and the exit angle after passing through the calculation layer for each calculation layer, and record it as the single-layer angular deviation of the calculation layer. Accumulate the single-layer angular deviations of all calculation layers that the sound ray passes through from the water surface to each target depth layer by layer to obtain the cumulative angular deviation at each target depth, and summarize them to form an angular deviation sequence. In one embodiment, the sound velocity profile data is divided into layers of varying depths according to a layer thickness sequence, and these layers are designated as computational layers. During ray tracking, the incident angle of the sound ray entering each computational layer is recorded. and the exit angle after passing through the computing layer The absolute value of the difference between the angle of incidence and the angle of departure. This is denoted as the single-layer angular deviation of the computational layer. The single-layer angular deviation reflects the degree to which the sound velocity gradient deflects the direction of sound ray propagation within the computational layer. In the core of the analytic layer with a large normalized sound velocity gradient, the single-layer angular deviation of each computational layer is relatively large, while in the homogeneous layer where the normalized sound velocity gradient is close to zero, the single-layer angular deviation of each computational layer is close to zero.

[0037] For each target depth in both low-frequency and high-frequency velocity data, the single-layer angular deviation of all computational layers through which the sound ray propagates from the water surface to that target depth is accumulated layer by layer to obtain the cumulative angular deviation at that target depth. The cumulative angular deviation generally increases with increasing target depth, and its rate of increase is significantly higher than that of the uniform layer when passing through the depth range of interlayers due to the larger single-layer angular deviation of each computational layer.

[0038] After calculating the cumulative angular deviation for each target depth across the entire water depth, the results are arranged and summarized by depth to form an angular deviation sequence. The angular deviation sequence covers all depth layers of both the low-frequency and high-frequency corrected velocity fields. Within the interlayer depth range, the cumulative angular deviation of each layer is significantly greater than that of the uniform layer.

[0039] Step S34: Add the cumulative angular deviation of each layer in the angular deviation sequence to 1, use the sum as the denominator and 1 as the numerator, and use the resulting ratio as the confidence weight value to form a confidence weight sequence.

[0040] In one embodiment, the cumulative angular deviation value of each depth layer in the angular deviation sequence is added to 1. The ratio obtained by adding the result as the denominator and using 1 as the numerator is used as the confidence weight value of that layer. When the cumulative angular deviation value is zero, the confidence weight value is 1, indicating that the sound ray propagation direction of that depth layer has not been deflected and the residual uncertainty after coordinate correction of the measurement unit is the lowest. The larger the cumulative angular deviation value, the smaller the confidence weight value, indicating that the sound ray bending amplitude of that depth layer is greater and the residual uncertainty after coordinate correction is higher.

[0041] After calculating the confidence weight values ​​for each depth layer of the diagonal deviation sequence, the layers are arranged and summarized by depth to form a confidence weight sequence. The depth distribution of the confidence weight sequence corresponds perfectly to that of the diagonal deviation sequence. The confidence weight values ​​decrease monotonically with the increase of the cumulative diagonal deviation within a range greater than 0 and not exceeding 1. The confidence weight values ​​of each layer within the depth range of the skip layer are significantly lower than those of the uniform layer, and the core layer of the skip layer has the smallest confidence weight value.

[0042] like Figure 3 As shown, the left side is a schematic diagram of acoustic ray bending correction, and the right side is a schematic diagram of angular deviation sequence. The survey vessel is located at the water surface, and the water body consists of an upper homogeneous layer, a mesophilic depth range, and a lower homogeneous layer. The upper and lower boundaries of the mesophilic layers are marked at their respective depths. Due to the significant increase in the sound velocity gradient within the mesophilic region, the sound wave propagation path undergoes significant refraction in this region, resulting in a deviation between the nominal coordinates under the traditional straight-line propagation assumption and the actual measured coordinates.

[0043] In the left figure, the gray dashed line represents the nominal path obtained under the straight-line propagation assumption without correction; the red solid line represents the actual path obtained after performing ray tracking based on the sound velocity profile data. The actual path bends within the jump-layer interval and ultimately falls at the actual coordinate position. The difference in horizontal distance between the nominal and actual coordinates represents the spatial coordinate deviation, which directly affects the ADCP velocity inversion results. The ray tracking correction method of this invention can correct the actual spatial position of each distance unit, improving the spatial accuracy of the velocity measurement results.

[0044] As the sound ray travels through each computational layer, the angle of incidence when the sound ray enters that layer is recorded. Angle of departure from that layer The absolute difference between the two Defined as single-layer angular deviation. The enlarged circle in the figure illustrates how the single-layer angular deviation is calculated within the mezzanine layer. Due to the larger sound velocity gradient within the mezzanine layer, the single-layer angular deviation is significantly higher than that of the upper and lower uniform layers.

[0045] The right figure shows the angular deviation sequence, with the target depth as the vertical axis and the cumulative angular deviation (the sum of the angular deviations of each layer during propagation from the water surface to that depth) as the horizontal axis. It can be seen that the cumulative angular deviation changes relatively little within the upper homogeneous layer; as the sound ray enters the depth range of the mezzanine, the cumulative angular deviation increases rapidly, and the slope of the curve increases significantly; after passing through the mezzanine and entering the lower homogeneous layer, the cumulative angular deviation tends to stabilize. In the deepest example, the cumulative angular deviation is approximately 8°.

[0046] The angular bias sequence is used to generate the confidence weight sequence. The larger the cumulative angular bias, the higher the observation uncertainty of the corresponding depth layer, and the lower the weight assigned to the data fusion. Through this mechanism, the physical error of acoustic ray propagation can be explicitly introduced into the fusion calculation process, making the final output fused velocity profile more consistent with real ocean environmental conditions.

[0047] Furthermore, step S4 includes the following steps: Step S41: Based on the depth distribution of the low-frequency corrected velocity field, construct candidate velocity profiles for each depth layer with undetermined velocity values. In one embodiment, candidate velocity profiles are constructed based on the depth distribution of the low-frequency corrected velocity field. The candidate velocity profiles have the same depth layer divisions as the low-frequency corrected velocity field, covering the entire water depth range from the water surface to the maximum detection depth. Each depth layer corresponds to a velocity value to be determined. The number of depth layers in the candidate velocity profiles is consistent with that in the low-frequency corrected velocity field, and the velocity value to be determined for each depth layer is determined by minimizing the target velocity.

[0048] Step S42: Sum the squared deviations of the high-frequency corrected velocity field and the candidate velocity profile within the depth range of the interlayer to form a high-frequency fitting term; when high-frequency velocity data is not available, the high-frequency fitting term is not calculated. In one embodiment, when high-frequency velocity data is available, the high-frequency corrected velocity field and the candidate velocity profile are compared layer by layer within the depth range of the cascade. The difference between the velocity value of each depth layer of the high-frequency corrected velocity field and the velocity value of the corresponding depth layer of the candidate velocity profile is calculated. The square of the difference of each layer is then summed over all depth layers within the depth range of the cascade to form a high-frequency fitting term.

[0049] The high-frequency fitting term corresponds to the first term in the cost function. Where H is the observation operator, The weight matrix is ​​constructed based on the ray tracking residual and the signal-to-noise ratio. For high-frequency correction of the flow velocity field, These are the weighting coefficients for the high-frequency fitting term. The diagonal elements are determined by the confidence weight values ​​of the corresponding depth layers in the confidence weight sequence. The smaller the confidence weight value, the lower the weight of that layer in the high-frequency fitting term. The role of the high-frequency fitting term is to force the candidate velocity profile to remain consistent with the high-frequency corrected velocity field within the cascade depth range, so that the fused velocity profile retains the fine velocity structure of the high-frequency corrected velocity field in that range. When high-frequency velocity data is unavailable, the high-frequency fitting term is not calculated.

[0050] The high-frequency fitting term is calculated only within the depth range of the strata. Its purpose is to force the candidate velocity profile to remain consistent with the high-frequency corrected velocity field within this range, so that the fused velocity profile retains the fine velocity structure of the high-frequency corrected velocity field in this interval. The high-frequency fitting term is not calculated when high-frequency velocity data is unavailable.

[0051] Step S43: Sum the squared deviations of the low-frequency corrected velocity field and the candidate velocity profile across the entire water depth range to form the low-frequency constraint term.

[0052] In one embodiment, the low-frequency corrected velocity field and the candidate velocity profile are compared layer by layer across the entire water depth range. The difference between the velocity value of the low-frequency corrected velocity field at each depth layer and the corresponding velocity value of the candidate velocity profile at the same depth layer is calculated. The squares of the differences at each layer are then summed across all depth layers from the water surface to the maximum detection depth to form the low-frequency constraint term. The low-frequency constraint term corresponds to the second term in the cost function. Where L is the smoothing downscaling operator, The diagonal weight matrix is ​​constructed based on the low-frequency corrected velocity field data quality (including signal-to-noise ratio, spatial coverage uniformity, and ray tracking residuals). To correct the flow velocity field at low frequencies, This represents the weighting coefficient for the low-frequency constraint term. The low-frequency constraint term is calculated across the entire water depth range. Its function is to ensure that the large-scale structure of the candidate velocity profile remains consistent with the low-frequency corrected velocity field covering the entire water depth, so that the merged velocity profile maintains the background velocity distribution reflected by the low-frequency corrected velocity field in the uniform layer outside the interlayer depth range.

[0053] Furthermore, step S4 also includes the following steps: Step S44: Use the regularization coefficients of each layer in the regularization coefficient sequence as multiplicative coefficients, and sum the squares of the velocity differences between adjacent depth layers of the candidate velocity profile in a weighted manner and summarize them over the entire water depth range to form a regularization term; For each pair of adjacent depth layers in the candidate velocity profile, the difference between the two velocity values ​​is calculated, squared, and multiplied by the regularization coefficient value of the corresponding depth layer in the regularization coefficient sequence to obtain the weighted squared velocity difference for that pair of adjacent layers. The weighted squared velocity differences for all pairs of adjacent depth layers across the entire depth range of the candidate velocity profile are summed to form the regularization term. The regularization term corresponds to the third term in the cost function. ,in The structure adaptive regularization operator is defined as follows: Γ(z) is the adaptive adjustment factor, calculated as follows: , The absolute value of the sound velocity gradient at each layer. To adjust the sensitivity coefficient, These are the weight coefficients for the regularization term. The sequence of regularization coefficients is generated in step S24 based on the layer-by-layer normalized sound velocity gradient as described above. Confirmed. At the core of the mezzanine. Too big A smaller regularization coefficient weakens the constraint on the vertical smoothness of candidate velocity profiles in this layer, allowing for larger vertical velocity variations to preserve strong velocity shear details; in a homogeneous layer... Approaching zero With a regularization coefficient close to 1, this layer exhibits the strongest smoothing constraint, effectively suppressing noise fluctuations within the uniform layer.

[0054] After step S44 is completed, the wavenumber spectrum constraint term also needs to be calculated. A Fourier transform is performed on the candidate velocity profile along the vertical direction to obtain the energy spectrum distribution of the candidate velocity profile in the vertical wavenumber domain. A reference energy spectrum was established based on the statistical characteristics of the vertical velocity energy spectrum in historical observation data. The reference energy spectrum decays according to a power-law in the high wavenumber region to reflect the physical and statistical laws of the ocean current field. The wavenumber spectrum constraint term corresponds to the fourth term in the cost function. ,in For vertical Fourier transform operators, For vertical wavenumber, The wavenumber domain weight matrix is... These are the weighting coefficients for the wavenumber spectrum constraint term. Larger values ​​are chosen in the high wavenumber region to suppress non-physical high wavenumber energy introduced by data sparsity or noise, while smaller values ​​are chosen in the low wavenumber region to maintain the energy of the large-scale background velocity structure. The role of the wavenumber spectrum constraint term is to ensure that the vertical energy spectrum of the fused velocity profile conforms to the statistical characteristics of ocean dynamic processes, and to avoid introducing non-physical spectral shapes into the fused results due to the spatial coverage of high-frequency velocity data being limited to the depth range of the strata.

[0055] Step S45: Take the sum of the high-frequency fitting term, the low-frequency constraint term, and the regularization term as the objective quantity, find the candidate velocity profile that minimizes the objective quantity in the solution set of the candidate velocity profile, and take it as the fused velocity profile; and record the square value of the single-layer deviation of the low-frequency constraint term at each depth layer when the objective quantity is at its minimum as the fusion analysis error of that layer. In one embodiment, the sum of the high-frequency fitting term, the low-frequency constraint term, the regularization term, and the wavenumber spectrum constraint term is used as the objective quantity, i.e., the complete cost function: The objective quantity is a function of the undetermined velocity values ​​at each depth layer of the candidate velocity profile, and is simultaneously constrained by the high-frequency corrected velocity field, the low-frequency corrected velocity field, the regularization coefficient sequence, and the reference energy spectrum. An iterative solution to the objective quantity is employed using the alternating direction multiplier method. This method decomposes the objective quantity into several independently solvable subproblems. In each iteration, the solutions to each subproblem are updated alternately, gradually approaching the global optimum. After iterative convergence, the objective quantity reaches its minimum value.

[0056] The candidate velocity profile corresponding to the minimum value of the target quantity is taken as the fused velocity profile. Within the depth range of the strata, the fused velocity profile retains the fine velocity structure of the high-frequency corrected velocity field due to the constraint of the high-frequency fitting term. In the homogeneous layer, the combined effect of the low-frequency constraint term and the regularization term maintains a smooth background distribution consistent with the low-frequency corrected velocity field. At the same time, due to the effect of the wavenumber spectrum constraint term, the vertical energy spectrum of the entire water depth conforms to the statistical characteristics of the ocean dynamic process, realizing a unified expression of the large-scale background field and the fine structure of the thin layer within the entire water depth range.

[0057] The squared value of the single-layer deviation of the low-frequency constraint term at each depth layer when the target quantity is at its minimum is denoted as the fusion analysis error for that layer. The fusion analysis error reflects the residual deviation between the fused velocity profile and the low-frequency corrected velocity field at each depth layer: within the interlayer depth range, the high-frequency fitting term pulls the candidate velocity profile toward the high-frequency corrected velocity field, causing the fused velocity profile to deviate from the low-frequency corrected velocity field, and the original value of the fusion analysis error for each layer is relatively large; in the homogeneous layer, the low-frequency constraint term dominates, making the fused velocity profile close to the low-frequency corrected velocity field, and the original value of the fusion analysis error for each layer is relatively small.

[0058] Step S46: Weight the fusion analysis error of each depth layer when the target quantity is minimized by the confidence weight values ​​of each layer in the confidence weight sequence, and summarize to form a fusion confidence field.

[0059] In one embodiment, the fusion analysis errors of each depth layer obtained in step S45 are weighted by region using the corresponding layer confidence weight values ​​in the confidence weight sequence. Using the depth range of the skip-layer interval as the boundary, the fusion analysis errors are divided by depth into fusion analysis errors of each layer in the skip-layer interval and fusion analysis errors of each layer in the non-skip-layer interval.

[0060] For each depth layer in the non-jumping layer interval, the confidence weight value of the corresponding layer in the confidence weight sequence is directly multiplied by the fusion analysis error of that layer to obtain the weighted error of each non-jumping layer. The acoustic ray curvature amplitude in the non-jumping layer interval is small, the cumulative angular deviation of each layer is small, the confidence weight value is close to 1, and the weighted error of the non-jumping layer is close to the original value of the fusion analysis error, reflecting the physical fact that the fusion result of the uniform layer has high reliability.

[0061] For each depth layer in the hopping layer interval, the fusion analysis error is obtained by dividing the corresponding layer's confidence weight value in the confidence weight sequence by the corresponding layer's regularization coefficient value in the regularization coefficient sequence, and then dividing again by the regularization coefficient value in the regularization coefficient sequence. At the core of the hopping layer, the confidence weight value is smaller due to the large cumulative angle deviation, and the regularization coefficient is smaller due to the large normalized sound velocity gradient. The two divisions significantly amplify the hopping layer weight error at this layer, quantitatively reflecting the physical fact that the acoustic ray bending amplitude is large, the smoothing constraint of the fusion solution is weak, and the uncertainty of the output result is high at the core of the hopping layer.

[0062] The weighted errors of non-interval layers and interval layers across the entire water depth are combined and arranged by depth to form a fused confidence field. The fused confidence field covers the entire water depth range of the fused velocity profile. The weighted fusion error of each layer has the largest value at the core of the interval layer and the smallest value at the homogeneous layer, which corresponds to the depth distribution trend of the normalized sound velocity gradient. It serves as a quantitative indicator of the reliability of the output results of each layer of the fused velocity profile and is passed to step S5 for labeling the confidence level of each layer.

[0063] Furthermore, step S46 includes: Using the depth range of the jump layer as the boundary, the fusion analysis error sequence is divided into the jump layer interval error sequence and the non-jump layer interval error sequence; For the fusion analysis error value of each layer in the non-jumping layer interval error sequence, multiply it by the corresponding layer confidence weight value in the confidence weight sequence to obtain the weighted error of each non-jumping layer. The fusion analysis error value of each layer in the step-layer interval error sequence is divided by the corresponding layer confidence weight value in the confidence weight sequence, and then divided by the corresponding layer regularization coefficient value in the regularization coefficient sequence to obtain the step-layer weighted error of each layer. The non-jump-layer weighted errors and jump-layer weighted errors of each layer at all water depths are combined and arranged by depth to form a fused confidence field.

[0064] In one embodiment, the obtained fusion analysis error is divided into two parts based on the depth range of the mezzanine: the fusion analysis errors of each layer with a depth falling within the interval from the upper boundary to the lower boundary of the mezzanine are summarized into a mezzanine interval error sequence, and the fusion analysis errors of each layer with a depth falling outside the mezzanine depth range are summarized into a non-mezzanine interval error sequence. The mezzanine interval error sequence and the non-mezzanine interval error sequence are combined to cover the entire depth range of the fused velocity profile, and the fusion analysis error of each layer belongs to only one sequence.

[0065] For each depth layer in the non-jumping layer interval error sequence, the fusion analysis error value of that layer is multiplied by the confidence weight value of the corresponding depth layer in the confidence weight sequence to obtain the non-jumping layer weighted error. Each depth layer in the non-jumping layer interval is located in a homogeneous layer. The cumulative angular deviation contribution of the jumping layer interval traversed by the sound ray from the water surface to this depth is limited, and the confidence weight value is close to 1. The non-jumping layer weighted error is close to the original value of the fusion analysis error, reflecting the physical fact that the sound ray bending amplitude is small in the homogeneous layer, the coordinate correction of the measurement unit is sufficient, and the fusion result has high reliability.

[0066] For each depth layer in the error sequence of the hierarchical interval, the fusion analysis error value of that layer is divided by the confidence weight value of the corresponding depth layer in the confidence weight sequence, and then divided by the regularization coefficient value of the corresponding depth layer in the regularization coefficient sequence to obtain the hierarchical weighted error of that layer. A larger normalized sound velocity gradient within the hierarchical interval leads to a larger cumulative angle deviation of the sound ray and a smaller confidence weight value. Using this as a divisor amplifies the hierarchical weighted error, reflecting the high residual uncertainty after coordinate correction due to the large curvature of the sound ray. Furthermore, it leads to a smaller regularization coefficient, which, again as a divisor, further amplifies the hierarchical weighted error, reflecting the weak smoothing constraint of this layer in step S44, insufficient restriction of the candidate velocity profile by the fusion solution, and a further increase in the uncertainty of the output result. The net effect of the two divisions is that the hierarchical weighted error at the core of the hierarchical layer is significantly greater than that at the boundary of the hierarchical layer, corresponding to the depth distribution trend of the normalized sound velocity gradient within the hierarchical layer.

[0067] The non-jump-layer weighted errors and jump-layer weighted errors of each layer across the entire water depth are combined and arranged by depth to form a fused confidence field. The fused confidence field covers the entire water depth range of the fused velocity profile and serves as a quantitative indicator of the confidence level of each layer in the fused velocity profile output results. It is output synchronously with the fused velocity profile and passed to step S5 for labeling the confidence level of each layer and determining the confidence level at the shear peak depth.

[0068] Furthermore, step S5 includes the following steps: Step S51: Extract the velocity values ​​of each depth layer within the depth range of the fused velocity profile, and arrange them according to depth to form a velocity sequence of the fused velocity interval; In one embodiment, velocity values ​​for each depth layer falling within the interval from the upper boundary to the lower boundary of the mezzanine are read from the fused velocity profile and arranged from shallowest to deepest, forming a mezzanine interval velocity sequence. The depth resolution of the mezzanine interval velocity sequence is consistent with that of the fused velocity profile within the mezzanine depth range. Due to the constraint of the high-frequency fitting term within the mezzanine depth range, the vertical resolution of the fused velocity profile is close to the level of the high-frequency corrected velocity field, and the mezzanine interval velocity sequence retains the fine velocity structure of the high-frequency corrected velocity field within this interval.

[0069] Step S52: Subtract the velocity values ​​of adjacent depth layers in the velocity sequence of the interlayer interval and divide by the corresponding depth interval to obtain the velocity shear value of each layer, thus forming the interlayer shear sequence; In one embodiment, for each pair of adjacent depth layers in the mezzanine velocity sequence, the difference between the shallow and deep velocity values ​​is divided by the depth interval between the two layers to obtain the velocity shear value between the adjacent layers, in units of seconds. The velocity shear values ​​are calculated for each pair of adjacent depth layers in the mezzanine velocity sequence, and then arranged and summarized by depth to form the mezzanine shear sequence.

[0070] The thermocline shear sequence reflects the rate of vertical velocity variation between different depth layers within the thermocline depth range, and is a quantitative characterization of the intensity of the dynamic processes within the thermocline. The system can derive characteristic parameters such as the maximum shear within the thermocline, and the thermocline shear sequence is the direct data source for this derived output.

[0071] Step S53: Take the velocity shear value with the largest absolute value in the shear sequence of the latitudinal layer as the maximum velocity shear value, and record the depth layer where the maximum velocity shear value is located as the shear peak depth; In one embodiment, the flow velocity shear values ​​at each depth layer in the thermocline shear sequence are traversed, and the absolute values ​​are compared. The flow velocity shear value with the largest absolute value is taken as the maximum flow velocity shear value, and the depth layer where this flow velocity shear value is located is recorded as the peak shear depth. The maximum flow velocity shear value and the peak shear depth together describe the intensity and location of the most drastic vertical change in flow velocity within the thermocline, and are one of the core output parameters of this invention for the analysis of ocean dynamic processes.

[0072] Step S54: Based on the range of weighted fusion error values ​​for each layer in the fusion confidence field, label the confidence level for each depth layer of the fusion velocity profile, and separately label the shear peak depth as either reliable or pending verification, and output the velocity profile product.

[0073] Of particular importance, step S54 includes the following steps: Step S541: Divide the difference between the minimum and maximum values ​​of the weighted fusion error of each layer at the full water depth in the fused confidence field into three segments, denoted as the high confidence interval, the medium confidence interval, and the low confidence interval. Step S542: Label the corresponding confidence level of each layer according to the high confidence interval, medium confidence interval and low confidence interval into which the weighted fusion error of each depth layer falls, forming a layer-by-layer confidence level sequence; Step S543: Read the weighted fusion error at the shear peak depth from the fusion confidence field and compare it with the upper limit of the high confidence interval; when the weighted fusion error at the shear peak depth falls into the high confidence interval, mark the maximum flow velocity shear value as the credible shear peak; when it falls into the medium confidence interval or the low confidence interval, mark the maximum flow velocity shear value as the shear peak to be verified. Step S544: Combine the fused velocity profile, layer-by-layer confidence level sequence, credible shear peak or shear peak to be verified and shear peak depth to output the velocity profile product.

[0074] In one embodiment, the difference between the minimum and maximum weighted fusion errors of all layers across the entire depth in the fused confidence field is used as the value range. This range is divided into three equal segments: the interval where the weighted fusion error does not exceed the minimum plus one-third of the range is designated as the high-confidence interval; the interval where the weighted fusion error exceeds the minimum plus one-third but does not exceed the minimum plus two-thirds of the range is designated as the medium-confidence interval; and the interval where the weighted fusion error exceeds the minimum plus two-thirds of the range is designated as the low-confidence interval. The division of confidence intervals is dynamically determined based on the value range of the fused confidence field itself, so that the confidence level division standard automatically adapts to the layer jump strength and acoustic ray curvature of the current expedition, avoiding the situation where a fixed threshold causes most depth layers to fall into the low-confidence interval and lose distinguishability in strong layer jump expeditions.

[0075] Based on the high-confidence, medium-confidence, or low-confidence intervals into which the weighted fusion error of each depth layer in the fused velocity profile falls, a corresponding confidence level is assigned to that layer, forming a layer-by-layer confidence level sequence. Layers at different depths in the homogeneous layer fall into the high-confidence interval due to their smaller weighted fusion error and are assigned a high-confidence level; layers at the core of the abrupt layer fall into the low-confidence interval due to their larger weighted fusion error and are assigned a low-confidence level; layers near the boundary of the abrupt layer are assigned a medium-confidence or low-confidence level depending on the magnitude of the weighted fusion error.

[0076] The weighted fusion error at the shear peak depth is read from the fused confidence field and compared with the upper limit of the high confidence interval. When the weighted fusion error at the shear peak depth falls within the high confidence interval, the maximum velocity shear value is marked as a reliable shear peak. When the weighted fusion error at the shear peak depth falls within the medium or low confidence interval, the maximum velocity shear value is marked as a shear peak to be verified. Since the shear peak depth where the maximum velocity shear value is located is usually near the megohmmeter core, and the normalized sound velocity gradient is the largest, the confidence weight is the smallest, and the regularization coefficient is the smallest at the megohmmeter core, the weighted fusion error of the fused confidence field is the largest at this depth. In most cases, the maximum velocity shear value is marked as a shear peak to be verified, prompting the user to further verify this key parameter.

[0077] By merging the fused velocity profile, layer-by-layer confidence level sequence, reliable shear peak or unverified shear peak, and shear peak depth, a velocity profile product with precise geographic coordinates and confidence level annotations is output. This product simultaneously provides a full-depth fused velocity profile and key dynamic parameters within the interlayer, and quantitatively annotates the confidence level of each layer and key parameters based on the fused confidence field, thus achieving the output goal of "a three-dimensional velocity profile product with precise geographic coordinates and high confidence."

Claims

1. A method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines, characterized in that, For multi-frequency ADCPs that include a low-frequency main ADCP and a high-frequency auxiliary ADCP, the following steps are included: Step S1: Use the low-frequency main ADCP to collect low-frequency velocity data and sound velocity profile data; when the gradient in the sound velocity profile data exceeds the preset trigger threshold, record the layer depth range and start the high-frequency auxiliary ADCP to collect high-frequency velocity data. Step S2: Calculate the normalized sound velocity gradient layer by layer for the sound velocity profile data, and generate a layer thickness sequence and a regularization coefficient sequence from the normalized sound velocity gradient of each layer. Step S3: Perform acoustic ray tracking on the low-frequency velocity data and high-frequency velocity data according to the layer thickness sequence, and output the low-frequency corrected flow velocity field, high-frequency corrected flow velocity field and angular deviation sequence; convert the angular deviation sequence into a confidence weight sequence; Step S4: Use the confidence weight sequence to weight the layer-by-layer deviation of the high-frequency corrected velocity field within the depth range to form a high-frequency fitting term; use the layer-by-layer deviation of the low-frequency corrected velocity field within the full water depth range to form a low-frequency constraint term; use the regularization coefficient sequence to weight the velocity difference between adjacent layers within the full water depth to form a regularization term; the velocity distribution corresponding to the minimum value of the sum of the three terms is used as the fused velocity profile; and output the fused analysis error of each depth layer at the minimum value, and summarize them to form a fused confidence field; Step S5: Extract the maximum velocity shear value within the depth range of the fused velocity profile, label the confidence level of each layer with the fused confidence field, and output the velocity profile product.

2. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines as described in claim 1, characterized in that, Step S1 includes the following steps: Step S11: Use the low-frequency main ADCP to transmit acoustic pulses to the full water depth and receive the echo signal to generate low-frequency velocity data. The full water depth acoustic pulse transmission means transmitting acoustic pulses to the water column range from the instrument position to the maximum detection depth of the target water body in one working cycle. Step S12: Use a sound velocity profiler to synchronously collect sound velocity values ​​at each layer of the entire water depth to form sound velocity profile data; Step S13: Subtract the sound velocity values ​​of adjacent depth layers in the sound velocity profile data and divide by the corresponding depth interval to obtain the local sound velocity gradient values ​​of each layer, forming a layer-by-layer sound velocity gradient.

3. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines as described in claim 2, characterized in that, Step S1 also includes the following steps: Step S14: Compare the gradient values ​​of each layer of the sound velocity gradient with the preset trigger threshold layer by layer. When the gradient values ​​of each layer in the continuous depth interval are not lower than the trigger threshold and the thickness of the depth interval is not less than 2 meters, the shallowest depth of the depth interval is recorded as the upper boundary of the jump layer and the deepest depth is recorded as the lower boundary of the jump layer, thus forming the jump layer depth range. Step S15: Send a start command to the high-frequency auxiliary ADCP, configure the observation window to be the interval between 0.5-2 meters above the upper boundary of the mezzanine and 0.5-2 meters below the lower boundary of the mezzanine, and collect high-frequency velocity data in this interval.

4. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines as described in claim 3, characterized in that, Step S2 includes the following steps: Step S21: Take the sound velocity value of the shallowest layer in the sound velocity profile data as the surface reference sound velocity; Step S22: Subtract the sound velocity values ​​of each adjacent depth layer in the sound velocity profile data, divide by the corresponding depth interval, and then divide by the surface reference sound velocity to obtain the normalized sound velocity gradient value of each layer. Take the absolute value and summarize to form the layer-by-layer normalized sound velocity gradient. Step S23: Determine the layer thickness of each layer based on the normalized sound velocity gradient values ​​of each layer in the layer-by-layer normalized sound velocity gradient, where the normalized sound velocity gradient value is inversely proportional to the layer thickness, thus obtaining the layer thickness sequence. Step S24: Determine the regularization coefficient for each layer of the same normalized sound velocity gradient value in the layer-by-layer normalized sound velocity gradient. The regularization coefficient is inversely proportional to the normalized sound velocity gradient value between 0 and 1, forming a regularization coefficient sequence.

5. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines according to claim 4, characterized in that, Step S3 includes the following steps: Step S31: Based on the layer thickness sequence, trace the sound path layer by layer for each distance unit of each beam in the low-frequency velocity data, and output the low-frequency corrected velocity field after coordinate correction of each measurement unit. Step S32: When high-frequency velocity data exists, trace the sound path layer by layer for each distance unit of each beam in the high-frequency velocity data using a layered thickness sequence, and output a high-frequency corrected flow velocity field.

6. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines according to claim 5, characterized in that, Step S3 also includes the following steps: Step S33: Divide the sound velocity profile data into layers of each depth according to the layer thickness sequence and record them as calculation layers. During the sound ray tracking process, calculate the absolute value of the difference between the incident angle of the sound ray entering the calculation layer and the exit angle after passing through the calculation layer for each calculation layer, and record it as the single-layer angular deviation of the calculation layer. Accumulate the single-layer angular deviations of all calculation layers that the sound ray passes through from the water surface to each target depth layer by layer to obtain the cumulative angular deviation at each target depth, and summarize them to form an angular deviation sequence. Step S34: Add the cumulative angular deviation of each layer in the angular deviation sequence to 1, use the sum as the denominator and 1 as the numerator, and use the resulting ratio as the confidence weight value to form a confidence weight sequence.

7. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines as described in claim 6, characterized in that, Step S4 includes the following steps: Step S41: Based on the depth distribution of the low-frequency corrected velocity field, construct candidate velocity profiles for each depth layer with undetermined velocity values. Step S42: Sum the squared deviations of the high-frequency corrected velocity field and the candidate velocity profile within the depth range of the interlayer to form a high-frequency fitting term; when high-frequency velocity data is not available, the high-frequency fitting term is not calculated. Step S43: Sum the squared deviations of the low-frequency corrected velocity field and the candidate velocity profile across the entire water depth range to form the low-frequency constraint term.

8. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines according to claim 7, characterized in that, Step S4 also includes the following steps: Step S44: Use the regularization coefficients of each layer in the regularization coefficient sequence as multiplicative coefficients, and sum the squares of the velocity differences between adjacent depth layers of the candidate velocity profile in a weighted manner and summarize them over the entire water depth range to form a regularization term; Step S45: Take the sum of the high-frequency fitting term, the low-frequency constraint term, and the regularization term as the objective quantity, find the candidate velocity profile that minimizes the objective quantity in the solution set of the candidate velocity profile, and take it as the fused velocity profile; and record the square value of the single-layer deviation of the low-frequency constraint term at each depth layer when the objective quantity is at its minimum as the fusion analysis error of that layer. Step S46: Weight the fusion analysis error of each depth layer when the target quantity is minimized by the confidence weight values ​​of each layer in the confidence weight sequence, and summarize to form a fusion confidence field.

9. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines according to claim 8, characterized in that, Step S46 includes: Using the depth range of the jump layer as the boundary, the fusion analysis error sequence is divided into the jump layer interval error sequence and the non-jump layer interval error sequence; For the fusion analysis error value of each layer in the non-jumping layer interval error sequence, multiply it by the corresponding layer confidence weight value in the confidence weight sequence to obtain the weighted error of each non-jumping layer. The fusion analysis error value of each layer in the step-layer interval error sequence is divided by the corresponding layer confidence weight value in the confidence weight sequence, and then divided by the corresponding layer regularization coefficient value in the regularization coefficient sequence to obtain the step-layer weighted error of each layer. The non-jump-layer weighted errors and jump-layer weighted errors of each layer at all water depths are combined and arranged by depth to form a fused confidence field.

10. The method for multi-frequency ADCP collaborative observation and high-resolution data fusion under the influence of deep thermoclines according to claim 9, characterized in that, Step S5 includes the following steps: Step S51: Extract the velocity values ​​of each depth layer within the depth range of the fused velocity profile, and arrange them according to depth to form a velocity sequence of the fused velocity interval; Step S52: Subtract the velocity values ​​of adjacent depth layers in the velocity sequence of the interlayer interval and divide by the corresponding depth interval to obtain the velocity shear value of each layer, thus forming the interlayer shear sequence; Step S53: Take the velocity shear value with the largest absolute value in the shear sequence of the latitudinal layer as the maximum velocity shear value, and record the depth layer where the maximum velocity shear value is located as the shear peak depth; Step S54: Based on the range of weighted fusion error values ​​for each layer in the fusion confidence field, label the confidence level for each depth layer of the fusion velocity profile, and separately label the shear peak depth as either reliable or pending verification, and output the velocity profile product.

Images