Positioning and leveling method, system and equipment of underwater leveling machine and underwater leveling machine

Abstract

The invention provides a positioning and leveling method, system and equipment of an underwater leveling machine and the underwater leveling machine, and relates to the technical field of ocean engineering. The method comprises the following steps: acquiring an original ranging data sequence of a fixed reference node; performing deviation compensation and correction on the original ranging data sequence to generate a target ranging value; performing time sequence filtering processing on the target distance measurement value of the distance measurement link between the underwater leveling machine and each fixed reference node to obtain a smooth distance measurement sequence; selecting data at multiple moments from the smooth ranging sequence to construct a time window input matrix, taking the time window input matrix as the input of the target deep learning model, and outputting the predicted position of the underwater leveling machine; and generating a control instruction based on the predicted position so as to control the underwater leveling machine to carry out positioning and leveling. According to the method, by combining multi-stage data processing and deep learning prediction, distance measurement errors and environmental interference are eliminated, so that the positioning and leveling accuracy of the underwater leveling machine can be improved.

Description

Technical Field

[0001] This application relates to the field of marine engineering technology, specifically to the positioning and leveling method, system, equipment, and underwater leveling machine. Background Technology

[0002] In typical marine engineering construction scenarios such as port and wharf foundation leveling, caisson / submerged tube installation foundation treatment, gravity structure foundation leveling, and foundation preparation before underwater placement of large-tonnage components, underwater leveling machines are typically used to push, scrape, and re-level the surface of riprap foundations to ensure that the foundation surface elevation meets design requirements and has good flatness. To achieve controllable, repeatable, and traceable construction quality, the leveling machine must maintain its position and attitude information in the construction coordinate system in real time. This positioning information is used not only to guide the leveling machine along the predetermined path and control the elevation and inclination angle of the scraper / leveling beam, but also to write the information of "actual working area—leveling depth—residual high / shallow points" into the digital construction platform in real time, forming a closed loop for quality acceptance. Therefore, positioning accuracy and stability have become the core foundation for the intelligent and automated construction capabilities of underwater leveling equipment.

[0003] Current positioning technologies heavily rely on ranging signals, which are exceptionally vulnerable in complex underwater environments. Underwater leveling areas typically contain numerous hard reflective targets such as rock piles, guide piles, caisson edges, and mooring structures. These structures cause severe multipath reflections and obstructions to the underwater acoustic signal, leading to drastic fluctuations or even jumps in instantaneous ranging values. Simultaneously, strong noise introduced by local currents, hoisting operations, and rockfalls further contributes to the intermittent degradation of ranging signal quality. This high sensitivity to the environment renders the raw data used for positioning calculations inherently unreliable. Furthermore, existing positioning and leveling systems often suffer from robustness deficiencies. Most systems employ a single-source positioning approach: "underwater acoustic ranging + geometric calculation" or "IMU calculation + manual correction." If the primary signal source (such as the underwater acoustic ranging path) is obstructed by the leveling machine's legs or temporary work hooks and caisson structures, the system loses reliable coordinate reference for that period. Although inertial measurement units (IMUs) can perform short-term displacement calculations to attempt compensation, they inevitably have cumulative drift errors. Over time, the calculated position will deviate significantly from the actual position, making it impossible to support accurate positioning over long periods on its own. Therefore, existing technologies suffer from low accuracy in positioning and leveling of underwater leveling machines. Summary of the Invention

[0004] Based on this, this application provides a positioning and leveling method, system, equipment, and underwater leveling machine. By combining multi-stage data processing and deep learning prediction, ranging errors and environmental interference are eliminated, thereby improving the positioning and leveling accuracy of the underwater leveling machine.

[0005] In a first aspect, embodiments of this application provide a positioning and leveling method for an underwater leveling machine, comprising: The process involves: acquiring the original ranging data sequence from fixed reference nodes; performing deviation compensation and correction on the original ranging data sequence to generate target ranging values; using the target ranging values ​​to correct each original ranging data sequence; performing time-series filtering on the target ranging values ​​of the ranging link between the underwater leveling machine and each fixed reference node to obtain a smoothed ranging sequence; selecting data from multiple moments in the smoothed ranging sequence to construct a time window input matrix; using the time window input matrix as input to a target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model; and generating control commands based on the predicted position to control the underwater leveling machine for positioning and leveling.

[0006] In some embodiments, the step of selecting data from multiple time points in the smoothed ranging sequence to construct a time window input matrix, and using the time window input matrix as input to a target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model, includes: Using a fixed time window with multiple sampling periods, the target smoothed ranging values ​​at the most recent times are sequentially extracted from the smoothed ranging sequence corresponding to each fixed reference node. The input tensor, whose dimension is the product of the target smoothed ranging values ​​at the multiple times and the fixed reference node, is constructed based on the target smoothed ranging values ​​at the multiple times. In the input tensor, each row represents the distance from the underwater leveling machine to all the fixed reference nodes at the same time, and each column represents the ranging sequence of a fixed reference node at multiple consecutive times. Temporal features are extracted from the input tensor, and the three-dimensional coordinates of the underwater leveling machine in the global coordinate system are regressed and output. The three-dimensional coordinates represent the predicted position of the underwater leveling machine.

[0007] In some embodiments, the step of performing deviation compensation and correction on the original ranging data sequence to generate the target ranging value includes: A deviation regression model is established using static calibration data; based on the deviation regression model, the fixed deviation value and proportional error of the ranging link between the underwater leveling machine and each fixed reference node are determined, and the target ranging value is generated based on the fixed deviation value and the proportional error.

[0008] In some embodiments, performing time-series filtering on the target ranging values ​​of the ranging link between the underwater leveling machine and each fixed reference node to obtain a smooth ranging sequence includes: establishing a linear state-space model corresponding to each of the original ranging data sequences, predicting the true distance value through the recursive prediction and update process of the linear state-space model, and generating the smooth ranging sequence based on the true distance value.

[0009] In some embodiments, establishing a linear state-space model corresponding to each of the original ranging data sequences, predicting the true distance value through the recursive prediction and update process of the linear state-space model, and generating the smoothed ranging sequence based on the true distance value includes: establishing a linear state-space model corresponding to each of the original ranging data sequences; the state variable of the linear state-space model is the true distance value to be predicted, and the observation variable is the target ranging value at the current time; based on the posterior estimate of the state variable at the previous time and the corresponding covariance matrix, combined with a preset process noise covariance, calculating the prior estimate of the state variable at the current time and the corresponding prior... The covariance matrix is ​​used for prediction. A prediction process is performed by constructing an innovation based on the difference between the observed variable at the current time and the predicted observation value obtained from the prior estimate, and calculating the Kalman gain based on the prior covariance matrix and a preset observation noise covariance. An update process is performed by weighting the innovation based on the Kalman gain to correct the prior estimate of the state variable, obtaining the posterior estimate of the state variable at the current time, and updating the posterior covariance matrix corresponding to the posterior estimate. The prediction and update processes are iteratively executed. Upon completion of the update, the posterior estimate of the state variable obtained at each time step is output as the smoothed ranging sequence.

[0010] In some embodiments, the linear state-space model includes:

[0011] The formulas for calculating Kalman gain include:

[0012] in, These are the prior estimates of the state variables at the current moment. This represents the posterior estimate of the state variable from the previous time step. For process noise, For the observed variables at the current moment, To observe the noise, For Kalman gain, The prior covariance matrix, For the observation matrix, Let be the transpose of the observation matrix. This is the preset observation noise covariance.

[0013] In some embodiments, generating control commands based on the predicted position to control the underwater leveling machine for positioning and leveling includes: automatically correcting the gait propulsion command when the predicted position indicates that the real-time pose of the underwater leveling machine deviates from the target trajectory; and adjusting the leveling tool holder height and attitude of the underwater leveling machine according to the actual elevation deviation.

[0014] Secondly, embodiments of this application provide a positioning and leveling system for an underwater leveling machine, comprising: The multi-reference ranging module is used to acquire the original ranging data sequence of fixed reference nodes; The data preprocessing and synchronization module is used to perform deviation compensation and correction on the original ranging data sequence to generate the target ranging value; The noise reduction module is used to establish a linear state-space model for each of the original ranging data sequences based on the target ranging, predict the true distance value through the recursive prediction and update process of the linear state-space model, and generate a smooth ranging sequence. The temporal deep learning localization module is used to select data from multiple moments in the smoothed ranging sequence to construct a time window input matrix, and use the time window input matrix as the input of the target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model. The control module is used to generate control commands based on the predicted position to control the underwater leveling machine to perform positioning and leveling.

[0015] Thirdly, embodiments of this application provide a computer device, the computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the methods described above.

[0016] Fourthly, embodiments of this application also provide an underwater leveling machine, which is equipped with the positioning and leveling system of the underwater leveling machine described above.

[0017] Compared with the prior art, the beneficial effects of this application are as follows: by acquiring the original ranging data sequence of fixed reference nodes, the original sequence is subjected to deviation compensation and correction to generate target ranging values ​​to eliminate systematic errors and random noise. By performing time-series filtering on the target ranging values ​​of each ranging link, a smooth ranging sequence is obtained to suppress short-term fluctuations and outliers, thereby enhancing the stability and continuity of the data. Data from multiple moments are selected from the smooth sequence to construct a time window input matrix, capturing the dynamic temporal characteristics of the ranging data. This matrix is ​​then input into the target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system. Based on the control commands generated from the accurate predicted position, the underwater leveling machine is guided to make fine adjustments, thereby improving the positioning and leveling accuracy of the underwater leveling machine. Attached Figure Description

[0018] Figure 1 A schematic diagram illustrating the steps of the positioning and leveling method for the underwater leveling machine provided in this application embodiment. Detailed Implementation

[0019] The present application will now be described in further detail with reference to experimental examples and specific embodiments. However, this should not be construed as limiting the scope of the subject matter of the present application to the following embodiments. All technologies implemented based on the content of the present application fall within the scope of protection of the present application.

[0020] In the description of the embodiments of this application, technical terms such as "first" and "second" only distinguish one entity or operation from another, and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary or secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0021] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0022] During the research process, the applicant found that the positioning methods commonly used in the industry at present mainly include: (1) relative positioning based on underwater acoustic ranging or acoustic reference stations; (2) attitude and displacement estimation based on inertial measurement units (IMUs); and (3) manual correction based on construction vessels, guide piles, or shoreline stations. These methods can basically meet the requirements under open deep water conditions, but in real leveling conditions near the shore, in shallow water, in harbor basins, in riprap piles, and in densely packed large steel components, there are significant unstable factors: underwater acoustic signals generate multipath reflections and blockages between complex structures, and instantaneous ranging values ​​will jump; local water flow, hoisting operations, and riprap fall will introduce strong noise, causing the effectiveness of the ranging signal to decrease intermittently; the IMU estimation method has unavoidable cumulative drift, which will significantly deviate from the actual position over time. These factors cause the positioning calculation of the leveling machine to often exhibit jitter, drift, or short-term distortion, making it difficult to support centimeter-level elevation control and reproducible path tracking. In addition, traditional positioning methods usually use the "single frame ranging result at the current moment" directly for geometric inverse solution of position, lacking the ability to model continuous time series. Once a single abnormal ranging occurs, the solution result will deviate from the true value instantly, and the continuity and smoothness of the trajectory output cannot be guaranteed.

[0023] Although existing technologies have attempted to improve positioning accuracy through multi-reference ranging, filtering algorithms, and neural network regression, the following problems still exist in the specific scenario of underwater leveling: 1. Ranging signals are highly sensitive to the environment, and positioning results are easily interfered with. Underwater leveling areas are often accompanied by a large number of hard reflective targets such as riprap piles, guide piles, caisson corners, mooring structures, and hoisting steel components, and the work area is usually close to the shoreline or harbor boundary. External global navigation signals (such as satellite positioning) are unavailable, and local ranging signals are blocked, subject to multipath reflection, and noise pollution. Under these conditions, the raw distance values ​​obtained by single underwater acoustic ranging or short-range ranging will fluctuate drastically or even jump. 2. Positioning systems based on a single sensor or a single algorithm lack robustness. Most existing positioning systems are based on "underwater acoustic ranging + geometric calculation" or "IMU calculation + manual correction," which belongs to the single-source positioning approach. If the main signal source is blocked in a local area (e.g., the leveling machine's pile legs, hooks, or caisson structure temporarily block the ranging path) or its quality deteriorates, the system will lose reliable coordinates during that period; and inertial calculations can only compensate for this in a short time, and will still inevitably drift over time.

[0024] Therefore, this application provides a positioning and leveling method for an underwater leveling machine. By combining multi-stage data processing and deep learning prediction, it eliminates ranging errors and environmental interference, thereby improving the positioning and leveling accuracy of the underwater leveling machine. Please refer to... Figure 1 , Figure 1This diagram illustrates the steps of the positioning and leveling method for an underwater leveling machine provided in this application embodiment. The steps of the positioning and leveling method for the underwater leveling machine may include: S1. Obtain the original ranging data sequence of the fixed reference node.

[0025] S2. Perform deviation compensation and correction on the original ranging data sequence to generate target ranging values; the target ranging values ​​are used to correct each original ranging data sequence.

[0026] S3. Perform time-series filtering on the target ranging values ​​of the ranging link between the underwater leveling machine and each fixed reference node to obtain a smooth ranging sequence.

[0027] S4. Select data from multiple moments in the smoothed ranging sequence to construct a time window input matrix. Use the time window input matrix as input to the target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model.

[0028] S5. Generate control commands based on the predicted position to control the underwater leveling machine for positioning and leveling.

[0029] In this embodiment, a fixed reference node refers to a ranging beacon with known three-dimensional coordinates that is pre-deployed in the construction area. That is, its spatial position is fixed in the construction coordinate system, providing a stable spatial reference for the moving screed machine. For example, in a harbor basin bed screed project, three or more such nodes can be set on the bank, guide piles, or existing structures. The raw ranging data sequence refers to a series of time-ordered distance observations obtained periodically by the moving tag node carried by the screed machine and each fixed reference node, based on the principle of Double-Sided Two-Way Ranging (DS-TWR) using Ultra Wide Band (UWB) signals. In other words, it is the raw distance data set of each ranging link before processing, containing noise and systematic errors.

[0030] The process of bias compensation and correction for the original ranging data sequence refers to the following steps during the static calibration phase: using control points with known coordinates, and assuming the underwater leveling machine is positioned at each control point, the original ranging values ​​between the underwater leveling machine and each fixed reference node are collected. The collected original data is compared with the actual geometric distance calculated based on the known coordinates, and linear regression or least squares fitting methods are used to determine the unique fixed bias compensation amount and proportional coefficient for each ranging link. In actual construction, based on the calibration parameters, the real-time acquired original ranging data sequence is calculated and compensated in real time to obtain the target ranging value that eliminates system errors. This refers to the calibrated distance data output after the aforementioned bias compensation and correction, which is closer to the actual geometric distance.

[0031] The ranging link is an independent ranging channel formed between the underwater leveling machine and a single fixed reference node. Each reference point corresponds to a data stream, and multiple links collectively provide spatial constraints. Temporal filtering specifically involves applying the Kalman filter algorithm to recursively estimate the time-series data. This is achieved by constructing a linear state-space model, performing prediction and update iterations, and suppressing random noise. The smoothed ranging sequence is the continuous and stable distance estimate output after filtering; its values ​​represent an optimized approximation of the true distance, for example, by smoothing the curve after removing instantaneous jumps.

[0032] "Multiple time points" refers to two or more consecutive sampling periods, such as T time points, to capture temporal dynamics. The time window input matrix is ​​a two-dimensional array formed by stacking the smoothed ranging data of the most recent T time points row by row, i.e., its dimension is T×M, where M represents the number of fixed reference nodes. The target deep learning model can extract long-term dependency features through its gating mechanism and output the predicted position of the underwater leveling machine in the global coordinate system. The global coordinate system refers to the unified spatial reference system of the construction area, i.e., its coordinate origin is fixed, and it is used to describe the position of the underwater leveling machine. The output predicted position of the target deep learning model can be the three-dimensional coordinate estimate of the underwater leveling machine in the global coordinate system, i.e., the [X, Y, Z] vector, representing the real-time pose of the underwater leveling machine.

[0033] In some embodiments, the underwater leveling machine is equipped with a mobile tag node. During step S1 of the above method, a complete signal exchange occurs between the mobile tag node and a fixed reference node. One node transmits a UWB pulse signal, and the other, upon receiving the signal, transmits a response pulse signal back to the initiator after a known, extremely short processing delay. By recording the round-trip propagation time of the pulse signal from transmission to reception and subtracting the known processing delays within both nodes, the true one-way propagation time of the electromagnetic wave in space can be calculated. Based on this precisely measured propagation time, combined with the electromagnetic wave propagation speed corrected for the underwater environment, the instantaneous straight-line distance between the underwater leveling machine and the fixed reference node can be obtained. The entire ranging process is periodically repeated between the underwater leveling machine and each fixed reference node. Each successful ranging interaction generates an initial distance observation value. All these distance observation values, arranged chronologically from different reference nodes, together constitute the initial ranging data sequence.

[0034] In some embodiments, the steps of the stage of bias compensation and correction of the original ranging data sequence may include: A deviation regression model is established using static calibration data; based on the deviation regression model, the fixed deviation value and proportional error of the ranging link between the underwater leveling machine and each fixed reference node are determined, and the target ranging value is generated based on the fixed deviation value and the proportional error.

[0035] Static calibration data refers to the collection of all raw distance measurement data and corresponding true geometric distance values ​​collected during the system preparation phase before construction. This is achieved by sequentially placing an underwater screed (equipped with moving distance measuring tag nodes) statically at several control points within the construction area whose spatial three-dimensional coordinates have been precisely measured. The process of establishing the deviation regression model involves placing the underwater screed sequentially at several control points with known spatial coordinates after setting up fixed reference nodes in the construction area. At each control point, a series of raw distance measurement data are collected between the screed and each fixed reference node through bilateral, bidirectional distance measurement. By comparing these measured raw distance values ​​with the true geometric distances calculated based on the precise coordinates of the control points and reference nodes, a "distance-true value" deviation regression model can be established for each independent "screed-reference node" distance measurement link.

[0036] For each newly acquired raw ranging value, the model parameters of the corresponding link are called to calculate and automatically deduct the calibrated fixed deviation value, and then scaled and corrected according to a determined scaling factor. The output result obtained after this compensation mechanism is the target ranging value, which, compared with the original data, has its systematic error effectively suppressed and is closer to the true geometric distance.

[0037] The target ranging values ​​obtained after correction of all reference node links are integrated and aligned according to a unified timestamp to form a standardized "distance-time" matrix. Each row of this matrix represents the set of target ranging values ​​from the underwater leveler to all reference nodes at the same time, providing complete spatial geometric constraint information; each column represents the sequence of target ranging values ​​of a reference node at different times, characterizing the temporal dynamics of the underwater leveler's motion.

[0038] In some embodiments, time-series filtering of the target ranging value includes: constructing a dedicated linear state-space model for each independent ranging time series, and performing a recursive prediction and update process through the model to dynamically estimate the true distance value and generate a high-quality smooth ranging sequence. Step S3 may include: establishing a linear state-space model corresponding to each of the original ranging data sequences, predicting the true distance value through the recursive prediction and update process of the linear state-space model, and generating the smooth ranging sequence based on the true distance value.

[0039] The specific implementation process includes: Establish a linear state-space model corresponding to each of the original ranging data sequences; the state variable of the linear state-space model is the true distance value to be predicted, and the observation variable is the target ranging value at the current moment.

[0040] Based on the posterior estimate of the state variable at the previous time step and the corresponding covariance matrix, combined with the preset process noise covariance, the prior estimate of the state variable at the current time step and the corresponding prior covariance matrix are calculated.

[0041] The prediction process is performed by constructing new information based on the difference between the observed variable at the current time and the observed predicted value obtained from the prior estimate, and calculating the Kalman gain based on the prior covariance matrix and the preset observation noise covariance.

[0042] The update process involves weighting the innovation based on the Kalman gain to correct the prior estimate of the state variable, obtaining the posterior estimate of the state variable at the current time, and updating the posterior covariance matrix corresponding to the posterior estimate. The prediction process and the update process are executed iteratively. When the update is completed, the posterior estimate of the state variable obtained at each time step is output as the smoothed ranging sequence.

[0043] The linear state-space model corresponding to each original ranging data sequence includes a state equation and an observation equation. The linear state-space model is as follows:

[0044] These are the prior estimates of the state variables at the current moment. This represents the posterior estimate of the state variable from the previous time step. This represents process noise, characterizing the inherent uncertainty of the model itself. For the observed variables at the current moment, To observe noise. At each new moment, the processing comprises two phases: prediction and update. The prediction process begins by utilizing the state information from the previous moment. Specifically, based on the posterior estimates of the state variables from the previous moment and their corresponding posterior covariance matrix, combined with a pre-defined process noise covariance, the prior estimates of the state variables at the current moment are calculated through mapping of the state equation. and its corresponding prior covariance matrix .

[0045] The prediction process is then executed, using the actual observation value at the current moment, i.e., the target distance value. This is used to correct the prediction results. The innovation is calculated, which is the difference between the current observation and the observed prediction obtained from the prior estimate (mapped through the observation matrix H, which is usually set to 1 in this model).

[0046] Based on the prior covariance matrix Calculate the Kalman gain with the preset observation noise covariance R. The calculation formula is as follows:

[0047] For Kalman gain, The prior covariance matrix, For the observation matrix, Let be the transpose of the observation matrix. The pre-defined observation noise covariance is used. Kalman gain characterizes the extent to which current observations should be trusted over model predictions when correcting state estimates. By weighting the innovation using Kalman gain, the prior estimates of the state variables are corrected to obtain more accurate posterior estimates of the state variables at the current time, and their posterior covariance matrix is ​​updated simultaneously. The posterior estimate is the optimal estimate of the true distance at the current time.

[0048] By iteratively executing the above prediction and update process, the posterior estimate of the state variable at each time step, corrected for the observed values, is output, thus forming the smooth ranging sequence. This sequence dynamically estimates the true distance, effectively mitigating random errors caused by multipath reflections, acoustic delays, and instantaneous jumps. The effect of this is to output a smooth, continuous, and reliable smooth ranging sequence. .

[0049] In some embodiments, the step of selecting data from multiple time points in the smoothed ranging sequence to construct a time window input matrix, and using the time window input matrix as input to a target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model, includes: Using a fixed time window of multiple sampling periods, the target smooth ranging values ​​at multiple recent moments are sequentially extracted from the smooth ranging sequence corresponding to each fixed reference node. The input tensor of the target smooth ranging values ​​at multiple moments is constructed based on the dimension of multiple moments multiplied by the fixed reference node. In the input tensor, each row represents the distance from the underwater leveling machine to all the fixed reference nodes at the same moment, and each column represents the ranging sequence of a fixed reference node at multiple consecutive moments. Temporal features are extracted from the input tensor, and the three-dimensional coordinates of the underwater leveling machine in the global coordinate system are regressed and output; the three-dimensional coordinates represent the predicted position of the underwater leveling machine.

[0050] The temporal window construction can select the denoised ranging data from the most recent T time points, and stack them into a T×M input tensor using fixed-length time windows (T sampling periods):

[0051] The target deep learning model can be a Long Short-Term Memory (LSTM) network. The input tensor is processed by the LSTM network through a multi-layer gating structure (forget gate, input gate, output gate), and the output is the predicted position of the leveling machine in the global coordinate system.

[0052] Optimize the loss function through training:

[0053] By continuously adjusting network parameters, high-precision temporal positioning mapping is achieved. Among these, It represents the actual three-dimensional coordinates of the underwater leveling machine in the global coordinate system at time t. It is the predicted position of the underwater leveling machine by the target deep learning model. This represents the loss function value.

[0054] The input tensor is subjected to temporal feature extraction using a target deep learning model, and the three-dimensional coordinates of the underwater leveler in the global coordinate system are regressed to represent the predicted position of the underwater leveler. By sending this predicted position to the motion / attitude controller of the underwater leveler, the gait propulsion command can be automatically corrected when the predicted position indicates that the real-time pose of the underwater leveler deviates from the target trajectory. This corrects the gait propulsion trajectory, adjusts the leveling height, roll angle, and pitch angle to ensure the target elevation, adjusts the leveling tool holder height and attitude of the underwater leveler according to the actual elevation deviation, and colors the "leveled area" in real time on the digital twin interface, overlaying it with the design elevation difference map to create a quality closure loop.

[0055] In the above implementation process, the original ranging data sequence of the fixed reference node is acquired, and the original sequence is subjected to deviation compensation and correction to generate target ranging values ​​to eliminate systematic errors and random noise. By performing time-series filtering on the target ranging values ​​of each ranging link, a smooth ranging sequence is obtained to suppress short-term fluctuations and outliers, thereby enhancing the stability and continuity of the data. Data from multiple moments are selected from the smooth sequence to construct a time window input matrix, which captures the dynamic temporal characteristics of the ranging data. This matrix is ​​then input into the target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system. Based on the control commands generated from the accurate predicted position, the underwater leveling machine is guided to make fine adjustments, thereby improving the positioning and leveling accuracy of the underwater leveling machine.

[0056] Based on the same concept, embodiments of this application also provide a positioning and leveling system for an underwater leveling machine, which may include: The multi-reference ranging module is used to acquire the original ranging data sequence of fixed reference nodes; The data preprocessing and synchronization module is used to perform deviation compensation and correction on the original ranging data sequence to generate the target ranging value; The noise reduction module is used to establish a linear state-space model for each of the original ranging data sequences based on the target ranging, predict the true distance value through the recursive prediction and update process of the linear state-space model, and generate a smooth ranging sequence. The temporal deep learning localization module is used to select data from multiple moments in the smoothed ranging sequence to construct a time window input matrix, and use the time window input matrix as the input of the target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model. The control module is used to generate control commands based on the predicted position to control the underwater leveling machine to perform positioning and leveling.

[0057] It should be understood that when the various modules of the system provided in the above embodiments are working, the division of each functional module in the above description is only used as an example. In actual applications, the above functions can be assigned to different functional modules as needed. That is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0058] The functional modules in the above embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of the embodiments of this application.

[0059] Based on the same concept, embodiments of this application also provide a computer device, which may include a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the methods described above.

[0060] Based on the same concept, this application also provides an underwater leveling machine, which is equipped with the positioning and leveling system described above.

[0061] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A positioning and leveling method for an underwater leveling machine, characterized in that, include: Obtain the raw ranging data sequence of the fixed reference node; The original ranging data sequence is subjected to deviation compensation and correction to generate the target ranging value; The target ranging value is used to correct each of the original ranging data sequences; The target ranging values ​​of the ranging link between the underwater leveling machine and each fixed reference node are subjected to time-series filtering to obtain a smooth ranging sequence; Data from multiple moments are selected from the smoothed ranging sequence to construct a time window input matrix. The time window input matrix is ​​used as the input to the target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model. Based on the predicted position, control commands are generated to control the underwater leveling machine to perform positioning and leveling.

2. The positioning and leveling method for the underwater leveling machine according to claim 1, characterized in that, The step of selecting data from multiple moments in the smoothed ranging sequence to construct a time window input matrix, and using the time window input matrix as input to the target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model, includes: Using a fixed time window of multiple sampling periods, the target smooth ranging values ​​at multiple recent moments are sequentially extracted from the smooth ranging sequence corresponding to each fixed reference node. The input tensor of the target smooth ranging values ​​at multiple moments is constructed based on the dimension of multiple moments multiplied by the fixed reference node. In the input tensor, each row represents the distance from the underwater leveling machine to all the fixed reference nodes at the same moment, and each column represents the ranging sequence of a fixed reference node at multiple consecutive moments. Temporal features are extracted from the input tensor, and the three-dimensional coordinates of the underwater leveling machine in the global coordinate system are regressed and output; the three-dimensional coordinates represent the predicted position of the underwater leveling machine.

3. The positioning and leveling method for the underwater leveling machine according to claim 1, characterized in that, The step of performing deviation compensation and correction on the original ranging data sequence to generate the target ranging value includes: A deviation regression model was established using static calibration data; Based on the deviation regression model, the fixed deviation value and proportional error of the ranging link between the underwater leveling machine and each fixed reference node are determined, and the target ranging value is generated based on the fixed deviation value and the proportional error.

4. The positioning and leveling method for the underwater leveling machine according to claim 1, characterized in that, The step of performing time-series filtering on the target ranging values ​​of the ranging link between the underwater leveling machine and each fixed reference node to obtain a smooth ranging sequence includes: A linear state-space model is established for each of the original ranging data sequences. The true distance value is predicted through the recursive prediction and update process of the linear state-space model, and the smooth ranging sequence is generated based on the true distance value.

5. The positioning and leveling method for the underwater leveling machine according to claim 4, characterized in that, The step of establishing a linear state-space model corresponding to each of the original ranging data sequences, predicting the true distance value through the recursive prediction and update process of the linear state-space model, and generating the smoothed ranging sequence based on the true distance value includes: Establish a linear state-space model corresponding to each of the original ranging data sequences; the state variable of the linear state-space model is the true distance value to be predicted, and the observation variable is the target ranging value at the current moment. Based on the posterior estimate of the state variable at the previous time step and the corresponding covariance matrix, combined with the preset process noise covariance, the prior estimate of the state variable at the current time step and the corresponding prior covariance matrix are calculated. The prediction process is performed by constructing information based on the difference between the observed variable at the current time and the observed predicted value obtained from the prior estimate, and calculating the Kalman gain based on the prior covariance matrix and the preset observation noise covariance. The update process involves weighting the innovation based on the Kalman gain to correct the prior estimate of the state variable, obtaining the posterior estimate of the state variable at the current time, and updating the posterior covariance matrix corresponding to the posterior estimate. The prediction process and the update process are executed iteratively. When the update is completed, the posterior estimate of the state variable obtained at each time step is output as the smoothed ranging sequence.

6. The positioning and leveling method for the underwater leveling machine according to claim 5, characterized in that, The linear state-space model includes: The formulas for calculating Kalman gain include: in, These are the prior estimates of the state variables at the current moment. This represents the posterior estimate of the state variable from the previous time step. For process noise, For the observed variables at the current moment, To observe the noise, For Kalman gain, The prior covariance matrix, For the observation matrix, Let be the transpose of the observation matrix. This is the preset observation noise covariance.

7. The positioning and leveling method for the underwater leveling machine according to claim 1, characterized in that, The step of generating control commands based on the predicted position to control the underwater leveling machine to perform positioning and leveling includes: When the predicted position indicates that the real-time pose of the underwater leveling machine deviates from the target trajectory, the walking propulsion command is automatically corrected. The height and attitude of the underwater leveling machine's leveling tool holder are adjusted according to the actual elevation deviation.

8. A positioning and leveling system for an underwater leveling machine, characterized in that, include: The multi-reference ranging module is used to acquire the original ranging data sequence of fixed reference nodes; The data preprocessing and synchronization module is used to perform deviation compensation and correction on the original ranging data sequence to generate the target ranging value; The noise reduction module is used to establish a linear state-space model for each of the original ranging data sequences based on the target ranging, predict the true distance value through the recursive prediction and update process of the linear state-space model, and generate a smooth ranging sequence. The temporal deep learning localization module is used to select data from multiple moments in the smoothed ranging sequence to construct a time window input matrix, and use the time window input matrix as the input of the target deep learning model to output the predicted position of the underwater leveling machine in the global coordinate system based on the target deep learning model. The control module is used to generate control commands based on the predicted position to control the underwater leveling machine to perform positioning and leveling.

9. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the method as claimed in any one of claims 1 to 7.

10. An underwater leveling machine, characterized in that, The underwater leveling machine is equipped with the positioning and leveling system of the underwater leveling machine as described in claim 8.

Images