Accurate pile-forming control method for underwater sand compaction pile

Abstract

The invention relates to the field of foundation treatment engineering, and discloses an underwater compaction sand pile accurate pile-forming control method which comprises the following steps that geological and sea condition investigation is conducted on a construction area, and after preprocessing is conducted, four fields of coupling data are formed; respectively constructing a control equation of each field based on the four-field coupling data, and constructing a training data set; and a fusion model combining the depth Gaussian process and the fractional order PID is constructed according to the training data set, construction parameters including the in-pipe air pressure and the pipe drawing speed are solved in real time based on a mixed agent model, construction operation is dynamically adjusted according to the construction parameters, and underwater sand compaction pile forming is completed. Four-field parameters and construction parameters are synchronously collected and subjected to standardized preprocessing, multi-working-condition simulation data and field pile testing data are fused to form a training data set, and a mixed agent model is built and optimized, so that the influence of the multi-field coupling effect on the pile forming quality is quantified, and closed-loop dynamic control over the underwater sand compaction pile construction process is achieved.

Description

Technical Field

[0001] This invention relates to the field of foundation treatment engineering technology, specifically a method for precise pile formation control of underwater compaction sand piles. Background Technology

[0002] With the rapid development of infrastructure construction such as transportation and energy in coastal areas, underwater compaction sand piles, as an efficient foundation reinforcement method, are widely used in projects such as cross-sea bridges, offshore wind power foundations, and port terminals. Its core principle is to squeeze sand into the underwater soft soil foundation to form a dense sand pile composite foundation, thereby improving the foundation bearing capacity, reducing settlement, and adapting to the engineering construction needs of complex underwater geological environments.

[0003] Existing underwater compaction sand pile construction typically involves first obtaining basic data such as soil layer distribution and physical and mechanical parameters of the construction area through geological survey drilling. Then, combined with the past experience of engineers, core construction parameters such as in-tube air pressure and pipe extraction speed are preset. During construction, an open-loop operation mode is often adopted, and local construction data is collected and monitored through sensors. This approach fails to fully consider the coupling effect of underwater geological stress field, sea state current field, particle flow field and pore water seepage field, and cannot adapt to actual working conditions such as random disturbance of soil layer parameters and dynamic changes in the hydrological environment, resulting in insufficient stability of pile quality. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a precise pile-forming control method for underwater compacted sand piles, solving the problem of insufficient pile quality stability in existing underwater compacted sand pile forming processes.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for precise pile formation control of underwater compaction sand piles, comprising the following steps:

[0006] S1. Conduct geological and marine condition surveys of the construction area, and simultaneously collect data on geological stress field, marine flow field, particle flow field, pore water seepage field and construction parameters of the construction area. After preprocessing, the data are coupled into four fields.

[0007] S2. Based on the four-field coupled data, construct the control equations for each field, solve them simultaneously to obtain the coupling relationship between compaction and construction parameters, and integrate simulation data and field test pile data to construct a training dataset.

[0008] S3. Construct a fusion model combining deep Gaussian process and fractional-order PID based on the training dataset, and train and optimize it to obtain a hybrid proxy model;

[0009] S4. Based on the hybrid agent model, the construction parameters including the air pressure inside the pipe and the pipe pulling speed are solved in real time, and the construction operation is dynamically adjusted according to the construction parameters to complete the underwater compaction sand pile.

[0010] By adopting the above technical solution, and simultaneously collecting four field parameters and construction parameters and performing standardized preprocessing, the control equations and coupling relationships of each field are constructed. Multi-condition simulation data and field test pile data are integrated to form a training dataset. A hybrid surrogate model combining deep Gaussian process and fractional-order PID is built and optimized. The construction parameters are solved in real time and the construction operation is dynamically adjusted, thereby quantifying the impact of multi-field coupling on pile quality, improving the model's adaptability to complex working conditions and the accuracy of parameter prediction, realizing closed-loop dynamic control of the underwater compaction sand pile construction process, and solving the problem of insufficient pile quality stability in existing underwater compaction sand piles.

[0011] Preferably, in step S1, the formation of four-field coupled data specifically includes the following steps:

[0012] By combining borehole sampling with laboratory tests, the effective unit weight, cohesion, and coefficient of earth pressure at rest of the soil layer were obtained. The distribution characteristics of the effective unit weight of the soil layer were statistically analyzed to form geological stress field data.

[0013] Collect meteorological and hydrological data of the construction area, monitor wave amplitude, wavelength, wave period and tidal characteristics, establish the correlation between sea state and ship attitude, and form sea state flow field data;

[0014] The moisture content of the sand at the outlet of the sand chamber and the height of the sand surface inside the casing are collected simultaneously to form particle flow field data;

[0015] Collect pore water pressure in the casing to generate pore water seepage field data;

[0016] Simultaneously collect data on soil lateral pressure, ship attitude, wave conditions, sand level inside the pipe, air pressure inside the pipe, pipe extraction speed, and longitudinal strain of the pile to form construction parameter data;

[0017] The geological stress field, sea state flow field, particle flow field, pore water seepage field, and construction parameter data are timestamped based on GPS second pulse signals. Abnormal data are removed using the 3σ criterion, and data noise is reduced through filtering algorithms to obtain standardized four-field coupled data.

[0018] Preferably, in step S2, constructing the training dataset specifically includes the following steps:

[0019] Based on the four-field coupled data, single-field control equations were constructed for the corresponding geological stress field, sea state flow field, particle flow field, and pore water seepage field.

[0020] Based on the interaction relationship of the parameters of each single field control equation, the single field control equations are integrated to establish the coupling relationship between compaction and construction parameters.

[0021] Based on the coupling relationship, a two-way coupling simulation tool is used to simulate different soil layer types, wave levels, air pressures, and pipe pulling speeds, generating simulation data containing four field parameters, construction parameters, and corresponding compaction.

[0022] Collect four-field coupling data, construction parameter data, and actual density measurement data during the on-site pile test process, merge them with simulation data, and divide them into training set, validation set, and test set according to a preset ratio to form a training dataset.

[0023] Preferably, the construction of the single-field governing equations corresponding to the geological stress field, sea state flow field, particle flow field, and pore water seepage field specifically includes the following steps:

[0024] Based on geological stress field data and combined with the random disturbance characteristics of soil layer parameters, stress calculation equations are constructed.

[0025] Based on sea state and current field data, and combined with the principles of ship dynamics and wave mechanics, a coupled equation for ship attitude and wave parameters is constructed.

[0026] Based on particle flow field data, and combined with the bonding effect and memory characteristics between sand particles, a correlation equation between particle contact force and density evolution is constructed.

[0027] Based on pore water seepage field data and combined with the seepage law, a dynamic variation equation for porosity and pore water pressure is constructed.

[0028] Preferably, in step S3, obtaining the hybrid agent model specifically includes the following steps:

[0029] Based on the feature dimensions of the training dataset, a deep Gaussian process model containing an input layer, a hidden layer, and an output layer is constructed, and a hybrid kernel function is used to capture the nonlinear features of different field parameters.

[0030] Based on nonlinear characteristics, an optimization algorithm is used to minimize the prediction error loss function, and a deep Gaussian process model is iteratively trained and the kernel function parameters are optimized until the fitting accuracy of the validation set meets the preset requirements.

[0031] The proportional coefficient, integral coefficient, derivative coefficient, and fractional order are determined by a joint tuning method. Based on the density prediction deviation, the correction logic is designed to construct a fractional-order PID correction module.

[0032] The trained deep Gaussian process model is fused with a fractional-order PID correction module to form a hybrid surrogate model.

[0033] Preferably, the method of using hybrid kernel functions to capture the nonlinear characteristics of different field parameters specifically includes the following steps:

[0034] Based on the input feature types of the deep Gaussian process model, the Matérn kernel function is selected as the corresponding sub-kernel function for different nonlinear characteristics of geological stress field, sea state flow field, particle flow field, pore water seepage field and construction parameters.

[0035] The sub-kernel functions are combined in a product form to form a hybrid kernel function, and the parameters of each sub-kernel function are initialized based on the statistical characteristics of the training dataset.

[0036] During the training of the deep Gaussian process model, the parameters of each sub-kernel function are optimized simultaneously, enabling the hybrid kernel function to capture the nonlinear correlation between different field parameters and density.

[0037] Preferably, in step S4, the dynamic adjustment of construction operations based on construction parameters specifically includes the following steps:

[0038] By combining the coupling relationship, an objective function containing a compaction deviation control term and a parameter adjustment smoothing term is constructed;

[0039] Define the upper and lower limits of the air pressure inside the tube, the range of the tube removal speed, and the allowable value of the compaction to form constraint conditions;

[0040] The real-time collected four-field coupling data is input at a preset cycle and then input into the hybrid surrogate model to obtain the density prediction value.

[0041] Based on the predicted compaction value, combined with the objective function and constraints, a robust optimization algorithm is used to iteratively solve the construction parameters, including the air pressure inside the pipe and the pipe pulling speed, and the construction operation is dynamically adjusted in real time according to the construction parameters.

[0042] Preferably, the step of iteratively solving for the construction parameters including the air pressure inside the pipe and the pipe pulling speed using a robust optimization algorithm specifically includes the following steps:

[0043] Based on the predicted density, objective function, and constraints, and combining the robust expectation improvement criterion and Monte Carlo sampling method, an optimization objective with robustness and constraints is constructed.

[0044] The L-BFGS algorithm is used to iteratively solve the optimization objective. In each iteration, the objective value is updated based on the hybrid surrogate model. The iteration stops when the preset convergence accuracy or number of iterations is met. The output includes construction parameters including the air pressure inside the pipe and the pipe pulling speed.

[0045] This invention provides a method for precise control of underwater compaction sand pile formation. It has the following beneficial effects:

[0046] 1. This invention simultaneously collects four field parameters and construction parameters and performs standardized preprocessing to construct control equations and coupling relationships for each field. It integrates multi-condition simulation data and field test pile data to form a training dataset, builds and optimizes a hybrid surrogate model combining deep Gaussian process and fractional-order PID, solves construction parameters in real time and dynamically adjusts construction operations, thereby quantifying the impact of multi-field coupling on pile quality, improving the model's adaptability to complex working conditions and the accuracy of parameter prediction, realizing closed-loop dynamic control of the underwater compaction sand pile construction process, and solving the problem of insufficient pile quality stability in existing underwater compaction sand pile construction.

[0047] 2. This invention constructs control equations for each field based on four-field coupled data and solves them simultaneously to obtain the coupling relationship between compaction and construction parameters, thus characterizing the intrinsic correlation between multiple field parameters and pile quality. At the same time, it integrates simulation data from multiple working conditions and actual test pile data from the field to construct a training dataset, covering scenarios with different geological conditions, sea conditions, and combinations of construction parameters, thereby improving the richness of training samples and enabling the model to fully learn the parameter correlation rules under complex working conditions, thus possessing stronger generalization ability.

[0048] 3. This invention uses a hybrid surrogate model that combines a deep Gaussian process with a fractional-order PID. The deep Gaussian process, with its powerful nonlinear fitting capability, can efficiently approximate the complex nonlinear relationships caused by multi-field coupling. The fractional-order PID correction module can compensate for model prediction deviations, suppress fluctuation effects, and further improve the accuracy of density prediction and construction parameter solution through training and optimization of model parameters, thereby improving the accuracy and stability of construction parameter prediction. Attached Figure Description

[0049] Figure 1 This is a flowchart of the precise pile formation control method for underwater compacted sand piles proposed in this invention.

[0050] Figure 2 This is a diagram illustrating the architecture of the underwater compaction sand pile precision pile formation control system proposed in an embodiment of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Example 1:

[0053] In a first embodiment of the present invention, the present invention provides a method for precise pile formation control of underwater compaction sand piles, such as... Figure 1As shown, it includes the following steps:

[0054] S1. Conduct geological and marine condition surveys of the construction area, and simultaneously collect data on geological stress field, marine flow field, particle flow field, pore water seepage field and construction parameters of the construction area. After preprocessing, the data are coupled into four fields.

[0055] Furthermore, in step S1, four-field coupled data is generated, specifically including the following steps:

[0056] By combining borehole sampling with laboratory tests, the effective unit weight, cohesion, and coefficient of earth pressure at rest of the soil layer were obtained. The distribution characteristics of the effective unit weight of the soil layer were statistically analyzed to form geological stress field data.

[0057] Collect meteorological and hydrological data of the construction area, monitor wave amplitude, wavelength, wave period and tidal characteristics, establish the correlation between sea state and ship attitude, and form sea state flow field data;

[0058] The moisture content of the sand at the outlet of the sand chamber and the height of the sand surface inside the casing are collected simultaneously to form particle flow field data;

[0059] Collect pore water pressure in the casing to generate pore water seepage field data;

[0060] Simultaneously collect data on soil lateral pressure, ship attitude, wave conditions, sand level inside the pipe, air pressure inside the pipe, pipe extraction speed, and longitudinal strain of the pile to form construction parameter data;

[0061] The geological stress field, sea state flow field, particle flow field, pore water seepage field, and construction parameter data are timestamped based on GPS second pulse signals. Abnormal data are removed using the 3σ criterion, and data noise is reduced through filtering algorithms to obtain standardized four-field coupled data.

[0062] Specifically, in this step, geological stress field data is obtained through borehole sampling and laboratory tests. After sampling at preset intervals, parameters such as effective unit weight, cohesion, and internal friction angle of the soil layer are obtained through laboratory tests, and their distribution characteristics are statistically analyzed to form data. Marine state flow field data relies on meteorological and hydrological equipment to collect historical and real-time data, monitor parameters such as wave amplitude and wavelength, and establish a correlation between hull dynamics and hull attitude. Particle flow field data is collected by a humidity sensor at the sand chamber outlet and a sand surface gauge at the top of the casing to collect sand moisture content and sand surface height, respectively. Pore water seepage field data is collected by a pressure sensor at the bottom of the casing to collect pore water pressure. Construction parameter data is simultaneously acquired by multiple sensors, including soil lateral pressure, hull attitude, air pressure inside the pipe, pipe extraction speed, and longitudinal strain of the pile.

[0063] Data preprocessing aligns the timestamps with GPS second pulse signals and uses the 3σ criterion to remove outliers, satisfying the requirements of... ,in For a single data entry, The mean of the data. The standard deviation is given. Noise is reduced through extended Kalman filtering; the state equation is: The observation equation is In the formula This is a state vector containing parameters such as soil lateral pressure and pore water pressure. To control the input, For process noise, For the observation vector, To mitigate noise, multi-source data is substituted into the equation for iterative calculation, and the filtered data is output to form standardized four-field coupled data, ensuring data reliability.

[0064] S2. Based on the four-field coupled data, construct the control equations for each field, solve them simultaneously to obtain the coupling relationship between compaction and construction parameters, and integrate simulation data and field test pile data to construct a training dataset.

[0065] Furthermore, in step S2, the training dataset is constructed, specifically including the following steps:

[0066] Based on the four-field coupled data, single-field control equations were constructed for the corresponding geological stress field, sea state flow field, particle flow field, and pore water seepage field.

[0067] Based on the interaction relationship of the parameters of each single field control equation, the single field control equations are integrated to establish the coupling relationship between compaction and construction parameters.

[0068] Based on the coupling relationship, a two-way coupling simulation tool is used to simulate different soil layer types, wave levels, air pressures, and pipe pulling speeds, generating simulation data containing four field parameters, construction parameters, and corresponding compaction.

[0069] Collect four-field coupling data, construction parameter data, and actual density measurement data during the on-site pile test process, merge them with simulation data, and divide them into training set, validation set, and test set according to a preset ratio to form a training dataset.

[0070] Furthermore, the single-field governing equations for the corresponding geological stress field, sea state flow field, particle flow field, and pore water seepage field are constructed respectively, specifically including the following steps:

[0071] Based on geological stress field data and combined with the random disturbance characteristics of soil layer parameters, stress calculation equations are constructed.

[0072] Based on sea state and current field data, and combined with the principles of ship dynamics and wave mechanics, a coupled equation for ship attitude and wave parameters is constructed.

[0073] Based on particle flow field data, and combined with the bonding effect and memory characteristics between sand particles, a correlation equation between particle contact force and density evolution is constructed.

[0074] Based on pore water seepage field data and combined with the seepage law, a dynamic variation equation for porosity and pore water pressure is constructed.

[0075] Specifically, in this step, constructing the single-field governing equations relies on the characteristics of the four-field coupled data and is gradually implemented by combining the physical and mechanical principles of each field. Generally, the stress calculation equations for the geological stress field need to consider the random disturbance characteristics of soil parameters, and their expression is: ,in For depth Lateral pressure of the soil layer at that location The effective unit weight of the soil layer follows a random distribution. The coefficient of earth pressure at rest. By substituting geological stress field data into the equation to determine soil cohesion, the relationship between depth and lateral pressure in the soil layer can be quantified.

[0076] The coupled equations of the sea state flow field are constructed based on the principles of ship dynamics and wave mechanics. By integrating sea state parameters such as wave amplitude, wavelength, and wave period with structural parameters such as ship mass and moment of inertia, a dynamic relationship between ship attitude and wave parameters is established. No additional complex formula derivation is required; the equation coefficients can be calibrated using measured data.

[0077] The correlation equations for the particle flow field need to reflect the bonding and memory characteristics between sand particles. Fractional Caputo derivatives are used to describe the density evolution, and the equation is as follows: ,in For Caputo fractional derivative, For fractional orders, the order reflects the memory effect. For density, The total normal contact force of the particles. The reaction force of pore water, For the volume of the particle aggregate, The critical compaction stress of the sand is given. Inputting particle flow field data will output the variation law of compaction over time.

[0078] The dynamic change equation of the pore water seepage field, combined with the seepage law, is expressed as follows: ,in As porosity The changing permeability coefficient Pore ​​water pressure, The porosity change rate is used to solve the spatiotemporal distribution of pore water pressure using pore water seepage field data.

[0079] After the individual field control equations were constructed, they were fused based on the interaction relationships between parameters to establish a coupling relationship between compaction and construction parameters such as pipe pressure and pipe extraction speed. Subsequently, based on this coupling relationship, a two-way coupling simulation tool was used to simulate different soil types, wave levels, air pressures, and pipe extraction speeds, generating a large amount of simulation data. At the same time, four-field coupling data, construction parameter data, and measured compaction data from the field pile testing process were collected and merged with the simulation data. These data were then divided into training, validation, and test sets according to a preset ratio to form a training dataset, providing comprehensive data support for subsequent model training.

[0080] S3. Construct a fusion model combining deep Gaussian process and fractional-order PID based on the training dataset, and train and optimize it to obtain a hybrid proxy model;

[0081] Furthermore, in step S3, the hybrid agent model is obtained, specifically including the following steps:

[0082] Based on the feature dimensions of the training dataset, a deep Gaussian process model containing an input layer, a hidden layer, and an output layer is constructed, and a hybrid kernel function is used to capture the nonlinear features of different field parameters.

[0083] Based on nonlinear characteristics, an optimization algorithm is used to minimize the prediction error loss function, and a deep Gaussian process model is iteratively trained and the kernel function parameters are optimized until the fitting accuracy of the validation set meets the preset requirements.

[0084] The proportional coefficient, integral coefficient, derivative coefficient, and fractional order are determined by a joint tuning method. Based on the density prediction deviation, the correction logic is designed to construct a fractional-order PID correction module.

[0085] The trained deep Gaussian process model is fused with a fractional-order PID correction module to form a hybrid surrogate model.

[0086] Furthermore, a hybrid kernel function is used to capture the nonlinear characteristics of different field parameters, specifically including the following steps:

[0087] Based on the input feature types of the deep Gaussian process model, the Matérn kernel function is selected as the corresponding sub-kernel function for different nonlinear characteristics of geological stress field, sea state flow field, particle flow field, pore water seepage field and construction parameters.

[0088] The sub-kernel functions are combined in a product form to form a hybrid kernel function, and the parameters of each sub-kernel function are initialized based on the statistical characteristics of the training dataset.

[0089] During the training of the deep Gaussian process model, the parameters of each sub-kernel function are optimized simultaneously, enabling the hybrid kernel function to capture the nonlinear correlation between different field parameters and density.

[0090] Specifically, the construction of the hybrid agent model in this step needs to be based on the training dataset and follow the progressive logic of model building, training optimization, correction module construction and fusion integration to ensure that the model can accurately capture the complex nonlinear relationship between the four field parameters and density.

[0091] Generally, the construction of a deep Gaussian process model needs to match the feature dimension of the training dataset. The input layer dimension is set to the number of features in the training dataset, which fully covers all key features of the geological stress field, sea state flow field, particle flow field, pore water seepage field and construction parameters. The hidden layer adopts a multi-layer progressive structure design, with the number of nodes in each layer decreasing according to a preset ratio to gradually extract higher-order nonlinear features. The output layer dimension is set to 1, which directly corresponds to the density prediction value, realizing the mapping from input features to density.

[0092] The nonlinear characteristics of different field parameters are captured using a hybrid kernel function, as follows:

[0093] Based on the input feature types of the deep Gaussian process model, the Matérn kernel function is selected as the corresponding sub-kernel function for the differentiated nonlinear features of the soil layer parameters of the geological stress field, the wave parameters of the sea state flow field, the particle contact parameters of the particle flow field, the pressure parameters of the pore water seepage field, and the air pressure velocity parameters of the construction parameters.

[0094] The sub-kernel functions are combined in a product form to form a hybrid kernel function, the expression of which is: ,in, For hybrid kernel functions, and The input feature vectors correspond to two different combinations of four-field parameters and construction parameters, respectively. This represents the Matérn sub-kernel function corresponding to the geological stress field characteristics. The Matérn sub-kernel function corresponding to the sea state flow field characteristics. The Matérn subkernel function corresponding to the characteristics of the particulate flow field. The Matérn sub-kernel function corresponding to the characteristics of the pore water seepage field. The Matérn sub-kernel functions are defined for the construction parameter characteristics. Based on the statistical characteristics of the training dataset, parameters such as length scale and amplitude are initialized for each sub-kernel function. During the training of the deep Gaussian process model, these sub-kernel function parameters are optimized synchronously, enabling the hybrid kernel function to accurately characterize the nonlinear relationship between different field parameters and compaction.

[0095] During the model training phase, based on the nonlinear features captured by the hybrid kernel function, the Adam optimization algorithm is used to minimize the prediction error loss function to improve the model fitting accuracy. The loss function is in the form of mean squared error, and its expression is: ,in, This is the loss value. The number of samples in the training dataset. For the first The measured density value of each sample. For the first The deep Gaussian process model predicts the compactness of each sample. The training dataset is input into the deep Gaussian process model, and the model parameters and kernel function parameters are iteratively updated until the fitting accuracy of the validation set meets the preset requirements, thus completing the model training.

[0096] The fractional-order PID correction module was constructed by determining the core parameters, including the proportional coefficient, through a joint tuning method combining the Ziegler-Nichols method and particle swarm optimization. Integral coefficient Differential coefficients and fractional order The correction logic is designed based on the density prediction deviation, and the correction output formula is as follows: ,in, The corrected density. This represents the predicted compactness output by the deep Gaussian process model. The density prediction deviation is the difference between the target density and the actual density. The difference, This is a Caputo fractional integral operator used to accumulate bias and compensate for system lag. The Caputo fractional differential operator is used to predict the trend of deviation changes and suppress fluctuations. By substituting the deviation between the predicted compactness output by the deep Gaussian process model and the target compactness into the above formula, the corrected compactness can be obtained, thus compensating for the prediction deviation.

[0097] The trained deep Gaussian process model is fused with a fractional-order PID correction module in a series manner. Specifically, the output of the deep Gaussian process model serves as the input to the fractional-order PID correction module, which processes the output to obtain the final density output, thus forming a hybrid surrogate model. This fusion method fully leverages the nonlinear fitting capability of the deep Gaussian process model and the bias correction capability of the fractional-order PID, improving the accuracy and stability of the model output.

[0098] S4. Based on the hybrid agent model, the construction parameters including the air pressure inside the pipe and the pipe pulling speed are solved in real time, and the construction operation is dynamically adjusted according to the construction parameters to complete the underwater compaction sand pile.

[0099] Furthermore, in step S4, the construction operation is dynamically adjusted based on the construction parameters, specifically including the following steps:

[0100] By combining the coupling relationship, an objective function containing a compaction deviation control term and a parameter adjustment smoothing term is constructed;

[0101] Define the upper and lower limits of the air pressure inside the tube, the range of the tube removal speed, and the allowable value of the compaction to form constraint conditions;

[0102] The real-time collected four-field coupling data is input at a preset cycle and then input into the hybrid surrogate model to obtain the density prediction value.

[0103] Based on the predicted compaction value, combined with the objective function and constraints, a robust optimization algorithm is used to iteratively solve the construction parameters, including the air pressure inside the pipe and the pipe pulling speed, and the construction operation is dynamically adjusted in real time according to the construction parameters.

[0104] Furthermore, a robust optimization algorithm is used to iteratively solve for the construction parameters, including the air pressure inside the pipe and the pipe pulling speed. This process specifically includes the following steps:

[0105] Based on the predicted density, objective function, and constraints, and combining the robust expectation improvement criterion and Monte Carlo sampling method, an optimization objective with robustness and constraints is constructed.

[0106] The L-BFGS algorithm is used to iteratively solve the optimization objective. In each iteration, the objective value is updated based on the hybrid surrogate model. The iteration stops when the preset convergence accuracy or number of iterations is met. The output includes construction parameters including the air pressure inside the pipe and the pipe pulling speed.

[0107] Specifically, the process of solving construction parameters and dynamically adjusting construction operations based on the hybrid proxy model in this step relies on the previously established coupling relationship between compaction and construction parameters, as well as the hybrid proxy model. It follows a logical progressive process of goal and constraint construction, real-time data input, robust optimization solution, and dynamic construction adjustment to ensure the real-time and accuracy of construction parameters and guarantee the quality of pile formation.

[0108] The objective function needs to be constructed in close conjunction with the coupling relationship between compaction and construction parameters obtained above, taking into account both pile quality and construction stability. A compaction deviation control term and a parameter adjustment smoothing term are introduced, and its expression is as follows: ,in, The objective function value, The air pressure inside the pipe, For the tube removal speed, For construction depth, For the pile length, Depth of output for hybrid proxy model Predicted density value, For target density, and These are weighting coefficients that respectively adjust the smoothness of the pipe pressure and pipe pulling speed. Integrating the coupling relationship into the objective function ensures that the optimization direction always revolves around achieving the required compaction and smooth construction. Different inputs... and Combining these values ​​yields the corresponding objective function value, providing a criterion for subsequent optimization.

[0109] The setting of constraints needs to be combined with the actual construction capabilities and pile quality requirements of the project, and the boundary limits of each parameter should be clearly defined. Generally, the upper and lower limits of the air pressure inside the pipe are determined based on the characteristics of the sand and the strength of the casing, to avoid the casing being damaged due to excessively high air pressure or the sand not being compacted due to excessively low air pressure; the range of pipe pulling speed is set based on construction efficiency and the adequacy of sand compaction, to prevent insufficient compaction due to excessively high speed or construction progress due to excessively slow speed; the allowable value of compaction is set to be no less than the minimum compaction requirement of the engineering design, forming a complete set of constraints to ensure that the parameters of the optimization solution meet the actual needs of the project.

[0110] Real-time data input must be performed according to a preset cycle, and the cycle setting must balance data timeliness and computational efficiency. The collected real-time four-field coupled data must be preprocessed according to the S1 method, including timestamp alignment, outlier removal, and noise filtering, before being input into the hybrid proxy model. The hybrid proxy model performs inference calculations on the input data based on the trained and optimized parameters, and outputs the density prediction value at the corresponding construction depth, providing the core input basis for subsequent optimization solutions.

[0111] Robust optimization solutions require the construction of an optimization objective with robustness and constraints based on the predicted compaction value, the aforementioned objective function and constraints, and the robust expected improvement criterion and Monte Carlo sampling method, to address the impact of random disturbances in soil parameters. The expression for the robust expected improvement function is: ,in, For robust expected improvement value, For soil layer parameters, random disturbance variables, Represents the effect of random disturbance variables Expectations This represents the minimum objective function value currently obtained. A large number of random perturbation samples are generated using the Monte Carlo sampling method to approximate the expected value. The constraints are then transformed into penalty terms and incorporated into the optimization objective, forming a solvable optimization objective with robustness and constraints.

[0112] The L-BFGS algorithm was then used to iteratively solve the optimization objective. During each iteration, the current parameter combination was input into the hybrid surrogate model to update the density prediction and objective function values, gradually approaching the optimal parameters. The iteration stopped and the current pipe pressure was locked when the difference between the objective function values ​​of two adjacent iterations was less than the preset convergence accuracy, or when the number of iterations reached the preset maximum. With tube removal speed Output as the optimal construction parameters.

[0113] After the optimal construction parameters are output, they are sent to the construction control system in real time. The opening of the air pressure valve is dynamically adjusted to regulate the air pressure inside the pipe, and the speed of the pipe-pulling winch driven by the servo motor is adjusted to regulate the pipe-pulling speed. At the same time, the process of real-time data acquisition, hybrid proxy model prediction, robust optimization solution, and parameter adjustment is repeated at preset depth intervals until the construction of a single underwater compacted sand pile is completed. This dynamic adjustment process can respond in real time to changes in the four parameters during construction, ensuring that the pile compaction at each depth meets the design requirements and improving the stability of the pile quality.

[0114] Example 2:

[0115] In a second embodiment of the present invention, the present invention provides a precise pile-forming control system for underwater compaction sand piles, such as... Figure 2 As shown, it includes the following modules:

[0116] Exploration and data acquisition module: used to conduct geological and marine condition surveys of the construction area, and simultaneously acquire geological stress field, marine flow field, particle flow field, pore water seepage field and construction parameter data of the construction area, and after preprocessing, form four-field coupled data;

[0117] The joint equation fusion module is used to construct the control equations for each field based on the four coupled data, obtain the coupling relationship between compaction and construction parameters, and fuse simulation data and field test pile data to construct a training dataset.

[0118] The optimization module is used to build a fusion model that combines a deep Gaussian process with a fractional-order PID based on the training dataset, and to train and optimize it to obtain a hybrid surrogate model.

[0119] Construction Solving Module: This module is used to solve construction parameters, including the air pressure inside the pipe and the pipe pulling speed, in real time based on a hybrid proxy model. It also dynamically adjusts the construction operation according to the construction parameters to complete the underwater compaction sand pile formation.

[0120] In the underwater compaction sand pile construction of a cross-sea bridge foundation reinforcement project in a coastal area, the geological strata in the construction area consisted of alternating layers of silty clay and loose sand, with poor soil layer parameter stability. Furthermore, the area was significantly affected by tides, and the construction period frequently encountered wind and waves, causing fluctuations in the ship's attitude, resulting in complex construction conditions. Traditional construction methods did not consider the coupled effects of geological stress field, sea state current field, etc., relying solely on fixed construction parameters. This led to uneven pile compaction after pile formation, and the composite foundation bearing capacity did not meet design requirements, necessitating additional pile reinforcement work, which not only increased construction costs but also extended the construction period. To solve these problems, the underwater compaction sand pile precision pile formation control system provided by this invention was adopted, the architecture of which is as follows: Figure 2 As shown. The specific implementation process of this system is as follows:

[0121] First, the survey and acquisition module deploys multiple types of sensors on the construction vessel and casing to cover the collection needs of geological, sea state, and construction parameters. It also performs borehole sampling in the construction area and obtains key geological parameters through laboratory tests. Simultaneously, it monitors sea state data and construction parameters such as in-tube air pressure and pipe extraction speed. The collected data undergoes timestamp alignment, outlier removal, and filtering preprocessing to form standardized four-field coupled data.

[0122] Subsequently, the joint fusion module constructs control equations for each field based on the four-field coupled data output by the exploration and acquisition module, and obtains the coupling relationship between compaction and construction parameters. Simulation data is generated by simulating multiple working conditions through a two-way coupling simulation tool, and the measured data of on-site test piles are integrated. The training set, validation set and test set are divided according to a preset ratio to complete the construction of a training dataset covering multiple working conditions.

[0123] Next, based on the training dataset output by the joint fusion module, the optimization module constructs a hybrid proxy model that integrates deep Gaussian process and fractional-order PID. The optimization algorithm iteratively trains the model parameters, optimizes the kernel function and key PID parameters, and ensures that the model prediction accuracy meets the preset requirements. After the trained model is lightweighted, it is deployed to the edge computing server to ensure that the inference response speed meets the real-time control requirements, thus achieving efficient approximation of the complex relationship of four-field coupling.

[0124] Finally, the construction module receives real-time four-field coupled data output by the exploration and acquisition module at a preset cycle, inputs it into the hybrid proxy model to obtain the density prediction value, and combines the preset objective function and constraints to iteratively solve the optimal pipe pressure and pipe pulling speed through a robust optimization algorithm. The results are then sent to the construction execution mechanism in real time to dynamically adjust the opening of the air pressure valve and the speed of the pipe pulling winch. The optimal parameters are updated at preset depth intervals to complete the underwater compaction sand pile, thereby improving the accuracy of pile formation under complex geological and marine conditions.

[0125] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for precise pile formation control of underwater compaction sand piles, characterized in that, Includes the following steps: S1. Conduct geological and marine condition surveys of the construction area, and simultaneously collect data on geological stress field, marine flow field, particle flow field, pore water seepage field and construction parameters of the construction area. After preprocessing, the data are coupled into four fields. S2. Based on the four-field coupled data, construct the control equations for each field, solve them simultaneously to obtain the coupling relationship between compaction and construction parameters, and integrate simulation data and field test pile data to construct a training dataset. S3. Construct a fusion model combining deep Gaussian process and fractional-order PID based on the training dataset, and train and optimize it to obtain a hybrid proxy model; S4. Based on the hybrid agent model, the construction parameters including the air pressure inside the pipe and the pipe pulling speed are solved in real time, and the construction operation is dynamically adjusted according to the construction parameters to complete the underwater compaction sand pile.

2. The method for precise pile formation control of underwater compaction sand piles according to claim 1, characterized in that: In step S1, the formation of four-field coupled data specifically includes the following steps: By combining borehole sampling with laboratory tests, the effective unit weight, cohesion, and coefficient of earth pressure at rest of the soil layer were obtained. The distribution characteristics of the effective unit weight of the soil layer were statistically analyzed to form geological stress field data. Collect meteorological and hydrological data of the construction area, monitor wave amplitude, wavelength, wave period and tidal characteristics, establish the correlation between sea state and ship attitude, and form sea state flow field data; The moisture content of the sand at the outlet of the sand chamber and the height of the sand surface inside the casing are collected simultaneously to form particle flow field data; Collect pore water pressure in the casing to generate pore water seepage field data; Simultaneously collect data on soil lateral pressure, ship attitude, wave conditions, sand level inside the pipe, air pressure inside the pipe, pipe extraction speed, and longitudinal strain of the pile to form construction parameter data; The geological stress field, sea state flow field, particle flow field, pore water seepage field, and construction parameter data are timestamped based on GPS second pulse signals. Abnormal data are removed using the 3σ criterion, and data noise is reduced through filtering algorithms to obtain standardized four-field coupled data.

3. The method for precise pile formation control of underwater compaction sand piles according to claim 1, characterized in that: In step S2, constructing the training dataset specifically includes the following steps: Based on the four-field coupled data, single-field control equations were constructed for the corresponding geological stress field, sea state flow field, particle flow field, and pore water seepage field. Based on the interaction relationship of the parameters of each single field control equation, the single field control equations are integrated to establish the coupling relationship between compaction and construction parameters. Based on the coupling relationship, a two-way coupling simulation tool is used to simulate different soil layer types, wave levels, air pressures, and pipe pulling speeds, generating simulation data containing four field parameters, construction parameters, and corresponding compaction. Collect four-field coupling data, construction parameter data, and actual compaction data during the on-site pile test process, merge them with simulation data, and divide them into training set, validation set, and test set according to a preset ratio to form a training dataset.

4. The method for precise pile formation control of underwater compaction sand piles according to claim 3, characterized in that: The construction of the single-field governing equations corresponding to the geological stress field, sea state flow field, particle flow field, and pore water seepage field respectively includes the following steps: Based on geological stress field data and combined with the random disturbance characteristics of soil layer parameters, stress calculation equations are constructed. Based on sea state and current field data, and combined with the principles of ship dynamics and wave mechanics, a coupled equation for ship attitude and wave parameters is constructed. Based on particle flow field data, and combined with the bonding effect and memory characteristics between sand particles, a correlation equation between particle contact force and density evolution is constructed. Based on pore water seepage field data and combined with the seepage law, a dynamic variation equation for porosity and pore water pressure is constructed.

5. The method for precise pile formation control of underwater compaction sand piles according to claim 1, characterized in that: In step S3, obtaining the hybrid agent model specifically includes the following steps: Based on the feature dimensions of the training dataset, a deep Gaussian process model containing an input layer, a hidden layer, and an output layer is constructed, and a hybrid kernel function is used to capture the nonlinear features of different field parameters. Based on nonlinear characteristics, an optimization algorithm is used to minimize the prediction error loss function, and a deep Gaussian process model is iteratively trained and the kernel function parameters are optimized until the fitting accuracy of the validation set meets the preset requirements. The proportional coefficient, integral coefficient, derivative coefficient, and fractional order are determined by a joint tuning method. Based on the density prediction deviation, the correction logic is designed to construct a fractional-order PID correction module. The trained deep Gaussian process model is fused with a fractional-order PID correction module to form a hybrid surrogate model.

6. The method for precise pile formation control of underwater compaction sand piles according to claim 5, characterized in that: The method of using hybrid kernel functions to capture the nonlinear characteristics of different field parameters specifically includes the following steps: Based on the input feature types of the deep Gaussian process model, the Matérn kernel function is selected as the corresponding sub-kernel function for different nonlinear characteristics of geological stress field, sea state flow field, particle flow field, pore water seepage field and construction parameters. The sub-kernel functions are combined in a product form to form a hybrid kernel function, and the parameters of each sub-kernel function are initialized based on the statistical characteristics of the training dataset. During the training of the deep Gaussian process model, the parameters of each sub-kernel function are optimized simultaneously, enabling the hybrid kernel function to capture the nonlinear correlation between different field parameters and density.

7. The method for precise pile formation control of underwater compaction sand piles according to claim 1, characterized in that: In step S4, the dynamic adjustment of construction operations based on construction parameters specifically includes the following steps: By combining the coupling relationship, an objective function containing a compaction deviation control term and a parameter adjustment smoothing term is constructed; Define the upper and lower limits of the air pressure inside the tube, the range of the tube removal speed, and the allowable value of the compaction to form constraint conditions; The real-time collected four-field coupling data is input at a preset cycle and then input into the hybrid surrogate model to obtain the density prediction value. Based on the predicted compaction value, combined with the objective function and constraints, a robust optimization algorithm is used to iteratively solve the construction parameters, including the air pressure inside the pipe and the pipe pulling speed, and the construction operation is dynamically adjusted in real time according to the construction parameters.

8. The method for precise pile formation control of underwater compaction sand piles according to claim 7, characterized in that: The process of iteratively solving for construction parameters, including the air pressure inside the pipe and the pipe pulling speed, using a robust optimization algorithm specifically includes the following steps: Based on the predicted density, objective function, and constraints, and combining the robust expectation improvement criterion and Monte Carlo sampling method, an optimization objective with robustness and constraints is constructed. The L-BFGS algorithm is used to iteratively solve the optimization objective. In each iteration, the objective value is updated based on the hybrid surrogate model. The iteration stops when the preset convergence accuracy or number of iterations is met. The output includes construction parameters including the air pressure inside the pipe and the pipe pulling speed.

Images