Attitude control method, system and equipment of underwater leveling machine and underwater leveling machine

Abstract

The invention provides an attitude control 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: establishing a dynamic model; constructing a physical information neural network based on the dynamic model; constructing a digital twinborn body of the leveling machine, and performing simulation verification on the attitude error prediction result in the digital twinborn body; based on a simulation verification result, taking control parameters of the underwater leveling machine as optimization variables, performing parallel optimization on a plurality of control targets of the underwater leveling machine, and generating a control solution set; and determining a target control parameter combination from the control solution set, and generating a control instruction for controlling the underwater leveling machine to perform attitude adjustment based on the target control parameter combination. According to the method, the reliability of attitude control of the underwater leveling machine can be improved through prediction of the integrated physical information neural network, simulation verification of the digital twin and multi-objective optimization.

Description

Technical Field

[0001] This application relates to the field of marine engineering technology, specifically to attitude control methods, systems, equipment, and underwater leveling machines. Background Technology

[0002] In underwater engineering projects such as seabed leveling, gravity-fed wharf construction, and subsea pipeline laying, the attitude control of the leveling machine is crucial for ensuring the flatness of the seabed and the accuracy of construction. Traditional underwater leveling machines typically employ manual remote control or proportional-integral-derivative (PID) control based on fixed parameters, adjusting the thrust of hydraulic cylinders to achieve multi-degree-of-freedom attitude adjustment of the leveling frame. However, the marine environment exhibits highly nonlinear and strongly coupled characteristics. During operation, the leveling machine is subject to complex hydrodynamic disturbances, uneven seabed reaction forces, and the inertia of the device itself, leading to significant changes in the system's dynamic characteristics. Existing control strategies are mostly based on simplified linear models or empirical adjustments, failing to fully reflect the nonlinear dynamic processes of the hydraulic system and environmental disturbances, thus exhibiting shortcomings in dynamic response, stability accuracy, and real-time performance.

[0003] Current research has attempted to improve attitude control accuracy through digital simulation or neural network prediction models, but most methods still suffer from problems such as isolated models, insufficient data utilization, and a single optimization mechanism. Purely data-driven methods are prone to overfitting and lack physical constraints, resulting in a lack of interpretability and engineering reliability in the prediction results. Therefore, there is a problem of low reliability in the attitude control of underwater leveling machines. Summary of the Invention

[0004] This application provides an attitude control method, system, device, and underwater leveling machine. By integrating physical information neural network prediction, digital twin simulation verification, and multi-objective optimization, the reliability of attitude control of the underwater leveling machine is improved.

[0005] In a first aspect, embodiments of this application provide an attitude control method for an underwater leveling machine, comprising: A dynamic model is established, representing the motion state and force relationship of the hydraulic drive mechanism of the underwater leveling machine. A physical information neural network (PIN) is constructed based on the dynamic model to obtain the attitude error prediction result of the underwater leveling machine based on the trained PIN. The dynamic model serves as the physical constraint condition for the PIN during training. A digital twin of the leveling machine is constructed, and the attitude error prediction result is simulated and verified within the digital twin. Based on the simulation verification results, multiple control objectives of the underwater leveling machine are optimized in parallel using the control parameters as optimization variables, generating a control solution set. A target control parameter combination is determined from the control solution set, and control commands for attitude adjustment of the underwater leveling machine are generated based on the target control parameter combination.

[0006] In some embodiments, establishing the dynamic model includes: The dynamic model is established based on the hydraulic cylinder thrust, bottom resistance, water flow disturbance, and gravity coupling effect of the hydraulic drive mechanism.

[0007] In some embodiments, a physical information neural network is constructed based on the dynamic model to obtain the attitude error prediction result of the underwater leveling machine based on the trained physical information neural network, including: A physical residual function is constructed based on the dynamic model; wherein the physical residual function characterizes the equation residuals of the dynamic model; The sum of squares of the physical residual function is added as a physical loss term to the overall loss function of the physical information neural network, so as to train the physical information neural network based on the overall loss function.

[0008] In some embodiments, the physical residual function includes:

[0009] in, For physical residuals, The effective area parameter is the equivalent area parameter of the hydraulic drive mechanism. and These represent the pressure difference between the i-th and j-th hydraulic cylinders, respectively. The equivalent mass parameters of the underwater leveling machine are... and Let represent the accelerations of the leveling arm in the i-th and j-th hydraulic cylinders in the direction of piston rod extension and retraction, respectively. This is the equivalent damping coefficient. and These represent the movement speeds of the leveling arm in the i-th and j-th hydraulic cylinders, respectively. This is the equivalent stiffness coefficient. and These represent the displacements of the leveling arm in the i-th and j-th hydraulic cylinders, respectively. and These represent the external forces applied to the i-th and j-th hydraulic cylinders, respectively.

[0010] In some embodiments, the overall loss function includes:

[0011]

[0012]

[0013] in, For the overall loss function, For data loss items, These are the weighting coefficients for the physical constraint loss. For physical loss items, These are the weighting coefficients for the regularization loss. For regularization terms, For the sample size, For the attitude error prediction results, This represents the true attitude error value. Let P be the set of all hydraulic cylinder pairs for which physical residuals need to be calculated, and P be the traversal index. This represents the k-th sampling time.

[0014] In some embodiments, using the control parameters of the underwater leveler as optimization variables, multiple control objectives of the underwater leveler are optimized in parallel to generate a control solution set. This includes: defining a multi-objective function using PID parameters as the optimization variables to minimize settling time, overshoot, and cumulative attitude error; initializing a population with real-number encoding; determining the objective function value corresponding to each individual in the population through the physical information neural network model; performing non-dominated sorting and generating reference points; generating new individuals using simulated binary crossover and polynomial mutation operations; iteratively updating the population; and, upon completion of the update, forming a Pareto optimal solution set as the control solution set based on the hypervolume convergence criterion.

[0015] In some embodiments, determining a target control parameter combination from the control solution set and generating control commands for attitude adjustment of the underwater leveling machine based on the target control parameter combination includes: Calculate the group utility value and individual regret value for each solution in the control solution set, and determine a compromise ranking value based on the group utility value and the individual regret value; if the compromise ranking value satisfies a preset condition, select the solution corresponding to the compromise ranking value as the target control parameter combination.

[0016] Secondly, embodiments of this application provide an attitude control system for an underwater leveling machine, characterized in that it includes: The dynamic modeling submodule is used to establish a dynamic model, which characterizes the motion state and force relationship of the hydraulic drive mechanism of the underwater leveling machine. A physical constraint deep learning module is used to construct a physical information neural network based on the dynamic model, so as to obtain the attitude error prediction result of the underwater leveling machine based on the trained physical information neural network; wherein, the dynamic model is the physical constraint condition of the physical information neural network during the training process. The virtual leveling simulation module is used to construct a digital twin of the leveling machine and to perform simulation verification of the attitude error prediction results in the digital twin. The multi-objective optimization module is used to perform parallel optimization of multiple control objectives of the underwater leveling machine based on the simulation verification results, using the control parameters of the underwater leveling machine as optimization variables, and generate a control solution set. The decision module is used to determine the target control parameter combination from the control solution set, and generate control commands to control the underwater leveling machine to adjust its attitude based on the target control parameter combination.

[0017] 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.

[0018] Fourthly, embodiments of this application also provide an underwater leveling machine, which is equipped with the attitude control system of the underwater leveling machine described above.

[0019] Compared with existing technologies, the beneficial effects of this application are as follows: by establishing a dynamic model to characterize the motion state and force relationship of the hydraulic drive mechanism, a physical information neural network is constructed based on this model, and physical constraints are incorporated into the training process, thereby enhancing the physical consistency and accuracy of attitude error prediction, improving the generalization ability of the model, constructing a digital twin to simulate and verify the prediction results, comprehensively evaluating the effectiveness of the control strategy in a virtual environment, using control parameters as optimization variables to perform parallel optimization of multiple control objectives, generating diverse control solution sets, supporting collaborative optimization of multi-objective balance in complex underwater environments, and generating control commands after determining the target control parameter combination from the solution set to guide the underwater leveling machine to perform precise attitude adjustment, thereby improving the reliability of attitude control of the underwater leveling machine. Attached Figure Description

[0020] Figure 1This is a schematic diagram illustrating the steps of the attitude control method for an underwater leveling machine provided in an embodiment of this application.

[0021] Figure 2 This is a schematic diagram of the physical architecture of the attitude control system provided in an embodiment of this application. Detailed Implementation

[0022] 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.

[0023] 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.

[0024] Please refer to Figure 1 , Figure 1 This diagram illustrates the steps of an attitude control method for an underwater leveling machine provided in this embodiment. The attitude control method for the underwater leveling machine may include: S1. Establish a dynamic model.

[0025] S2. Construct a physical information neural network based on the dynamic model, and obtain the attitude error prediction result of the underwater leveling machine based on the trained physical information neural network.

[0026] S3. Construct a digital twin of the leveling machine and perform simulation verification of the attitude error prediction results in the digital twin.

[0027] S4. Based on the simulation verification results, using the control parameters of the underwater leveler as optimization variables, multiple control objectives of the underwater leveler are optimized in parallel to generate a control solution set.

[0028] S5. Determine the target control parameter combination from the control solution set, and generate control commands for the underwater leveling machine to adjust its attitude based on the target control parameter combination.

[0029] In this embodiment, the dynamic model characterizes the motion state and force relationship of the hydraulic drive mechanism of the underwater leveling machine, which can be understood as a set of mathematical expressions that include hydraulic cylinder thrust, bottom resistance, water flow disturbance and gravity coupling effect.

[0030] The specific steps involved in establishing this model include: obtaining the structural parameters and external environmental parameters of the underwater leveling machine, and then deriving the dynamic model through mechanical analysis.

[0031] A Physical Information Neural Network (PIN) is a deep learning model that embeds physical dynamic equations as constraints into the training process. Its loss function consists of a data fitting term, a physical residual term, and a regularization term. The dynamic model serves as the physical constraint during the training process of the PIN, representing a portion of the physical residual term. The learning process of the PIN relies not only on data fitting but also on the consistency constraints of physical laws. Given an established dynamic model and historical operational data of an underwater leveling machine, an optimization algorithm minimizes the joint loss function combining data error and physical residuals to adjust the network parameters, resulting in an attitude prediction model that combines prediction accuracy and physical reliability.

[0032] A digital twin refers to a high-fidelity dynamic simulation model of a physical leveling machine in virtual space; that is, a virtual system capable of real-time data mapping and interaction with the physical equipment. The simulation verification process involves the virtual leveling simulation platform receiving the attitude prediction results output by the physical information neural network, and then simulating the dynamic response process of the leveling machine under this state in the virtual environment. This allows for the evaluation of the effectiveness of the control strategy under risk-free conditions and provides data support for optimization. Parallel optimization refers to using a multi-objective evolutionary algorithm to simultaneously process multiple competing performance indicators, i.e., finding a set of solutions that achieve the best balance among multiple objectives. The generation of the control solution set specifically includes, after the digital twin completes the simulation verification of the control strategy, using the control parameters of the underwater leveling machine as optimization variables, iteratively filtering the optimization variables through preset optimization parameters to obtain an optimal solution set containing multiple trade-offs, i.e., the control solution set.

[0033] In the description of the embodiments in this application, "multiple" means two or more, unless otherwise explicitly specified. For example, the multiple control objectives targeted by the above optimization process may include settling time, overshoot, and cumulative attitude error, etc.

[0034] Furthermore, the action of determining the target control parameter combination can specifically include, after obtaining the control solution set, calculating the comprehensive evaluation index of each solution in the control solution set according to the set multi-criteria decision method, and under the premise of satisfying specific decision conditions, selecting the parameter combination with the best comprehensive performance to obtain an efficient control command that can balance response speed, stability and accuracy, and finally sending it to the execution unit of the underwater leveling machine to drive the leveling machine to complete attitude adjustment.

[0035] In some embodiments, the specific implementation of establishing the dynamic model may include: establishing the dynamic model based on the hydraulic cylinder thrust, bottom resistance, water flow disturbance and gravity coupling effect of the hydraulic drive mechanism.

[0036] The hydraulic cylinder thrust is the primary power source for the mechanism's attitude adjustment, and its magnitude is determined by the differential pressure of the hydraulic system. Bottom resistance originates from the interaction between the leveler and the seabed foundation, reflecting the reaction force caused by the unevenness of the working surface. Water flow disturbance is a complex hydrodynamic load imposed on the equipment by the marine environment, exhibiting significant time-varying and nonlinear characteristics. Gravity coupling effects consider the leveler's own mass distribution and the different impacts of its gravity components on each hydraulic cylinder under tilted posture. By incorporating these four key physical effects into a unified mechanical analysis framework, a nonlinear dynamic model that accurately reflects the true dynamic characteristics of the leveler can be established.

[0037] For example, the established dynamic model includes a single-cylinder dynamic model and a multi-cylinder coupled attitude model.

[0038] The single-cylinder dynamics model provided in this embodiment is:

[0039] in: This refers to the effective piston area of ​​the hydraulic cylinder. For hydraulic differential pressure, ; The pressure in the rodless chamber of the hydraulic cylinder. For rod chamber pressure; This refers to the total mass of the leveling arm and the load. The acceleration of the leveling arm in the direction of piston rod extension and retraction; The damping coefficient; The speed of the leveling arm's movement; This is the stiffness coefficient; This refers to the displacement of the leveling arm relative to its equilibrium position; External forces can include seabed reaction forces or fluid disturbance forces.

[0040] In the multi-cylinder coupled attitude model provided in this application embodiment, the attitude adjustment of the multi-degree-of-freedom leveling system satisfies:

[0041] Among them, travel difference The attitude error of the leveling machine is characterized and serves as the core input for subsequent optimization of the control parameters of the physical information neural network and the underwater leveling machine.

[0042] In some embodiments, constructing a physical information neural network based on the dynamic model to obtain the attitude error prediction result of the underwater leveling machine based on the trained physical information neural network may include: A physical residual function is constructed based on the dynamic model; the sum of squares of the physical residual function is added as a physical loss term to the overall loss function of the physical information neural network, so as to train the physical information neural network based on the overall loss function.

[0043] In this embodiment, the physical residual function characterizes the equation residuals of the dynamic model. Constructing the physical residual function means, based on the established dynamic model, rearranging and combining the terms of the corresponding dynamic equations for any pair of coupled hydraulic cylinders i and j to form a residual expression. Theoretically, the value of this function should be zero. However, in practical applications, due to model simplification and measurement errors, when the predicted values ​​(such as predicted displacement, velocity, and acceleration) of the Physical Information Neural Network (PINN) are substituted into this equation, the residual value will be zero. It is no longer zero. The magnitude of its absolute value quantifies the degree to which the neural network prediction results violate known physical laws, in order to continuously evaluate the rationality of the attitude error prediction results output by the physical information neural network.

[0044] The PINN model structure includes an input layer, multiple fully connected networks (or LSTM-fusion layers), and an output layer. The input layer includes:

[0045] The output is:

[0046] For the first The physical residual function constructed for the cylinder group includes:

[0047] in, For physical residuals, The effective area parameter is the equivalent area parameter of the hydraulic drive mechanism. and These represent the pressure difference between the i-th and j-th hydraulic cylinders, respectively. The equivalent mass parameters of the underwater leveling machine are... and Let represent the accelerations of the leveling arm in the i-th and j-th hydraulic cylinders in the direction of piston rod extension and retraction, respectively. This is the equivalent damping coefficient. and These represent the movement speeds of the leveling arm in the i-th and j-th hydraulic cylinders, respectively. This is the equivalent stiffness coefficient. and These represent the displacements of the leveling arm in the i-th and j-th hydraulic cylinders, respectively. and These represent the external forces applied to the i-th and j-th hydraulic cylinders, respectively.

[0048] The overall loss function includes:

[0049]

[0050]

[0051] in, For the overall loss function, For data loss items, These are the weighting coefficients for the physical constraint loss. This is the physical loss weight, with values ​​ranging from 0.1 to 1.0; These are the weighting coefficients for the regularization loss. For regularization terms, For the sample size, For the attitude error prediction results, This represents the true attitude error value. Let P be the set of all hydraulic cylinder pairs for which physical residuals need to be calculated, and P be the traversal index. This represents the k-th sampling time.

[0052] PINN uses the ReLU activation function. Its network structure consists of four hidden layers, each with 128 neurons. The optimization process employs the Adam optimization algorithm with an initial learning rate of 0.1%. The number of training epochs is dynamically adjusted between 2,000 and 10,000, depending on performance on the validation set, to avoid undertraining or overfitting. During training, L2 regularization and Dropout are used simultaneously. L2 regularization constrains parameter values ​​by applying a penalty term with a coefficient of 1 / 100,000 to the network weights, while Dropout randomly masks some neurons in each training iteration with a 20% probability. The PINN network can be deployed on edge computing nodes, achieving a single prediction latency of <10ms.

[0053] In some embodiments, using the control parameters of the underwater leveling machine as optimization variables, multiple control objectives of the underwater leveling machine are optimized in parallel to generate a control solution set, including: Using PID parameters as the optimization variables, a multi-objective function is defined to minimize settling time, overshoot, and cumulative attitude error. A population with real-number encoding is initialized, and the objective function value for each individual in the population is determined using the physical information neural network model. Non-dominated sorting is performed to generate reference points, and new individuals are generated using simulated binary crossover and polynomial mutation operations. The population is iteratively updated. After the update is complete, a Pareto optimal solution set is formed based on the hypervolume convergence criterion as the control solution set.

[0054] Among them, the multi-objective optimization process will use the parameter set of PID parameters As the optimization variable, the objective function is defined as follows:

[0055] in, To adjust the time, For overshoot, , where represents the cumulative attitude error. These three objectives are often competing performance metrics in a control system; for example, reducing overshoot may prolong settling time. Therefore, the goal of multi-objective optimization is to find the optimal balance between them, rather than a single optimal solution.

[0056] The optimization process employs a non-dominated sorting genetic algorithm (NSGA-III). The optimization objective of NSGA-III is to optimize the PID parameters. To optimize the variables, a multi-objective function is defined to minimize settling time, overshoot, and cumulative attitude error; Define the objective function:

[0057] in, To adjust the time; This is the overshoot. To accumulate attitude error, , Let be the attitude error value at a specific time t.

[0058] The optimization process for NSGA-III includes: Population for initializing real number encoding population It contains N individuals, each of which is a set of PID parameter vectors. The population size can be set between 50 and 100. The objective function value (T, O, E) corresponding to each individual in the population is determined by the physical information neural network model. Non-dominated sorting is performed and reference points are generated. The solutions in the population are divided into different non-dominated levels, and a set of uniformly distributed reference points is generated to maintain the diversity of solutions on the Pareto front.

[0059] New individuals are generated using simulated binary crossover (SBX) and polynomial mutation operations, and the population is iteratively updated. The crossover probability can be set to 0.9, the mutation rate to 1 / dimension, and the distribution exponent. , The number of iterations can be set to 20 to 50 generations, with each generation repeating the evaluation, sorting, selection, and evolution operations. After the update is complete, a Pareto optimal solution set is formed based on the hypervolume convergence criterion. As the control solution set, the solution set contains numerous non-dominated solutions, each of which represents a superior trade-off between the three objectives of settling time, overshoot, and cumulative error, which cannot be surpassed by each other, thus providing a rich library of alternative solutions for subsequent decision-making and selection.

[0060] Furthermore, the method for determining the target control parameter combination from the control solution set can be to apply the VIKOR multi-criteria decision-making method for comprehensive optimization. Applying the VIKOR multi-criteria decision-making method for comprehensive optimization includes: For each candidate parameter solution in the control solution set A quantitative evaluation is performed to calculate the group utility value of each solution in the control solution set. and individual regret value The compromise ranking value is determined based on the group utility value and the individual regret value. ; For example, selecting the comprehensive optimal solution from the control solution set. The calculation formula is as follows:

[0061]

[0062] in, Used to iterate through all optimization targets. The value of corresponds to three specific control performance indicators, namely settling time. Overshoot and cumulative attitude error , To control a candidate solution in the solution set In the The specific numerical values ​​for each target For the first The ideal solution to an objective, that is, the best value that all candidate solutions can achieve on that objective. For the first The negative ideal solution to an objective, that is, the worst value that all candidate solutions can achieve on that objective. For the first The weight assigned to each target can be set to 1 / 3 by default. , which is a weighting coefficient used to balance the tendency between two decision-making strategies: maximizing group utility (i.e., pursuing the best overall performance, v close to 1) and minimizing individual regret (i.e., avoiding serious shortcomings, v close to 0). and The minimum and maximum population utility values ​​among all candidate solutions. and These are the minimum and maximum individual regret values, respectively.

[0063] Finally, all candidate solutions are based on The values ​​are sorted, and the smallest solution is the most advantageous solution at the beginning of the compromise sort, provided that the compromise sort values ​​meet a preset condition, i.e.

[0064] Its characterization is the solution ranked first. Do they satisfy simultaneously? The value is large enough and ranks highly in the G or R sort, thus ultimately determining it as the optimal control parameter. The solution corresponding to the compromise sorting value is selected as the target control parameter combination.

[0065] In the above implementation process, a dynamic model is established to characterize the motion state and force relationship of the hydraulic drive mechanism. Based on this model, a physical information neural network is constructed, and physical constraints are incorporated into the training process to enhance the physical consistency and accuracy of attitude error prediction, improve the generalization ability of the model, and construct a digital twin to simulate and verify the prediction results. The effectiveness of the control strategy is comprehensively evaluated in a virtual environment. With control parameters as optimization variables, multiple control objectives are optimized in parallel to generate diverse control solution sets, which support the collaborative optimization of multi-objective balance in complex underwater environments. After determining the combination of target control parameters from the solution set, control commands are generated to guide the underwater leveler to make precise attitude adjustments, thereby improving the reliability of attitude control of the underwater leveler.

[0066] Based on the same concept, embodiments of this application also provide an attitude control system for an underwater leveling machine, which may include: The system comprises the following modules: a dynamic modeling submodule for establishing a dynamic model representing the motion state and force relationships of the hydraulic drive mechanism of the underwater leveling machine; a physical constraint deep learning module for constructing a physical information neural network based on the dynamic model to obtain the attitude error prediction result of the underwater leveling machine based on the trained physical information neural network; wherein the dynamic model is the physical constraint condition of the physical information neural network during the training process; a virtual leveling simulation module for constructing a digital twin of the leveling machine and simulating and verifying the attitude error prediction result in the digital twin; a multi-objective optimization module for performing parallel optimization of multiple control objectives of the underwater leveling machine based on the simulation verification results and using the control parameters of the underwater leveling machine as optimization variables to generate a control solution set; and a decision module for determining the target control parameter combination from the control solution set and generating control commands for adjusting the attitude of the underwater leveling machine based on the target control parameter combination.

[0067] Furthermore, the attitude control system of the underwater leveling machine is designed to deeply integrate functional modules and physical architecture. Functionally, intelligent control is achieved through the collaborative efforts of the aforementioned five modules, which are systematically integrated into a physical architecture consisting of four parts. Please refer to [link / reference]. Figure 2 , Figure 2 This is a schematic diagram of the physical architecture of the attitude control system provided in an embodiment of this application.

[0068] The sensing and monitoring unit, acting as the system's "sensors," collects real-time operational status data of the leveling machine through deployed depth sensors, attitude gyroscopes, hydraulic pressure sensors, and bottom reaction force sensors, providing data input to all upper-level modules. The control unit embeds a main control module with a physical information neural network algorithm, undertaking core computational tasks such as attitude prediction, parameter optimization, and adaptive adjustment. It integrates physical constraint deep learning algorithms, multi-objective optimization, and decision-making modules, such as the PINN algorithm, NSGA-III algorithm, and VIKOR optimal selection algorithm. The virtual-real feedback channel serves as a "bridge" connecting the virtual and physical worlds, enabling bidirectional data synchronization and correction between the physical leveling machine and the virtual simulation model based on the digital twin model. This ensures consistency between the digital twin and the physical equipment, allowing the virtual leveling simulation module to operate effectively. The execution unit includes a multi-degree-of-freedom leveling mechanism composed of multiple hydraulic cylinders. It receives and executes control commands from the control unit, achieving precise adjustments in posture (forward / backward, left / right, and lifting) to achieve attitude leveling.

[0069] 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.

[0070] 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.

[0071] 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.

[0072] Based on the same concept, this application also provides an underwater leveling machine, which is equipped with the attitude control system of the underwater leveling machine described above.

[0073] 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 method of attitude control of an underwater screed, characterized in that, The method comprises the following steps: establishing a dynamic model, wherein the dynamic model represents a relationship between a motion state and a force of a hydraulic driving mechanism of the underwater screed; constructing a physical information neural network based on the dynamic model, and obtaining a posture error prediction result of the underwater screed based on the trained physical information neural network, wherein the dynamic model is a physical constraint condition of the physical information neural network in a training process; constructing a digital twin of the underwater screed, and simulating and verifying the posture error prediction result in the digital twin; based on a result of the simulation and verification, performing parallel optimization on multiple control targets of the underwater screed by taking control parameters of the underwater screed as optimization variables, and generating a control solution set; determining a target control parameter combination from the control solution set, and generating a control instruction for adjusting a posture of the underwater screed based on the target control parameter combination.

2. The method of claim 1, wherein, The step of establishing the dynamic model comprises: establishing the dynamic model based on a hydraulic cylinder thrust, a bottom resistance, a water flow disturbance and a gravity coupling effect of the hydraulic driving mechanism.

3. The method of claim 1, wherein, The step of constructing the physical information neural network based on the dynamic model, and obtaining the posture error prediction result of the underwater screed based on the trained physical information neural network comprises: constructing a physical residual function based on the dynamic model, wherein the physical residual function represents an equation residual of the dynamic model; adding a sum of squares of the physical residual function as a physical loss term into a total loss function of the physical information neural network, and training the physical information neural network based on the total loss function.

4. The method of claim 3, wherein, The physical residual function comprises: wherein, is the physical residual, is the equivalent area parameter of the hydraulic drive mechanism, and denote the pressure difference of the i-th and j-th hydraulic cylinder, respectively, is the equivalent mass parameter of the screed, and denote the acceleration of the screed arm in the direction of piston rod extension in the i-th and j-th hydraulic cylinder, respectively, is the equivalent damping coefficient, and denote the movement speed of the screed arm in the i-th and j-th hydraulic cylinder, respectively, is the equivalent stiffness coefficient, and denote the displacement of the screed arm in the i-th and j-th hydraulic cylinder, respectively, and denote the external force acting on the i-th and j-th hydraulic cylinder, respectively.

5. The method of claim 3, wherein, The total loss function comprises: wherein, is the total loss function, is the data loss term, is the weight coefficient of the physical constraint loss, is the physical loss term, is the weight coefficient of the regularization loss, is the regularization term, is the number of samples, is the pose error prediction result, is the true pose error value, is the traversal index, P is the set of all hydraulic cylinder pairs that need to calculate the physical defects, denotes the k-th sampling time.

6. The method of claim 1, wherein, The step of performing the parallel optimization on the multiple control targets of the underwater screed by taking the control parameters of the underwater screed as the optimization variables, and generating the control solution set based on the result of the simulation and verification comprises: taking PID parameters as the optimization variables, defining a multi-objective function to minimize a regulation time, an overshoot and a cumulative posture error, initializing a population of real number coding, determining a target function value corresponding to each individual in the population through the physical information neural network model, performing non-dominated sorting and generating a reference point, generating new individuals by using simulated binary crossover and polynomial mutation operations, and iteratively updating the population, and forming a Pareto optimal solution set as the control solution set according to a hyper volume convergence criterion in a case where the updating is completed. The step of determining the target control parameter combination from the control solution set, and generating the control instruction for adjusting the posture of the underwater screed based on the target control parameter combination comprises: calculating a group utility value and an individual regret value of each solution in the control solution set, and determining a trade-off ranking value based on the group utility value and the individual regret value, and selecting a solution corresponding to the trade-off ranking value as the target control parameter combination in a case where the trade-off ranking value meets a preset condition. The method comprises the following steps:

7. The method of claim 1, wherein, a dynamics modeling submodule is configured to establish a dynamic model, wherein the dynamic model represents a relationship between a motion state and a force of a hydraulic driving mechanism of the underwater screed; ​ ​ 8. A posture control system for an underwater screed, characterized by ​ ​ a physical constraint deep learning module configured to construct a physical information neural network based on the dynamic model, to obtain a posture error prediction result of the underwater screed machine based on the trained physical information neural network; wherein the dynamic model is a physical constraint condition for the physical information neural network in a training process; a virtual screed simulation module configured to construct a digital twin of the screed machine, and to simulate and verify the posture error prediction result in the digital twin; a multi-objective optimization module configured to perform parallel optimization on multiple control objectives of the underwater screed machine based on a result of the simulation and verification, to generate a control solution set, with control parameters of the underwater screed machine as optimization variables; a decision module configured to determine a target control parameter combination from the control solution set, and to generate a control instruction for controlling the underwater screed machine to perform posture adjustment based on the target control parameter combination.

9. A computer device, comprising: The computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. An underwater screed, characterized by The underwater screed machine is provided with the posture control system of the underwater screed machine according to claim 8.

Images