Energy-saving type water pile foundation mud circulating treatment method and system

Abstract

The invention relates to the crossing field of civil engineering and intelligent control, discloses an energy-saving type water pile foundation mud circulating treatment method and system, and aims to solve the problems of high energy consumption, poor adaptability and unstable treatment effect in the prior art. The method specifically comprises the following steps: collecting a slurry inlet state, an equipment operation state and an outlet quality parameter in real time; constructing a multi-dimensional real-time state characteristic matrix and establishing a dynamic association model; in combination with a weighted multi-objective optimization function which aims at minimizing the total power of the system and minimizing the deviation degree of the purification index, a cooperative regulation and control instruction is generated through an optimization algorithm; the slurry pump, the vibrating screen, the centrifugal machine and the adjusting valve are driven to operate cooperatively, and closed-loop self-adaptive control is achieved. Through full-link collaborative frequency conversion regulation and intelligent optimization, the energy consumption is remarkably reduced, and the slurry treatment quality stability and the system adaptive capacity are improved.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of civil engineering and intelligent control, and specifically relates to an energy-saving method and system for the recycling of mud for underwater pile foundations. Background Technology

[0002] As infrastructure construction continues to extend into water areas, tidal flats, and complex geological regions, offshore pile foundation construction has become a crucial component of major projects such as bridges, wharves, and offshore wind power. Drilling mud, as the core medium in the pile foundation drilling process, performs multiple functions including carrying slag, protecting the drilling wall, and cooling the drilling tools. Its recycling efficiency directly affects construction quality, progress, and environmental impact.

[0003] Traditional offshore mud circulation systems mostly employ fixed-flow pumping and a single sedimentation separation mode, relying on the experience of construction personnel to set discharge volume, concentration, and recirculation ratio. They lack the ability to dynamically perceive and coordinately control mud performance parameters such as density, viscosity, and sand content, as well as operational variables such as drilling depth, formation changes, and water level fluctuations. This traditional static operating mechanism frequently leads to mud performance imbalances, easily causing problems such as borehole collapse, pipe blockage, and insufficient cuttings carrying capacity. It also results in significant clean water replenishment and waste mud transportation, significantly increasing energy consumption and disposal costs.

[0004] In view of the technical defects in the existing technology, the existing mud treatment system is difficult to balance construction safety, economy and environmental protection in complex water conditions. There is an urgent need for an intelligent mud recycling treatment solution that deeply integrates multi-parameter sensing, energy-saving drive and process synergy. Summary of the Invention

[0005] To address the technical problems of high energy consumption, poor adaptability, and unstable treatment effects in existing underwater pile foundation mud treatment systems, this invention provides an energy-saving underwater pile foundation mud circulation treatment method and system. By constructing a dynamic correlation model between the real-time state of the mud and the operating parameters of the treatment equipment, and performing global collaborative control based on a multi-objective optimization function, the system achieves precise real-time optimization of energy consumption throughout the entire treatment process while ensuring the quality of mud purification, significantly improving the system's adaptability and operational economy.

[0006] According to one aspect of the present invention, an energy-saving method for treating mud circulation in underwater pile foundations is provided, comprising the following steps: The system collects real-time inlet status parameters of the mud to be treated during the construction of underwater pile foundations, operational status parameters of key treatment equipment in the mud circulation system, and outlet quality parameters of the treated mud. Inlet status parameters include inlet flow rate, inlet density, inlet viscosity, and inlet mud particle size distribution. Operational status parameters include input current and speed of the mud pump, input current and vibration frequency of the vibrating screen motor, inlet pressure of the desander, inlet pressure of the desilter, and differential speed and main motor input current of the centrifuge. Outlet quality parameters include outlet density, outlet viscosity, and outlet solids content.

[0007] Based on all collected parameters, a multidimensional real-time state feature matrix is ​​constructed through timestamp alignment and data normalization. Using the multidimensional real-time state feature matrix, a dynamic correlation model is established to characterize the nonlinear mapping relationship between inlet state parameters, operating state parameters and outlet quality parameters.

[0008] A weighted multi-objective optimization function is established with the objectives of minimizing the total system power consumption and minimizing the deviation of the mud purification index. The total system power consumption is calculated by summing the electrical parameters in the operating status parameters of each key treatment device, and the deviation of the mud purification index is calculated by comparing the real-time value of the outlet quality parameter with the preset target value.

[0009] The multidimensional real-time state feature matrix is ​​used as the input of the dynamic correlation model, and combined with the weighted multi-objective optimization function, the optimal set of coordinated control command matrix is ​​generated by iteratively solving through the preset optimization algorithm. The coordinated control command matrix includes the target speed of the mud pump, the target frequency of the vibrating screen motor, the target differential speed of the centrifuge, and the target opening degree of the flow regulating valve between each level of processing unit.

[0010] The coordinated control command matrix is ​​transmitted to the corresponding equipment execution units in the mud circulation treatment system. The equipment execution units drive each key processing equipment to operate according to the parameters set by the coordinated control command matrix, forming a closed-loop adaptive control of the mud treatment process.

[0011] As one embodiment of the present invention, real-time acquisition of inlet status parameters specifically includes: A Coriolis mass flow meter is installed on the main inlet pipe of the mud circulation treatment system to simultaneously obtain the inlet flow rate and inlet density of the mud to be treated; an online rotational viscometer is installed on the main inlet pipe to obtain the inlet viscosity of the mud to be treated; and an ultrasonic particle size analyzer is installed on the main inlet pipe to obtain the particle size distribution data of the mud and sand particles in the mud to be treated.

[0012] As one embodiment of the present invention, real-time acquisition of operating status parameters specifically includes: High-precision power monitoring modules are installed on the power supply circuits of the mud pump, vibrating screen vibrating motor, and centrifuge main motor to collect their three-phase input current and voltage in real time and calculate instantaneous power; the real-time speed of the mud pump, the real-time frequency of the vibrating screen vibrating motor, and the real-time differential speed of the centrifuge are directly read through the encoder interface or speed sensor built into the frequency converter driver equipped with each device; pressure transmitters are installed on the hydrocyclone inlet pipes of the desander and deslimer to obtain their inlet pressure.

[0013] As one embodiment of the present invention, the dynamic correlation model is specifically a recurrent neural network model based on a long short-term memory network. This model learns by using a large amount of multidimensional real-time state feature matrix data collected in history during the offline training phase, thereby accurately capturing the dynamic influence relationship between mud state changes and treatment effect and system energy consumption. The input layer of the model receives the multidimensional real-time state feature matrix, and the output layer outputs the predicted total power consumption value of the system and the outlet quality parameter value.

[0014] Furthermore, after establishing the weighted multi-objective optimization function, the method also includes: The mud purification index deviation penalty function compares the real-time density, viscosity, and solids content of the treated mud with a preset target range. When the detected value exceeds the target range, a penalty term is calculated using a preset piecewise nonlinear function based on the extent of the deviation. The weight coefficients in the weighted multi-objective optimization function are adaptively adjusted according to the different working stages of the underwater pile foundation construction, such as the excavation stage, the hole cleaning stage, or the pouring stage, to prioritize the protection of the core technical indicators at specific stages.

[0015] As one embodiment of the present invention, an iterative solution is performed using a preset optimization algorithm. Specifically, a sequential quadratic programming algorithm is adopted. In each control cycle, based on the current real-time state, the dynamic correlation model is linearized and approximated. Combined with a weighted multi-objective optimization function, the coordinated control instruction matrix that minimizes the objective function value is obtained.

[0016] According to another aspect of the present invention, an energy-saving underwater pile foundation mud circulation treatment system is provided, comprising: The multi-dimensional real-time state perception module is configured to collect in real-time inlet state parameters of the mud to be treated during the construction of underwater pile foundations, operating state parameters of each key treatment equipment in the mud circulation treatment system, and outlet quality parameters of the treated mud. It also performs timestamp alignment and data normalization on all collected parameters to construct a multi-dimensional real-time state feature matrix.

[0017] The dynamic correlation modeling and optimization decision-making module, connected to the multi-dimensional real-time state perception module, is used to establish a dynamic correlation model representing the nonlinear mapping relationship between inlet state parameters, operating state parameters and outlet quality parameters. It also establishes a weighted multi-objective optimization function with the objectives of minimizing the total power consumption of the system and minimizing the deviation of the mud purification index. Finally, based on the real-time input multi-dimensional real-time state feature matrix, the optimization function is solved through an optimization algorithm to generate a set of optimal collaborative control command matrices.

[0018] The multi-device collaborative execution module receives the collaborative control instruction matrix from the dynamic correlation modeling and optimization decision-making module, converts it into specific control signals, and sends them to the equipment execution units such as mud pumps, vibrating screens, centrifuges, and flow regulating valves in the mud circulation treatment system to drive their collaborative operation.

[0019] Furthermore, the multi-dimensional state real-time perception module specifically includes: The system includes a Coriolis mass flow meter, an online rotational viscometer, and an ultrasonic particle size analyzer installed on the main inlet pipe of the mud circulation treatment system; power monitoring modules installed on the power supply circuits of the mud pump, vibrating screen motor, and centrifuge main motor; speed and frequency sensors associated with the mud pump, vibrating screen, and centrifuge, respectively; and pressure transmitters installed on the inlet pipes of the desander and desilter, respectively. All sensors and monitoring modules communicate with the dynamic correlation modeling and optimization decision-making module via an industrial fieldbus.

[0020] As one embodiment of the present invention, the dynamic association modeling and optimization decision module is specifically an industrial control computer in terms of hardware. It adopts an embedded structure, with a main frequency of not less than 2.4 GHz and a memory of not less than 16 GB. It runs a pre-set control software program, which solidifies the dynamic association model, the weighted multi-objective optimization function, and the sequential quadratic programming optimization algorithm.

[0021] Furthermore, the multi-device collaborative execution module specifically includes: The system includes: a variable frequency drive connected to the mud pump drive motor; a variable frequency drive connected to the vibrating screen motor; a dual variable frequency DC bus drive system connected to the main and auxiliary motors of the centrifuge for precise differential speed control; and flow regulating valves and their actuators connected to pipelines between processing units at each level. All variable frequency drives and valve actuators are equipped with industrial fieldbus communication interfaces to receive and execute collaborative control commands from the dynamic correlation modeling and optimization decision-making module.

[0022] As one embodiment of the present invention, the system also includes a human-computer interaction and monitoring terminal, which is connected to the dynamic correlation modeling and optimization decision module via industrial Ethernet. The terminal is used to set the target value range of mud outlet quality, set the optimization weight coefficient under different working conditions, display the operating parameters and energy consumption curves of each part of the system in real time, and provide a manual intervention control interface.

[0023] In summary, this application includes at least one of the following beneficial technical effects: (1): By constructing a comprehensive multi-dimensional state perception system, this invention can accurately grasp the full picture of mud properties and system operation status in real time, providing a data foundation for refined control; the established dynamic correlation model reveals the complex nonlinear relationship between various system parameters, realizes accurate prediction of system behavior, and overcomes the lag and blindness of traditional control methods.

[0024] (2): Based on the weighted multi-objective optimization optimization decision-making mechanism, this application quantifies and unifies the two core objectives of energy saving and consumption reduction and ensuring processing quality, finds the globally optimal combination of operating parameters, and realizes the control paradigm shift from "passive adaptation" to "active optimization".

[0025] (3): In this application, by implementing full-link coordinated frequency conversion control of core energy-consuming equipment such as mud pumps, vibrating screens, and centrifuges, the total power output of the system is always precisely matched with the actual processing load, which fundamentally eliminates redundant energy consumption and achieves deep energy saving in system operation.

[0026] (4): The entire system constitutes a closed-loop adaptive control, which can automatically respond to the drastic fluctuations in mud properties caused by geological changes, continuously ensure the quality stability of the treated mud, improve construction efficiency, reduce the risk of work stoppage due to mud problems, reduce reliance on operator experience, and improve the automation and intelligence level of the entire processing process. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the overall technical solution of the present invention; Figure 2 This is a logical framework diagram of multi-dimensional state real-time perception, dynamic modeling optimization, and multi-device collaborative execution in this invention. Detailed Implementation

[0028] The energy-saving underwater pile foundation mud circulation treatment method and system disclosed in this application is based on the construction of a full-dimensional real-time perception system covering the mud inlet status, equipment operation status and outlet quality, and on this basis, a dynamic correlation model and multi-objective optimization function are established to achieve global coordinated control of the mud treatment process through closed-loop adaptive control.

[0029] The following, in conjunction with the appendix Figure 1-2 This application provides a detailed description of the energy-saving underwater pile foundation mud circulation treatment method and system.

[0030] First, the energy-saving method for treating mud circulation in underwater pile foundations includes the following steps: S1 collects in real time the inlet status parameters of the mud to be treated, the operating status parameters of each key treatment equipment in the mud circulation treatment system, and the outlet quality parameters of the treated mud during the construction of underwater pile foundations. S2, based on all collected parameters, constructs a multi-dimensional real-time state feature matrix through timestamp alignment and data normalization processing; S3. Using a multi-dimensional real-time state feature matrix, a dynamic correlation model is established to characterize the nonlinear mapping relationship between inlet state parameters, operating state parameters, and outlet quality parameters. S4. Establish a weighted multi-objective optimization function with the objectives of minimizing the total power consumption of the system and minimizing the deviation of the mud purification index. S5 takes the multi-dimensional real-time state feature matrix as the input of the dynamic correlation model, and combines it with a weighted multi-objective optimization function to iteratively solve the problem through a preset optimization algorithm to generate the optimal collaborative control instruction matrix. S6 transmits the coordinated control command matrix to the corresponding equipment execution units within the mud circulation treatment system. The equipment execution units then drive each key processing device to operate according to the parameters set by the coordinated control command matrix, forming a closed-loop adaptive control of the mud treatment process.

[0031] In step S1, the real-time collected data is divided into three categories: inlet status parameters, running status parameters, and outlet quality parameters.

[0032] The acquisition of inlet state parameters is accomplished by integrating a high-precision sensor on the main inlet pipe of the mud circulation treatment system. The specific installation and acquisition are as follows: the Coriolis mass flow meter is installed on the main inlet pipe. Its measurement principle is based on the Coriolis force generated when the fluid flows in the vibrating tube. This force is proportional to the mass flow rate. At the same time, the fluid density can be deduced by measuring the change in the natural frequency of the vibrating tube, thereby simultaneously obtaining the inlet flow rate and inlet density.

[0033] An online rotational viscometer uses a motor to drive a rotor to rotate at a constant speed in the mud, and measures the torque required to maintain that speed. This torque is linearly related to the mud viscosity, thus obtaining the inlet viscosity.

[0034] An ultrasonic particle size analyzer is used to emit high-frequency ultrasonic pulses into the mud and receive the echo signals scattered by the mud and sand particles. By analyzing the spectral attenuation characteristics and propagation time difference of the echoes, and combining the Mie scattering theory, the particle size distribution data of the mud and sand particles can be calculated. The Mie scattering theory is a classic optical scattering theory and one of the fundamental principles in the field of particle measurement. The typical particle size range covers 5 micrometers to 200 micrometers, with a resolution of 1 micrometer.

[0035] All three types of sensors have industrial-grade protection and a uniform sampling frequency of 10 Hz to ensure data continuity over time.

[0036] The collection of operating status parameters covers five key pieces of equipment: mud pump, vibrating screen, sand remover, desliming device, and centrifuge.

[0037] High-precision power monitoring modules are installed on the power supply circuits of the mud pump, vibrating screen motor, and centrifuge main motor. These modules use Hall effect current sensors and resistive voltage divider networks to collect three-phase input current and voltage in real time at a sampling frequency of 1kHz. The instantaneous power of each device is obtained using the instantaneous power calculation formula P=√3×U×I×cosφ (where U is the effective value of line voltage, I is the effective value of line current, and cosφ is the power factor).

[0038] The real-time speed of the mud pump is directly read through the encoder interface built into its frequency converter driver, with an encoder resolution of 1024 pulses per revolution. The real-time frequency of the vibrating screen motor is provided by the output frequency register of the frequency converter driver. The real-time differential speed of the centrifuge is calculated by the difference between the encoder signals of the main motor and the auxiliary motor. Piezoresistive pressure transmitters with a range of 0 to 2 MPa, an accuracy of 0.5% of full scale, and a sampling frequency of 5 Hz are installed on the hydrocyclone inlet pipes of the desander and deslimer to monitor the working pressure of the hydrocyclone, which directly reflects the hydrocyclone separation efficiency.

[0039] For the outlet quality parameters, the data are collected on the final outlet pipe of the mud circulation treatment system. The same Coriolis mass flow meter, online rotational viscometer and ultrasonic particle size analyzer as those on the inlet pipe are used to obtain the outlet density, outlet viscosity and outlet solids content, respectively.

[0040] The outlet solids content is directly calculated from the particle volume concentration measured by the ultrasonic analyzer, or indirectly calculated from the density difference between the inlet and outlet combined with the initial mud mix ratio. The two methods of direct conversion and cross-conversion are cross-validated to ensure the reliability of the data.

[0041] After the parameter collection is completed, step S2 of this application is to first align the timestamps of all collected parameter data.

[0042] Since various sensors have different sampling frequencies, a hardware synchronization triggering mechanism based on the system master clock is adopted to ensure that all data points have a unified time reference.

[0043] The data was then normalized to eliminate dimensional differences.

[0044] Specifically, for physical quantities such as flow rate, density, and pressure, the maximum and minimum values ​​are normalized, and the formula is x'=(x-x_min) / (x_max-x_min), where x_min and x_max are the minimum and maximum values ​​of the parameter in the historical operating data, respectively.

[0045] For electrical parameters such as current and power, Z-score standardization is used, with the formula x'=(x-μ) / σ, where μ and σ are the mean and standard deviation of the parameter under steady-state conditions, respectively.

[0046] The normalized data are arranged in time series to form a multidimensional real-time state feature matrix with dimensions N×M, where N is the number of time steps, M is the total number of features, and M is equal to the sum of the four dimensions of the inlet state parameters, the seven dimensions of the running state parameters, and the four dimensions of the outlet quality parameters, for a total of 14 dimensions.

[0047] In step S3, a dynamic correlation model is established using a multidimensional real-time state feature matrix to characterize the nonlinear mapping relationship between inlet state parameters, operating state parameters, and outlet quality parameters.

[0048] The constructed dynamic association model employs a recurrent neural network model based on a long short-term memory network. The recurrent neural network model consists of an input layer, two stacked LSTM hidden layers, and an output layer.

[0049] The input layer receives data for each time step of the multidimensional real-time state feature matrix.

[0050] The number of units in the LSTM hidden layer was set to 64 and 32, respectively. The tanh activation function and gating mechanism were used. The tanh activation function was used to generate candidate memories, with an output range of [-1, 1]. The three gating units of the gating mechanism used the sigmoid activation function. The weight matrices of the forget gate, input gate, and output gate were learned during offline training using the backpropagation algorithm.

[0051] The offline training data comes from at least three months of historical operation records, covering complete working conditions under different geological conditions and different construction stages, with a total data volume of no less than 1 million time step samples.

[0052] During offline training, mean squared error is used as the loss function, and an early stopping mechanism is introduced to prevent overfitting. After training, the model is stored in the memory of the industrial control computer. During online inference, the model receives the latest multidimensional real-time state feature matrix in a sliding window manner. The window length is 30 seconds, corresponding to 300 time steps (based on a 10Hz sampling rate), and outputs the predicted system state value within the next five seconds.

[0053] The output layer contains four nodes: total power consumption of the prediction system, outlet density, outlet viscosity, and outlet solids content.

[0054] In step S4, the established weighted multi-objective optimization function is defined as J = w1 × P_total + w2 × D_deviation, where P_total is the total power consumption of the system, D_deviation is the deviation of the mud purification index, and w1 and w2 are weight coefficients. The weight coefficients w1 and w2 are adaptively adjusted according to the construction conditions: during the excavation stage, w2 is set to 0.7 and w1 to 0.3 to prioritize treatment quality; during the borehole cleaning stage, w2 is increased to 0.9; during the pouring stage, when the system load is stable, w1 is increased to 0.6 to focus on energy saving.

[0055] The total power consumption of the system, P_total, is obtained by summing the instantaneous power of the mud pump, vibrating screen, and centrifuge, i.e., P_total = P_pump + P_vibrator + P_centrifuge.

[0056] The deviation of the mud purification index, D_deviation, is composed of three weighted sub-items: D_deviation = α × |ρ_out - ρ_target| / ρ_target + β × |η_out - η_target| / η_target + γ × |C_out - C_target| / C_target.

[0057] Where ρ_out, η_out, and C_out are the real-time values ​​of outlet density, viscosity, and solids content, respectively; ρ_target, η_target, and C_target are their preset target values; and α, β, and γ are the relative importance coefficients of each index, which are usually set according to construction specifications and are not fixed values. For example, during the hole cleaning stage, the weight of solids content γ is significantly increased.

[0058] When any outlet parameter exceeds the preset tolerance range (such as density deviation exceeding 5%), the mud purification index deviation penalty function is activated, multiplying D_deviation by an amplification factor greater than one. This factor increases exponentially with the deviation. Through the mud purification index deviation penalty function, the system can be ensured to have the best efficiency while maintaining the best operating quality.

[0059] In step S5, a sequential quadratic programming algorithm is used for iterative solution. This algorithm is executed once within each control cycle (cycle length is five seconds).

[0060] First, the multidimensional real-time state feature matrix at the current moment is input into the dynamic correlation model to obtain the Jacobian matrix of the system state, which is the partial derivative with respect to each control variable (slurry pump speed, vibration frequency, centrifuge differential speed, valve opening).

[0061] Then, the original nonlinear optimization problem is linearized at the current operating point, transforming it into a quadratic programming subproblem, specifically: min½Δu^THΔu+g^TΔu,stA_eqΔu=b_eq,A_ineqΔu≤b_ineq, where Δu is the increment vector of the control variable, H is the Hessian matrix approximation, g is the gradient vector, and A_eq and A_ineq are equality and inequality constraint matrices. The constraints include physical limits of the equipment (such as the upper limit of pump speed of 1,500 rpm), process safety boundaries (such as the differential speed of centrifuge must not be less than 5 rpm), and hard requirements for mud quality (such as the outlet solids content must be less than 5%).

[0062] Solving this quadratic programming problem yields the optimal control increment Δu*, which is then superimposed on the current control command to form a new collaborative control command matrix. This command matrix contains four core parameters: the target speed of the mud pump (in revolutions per minute), the target frequency of the vibrating screen motor (in Hertz), the target differential speed of the centrifuge (in revolutions per minute), and the target opening degree of the flow regulating valves between each stage of the treatment unit (in percentage).

[0063] In step S6, the coordinated control command matrix is ​​transmitted to the multi-device coordinated execution module via an industrial fieldbus (such as PROFINET or Modbus TCP). The variable frequency drive of the mud pump receives the target speed command and adjusts the output frequency through the internal PID controller to drive the motor for smooth speed change; the variable frequency drive of the vibrating screen responds to the target frequency command in the same way; the dual variable frequency DC bus drive system of the centrifuge receives the target differential speed command, the main frequency converter controls the speed of the main motor, and the auxiliary frequency converter controls the speed of the auxiliary motor, with the difference between the two precisely tracking the set value; the driver of the flow regulating valve receives the target opening command and drives the valve core to rotate to the specified position to adjust the mud distribution ratio between each processing unit. After all the execution actions are completed, the system enters the next control cycle and repeats steps S1 to S6 to form a closed-loop control.

[0064] In addition, the energy-saving underwater pile foundation mud circulation treatment system in this application includes a multi-dimensional real-time state perception module, a dynamic correlation modeling and optimization decision-making module, and a multi-device collaborative execution module.

[0065] The multi-dimensional real-time state perception module consists of the aforementioned sensors and monitoring modules. All devices are connected to the field junction box via shielded twisted-pair cables, and then aggregated to the dynamic correlation modeling and optimization decision-making module via an industrial switch.

[0066] The dynamic correlation modeling and optimization decision-making module is an embedded industrial control computer with a processor clock speed of 2.8GHz, a memory capacity of 32GB, a solid-state drive capacity of 1TB, an operating system based on a real-time Linux kernel, control software written in C++, and a model inference engine optimized based on TensorRT to ensure that the calculation latency within a five-second control cycle does not exceed 500 milliseconds.

[0067] The multi-device collaborative execution module includes a mud pump frequency converter, a vibrating screen frequency converter, a centrifuge dual frequency converter drive system, and a flow regulating valve driver. All drivers support the IEC 61850 communication protocol and have fault self-diagnosis and safe shutdown functions.

[0068] In addition, the system is equipped with a human-machine interaction and monitoring terminal, which is a 15-inch industrial touch screen that is connected to the control computer via gigabit Ethernet. The operating interface provides a parameter setting area (for inputting the target value of the export quality and the weight of the operating conditions), a real-time monitoring area (displaying the curves of each parameter and the energy consumption bar chart), and a manual intervention area (which can be switched to local control mode in case of emergency).

[0069] Finally, the principles of this application are summarized as follows: The core of this application is to solve the pain points of high energy consumption, poor adaptability, and unstable treatment effect of traditional underwater pile foundation mud circulation treatment through an intelligent closed loop of "full-dimensional real-time perception - dynamic correlation modeling - multi-objective optimization decision-making - multi-device closed-loop collaborative execution". Its principle is broken down into four core links: perception layer, modeling layer, optimization layer, and execution layer. Each link is progressive and interconnected. The following is a detailed explanation from the perspective of those skilled in the art, namely civil engineering and intelligent control: I. Perception Layer: Principle of Real-time Multi-dimensional State Perception This is the "sensory" basis of this application. The core is to comprehensively and synchronously collect key parameters of the entire mud treatment process through multiple types of industrial sensors and monitoring modules, and standardize the data into a computable feature matrix to solve the problems of traditional systems relying on experience for judgment and having one-sided data perception.

[0070] Parameter classification and collection: Inlet status parameters: Deploy a Coriolis mass flow meter (to measure flow rate and density), an online rotational viscometer (to measure viscosity), and an ultrasonic particle size analyzer (to measure sediment particle size distribution) in the main mud inlet pipeline to directly capture the original properties of the mud to be treated, providing a basis for the formulation of subsequent treatment strategies.

[0071] Operating status parameters: The electrical parameters (current, voltage, power) of the mud pump, vibrating screen, and centrifuge are collected through the power monitoring module. The speed / frequency of the equipment and the inlet pressure of the desander / desilter are obtained through the speed / frequency sensor and pressure transmitter, so as to monitor the equipment operating load and working conditions in real time.

[0072] Export quality parameters: Deploy the same type of sensor as the inlet at the final outlet of the mud to detect the density, viscosity, and solids content of the treated mud, and determine whether the treatment effect meets the standards.

[0073] Data preprocessing: Because different sensors have different sampling frequencies, timestamp alignment is achieved through hardware synchronization of the system's master clock to ensure a unified data time base; Maximum and minimum value normalization (physical quantities) and Z-score standardization (electrical parameters) are used to eliminate dimensional differences. Finally, a 14-dimensional multidimensional real-time state feature matrix (4-dimensional entry + 7-dimensional operation + 3-dimensional exit) is constructed according to the time series to provide standardized data input for subsequent modeling.

[0074] II. Modeling Layer: Dynamic Association Modeling Principle This is the "brain prediction center" of this application. The core is to establish a nonlinear mapping relationship between mud state, equipment operation and treatment effect / energy consumption through a recurrent neural network based on long short-term memory network (LSTM), so as to solve the problem of traditional systems "unable to predict the impact of parameter changes and lagging regulation".

[0075] Model selection criteria: The state of the mud in the underwater pile foundation (such as viscosity and particle size) and the equipment operating parameters are all time series data, and there are complex nonlinear relationships between the parameters (such as a sudden increase in mud viscosity will lead to an increase in centrifuge energy consumption and a decrease in treatment effect). The gating mechanism of LSTM network can effectively capture the long-term dependencies of time series and avoid the gradient vanishing problem of ordinary recurrent neural networks.

[0076] Model training and inference: Offline training: Using at least 3 months of historical working condition data (over 1 million records covering different geological and construction stages), the model is trained with mean squared error as the loss function. An early stopping mechanism is used to prevent overfitting, and the model parameters are finally solidified. Online inference: The model receives the real-time feature matrix through a 30-second sliding window, and outputs the predicted total power and output quality parameters of the system for the next 5 seconds, enabling early prediction of system behavior.

[0077] III. Optimization Layer: Weighted Multi-Objective Optimization and Optimization Decision-Making Principles This is the core of the decision-making process in this application. The core is to quantify the two major objectives of "energy saving" and "processing quality" through a weighted multi-objective optimization function, and combine it with an optimization algorithm to solve the globally optimal equipment control instructions, thereby solving the problem of "difficulty in balancing energy consumption and quality" in traditional systems.

[0078] Construction of weighted multi-objective optimization function: Define the optimization objective function: J = w1 × P_total + w2 × D_deviation, where: P_total is the total power consumption of the system (the sum of the power of the mud pump, vibrating screen, and centrifuge), representing the "energy saving" target; D_deviation is the deviation of mud purification indicators (the weighted sum of the relative deviations of outlet density, viscosity, and solid content from the target values), representing the "treatment quality" target; w1 and w2 are weighting coefficients that can be adaptively adjusted according to the construction conditions (e.g., in the excavation / hole cleaning stage, where the focus is on treatment quality, w2 is set to 0.7~0.9; in the pouring stage, where the focus is on energy saving, w1 is set to 0.6).

[0079] The supplementary role of the penalty function: When the export quality parameters exceed the preset tolerance range (such as density deviation exceeding 5%), the deviation is amplified by a piecewise nonlinear penalty function, and the optimization function is forced to prioritize the processing quality to avoid construction accidents (such as hole collapse or pipe blockage) caused by excessive energy saving.

[0080] Solution using optimization algorithms: The sequential quadratic programming algorithm is used to linearize the nonlinear dynamic correlation model at the current working point within each 5-second control cycle, transforming it into a quadratic programming subproblem to be solved, and finally generating a coordinated control command matrix (including the target speed of the mud pump, the target frequency of the vibrating screen, the target differential speed of the centrifuge, and the target opening degree of the regulating valve).

[0081] IV. Execution Layer: Multi-device Collaborative Execution and Closed-Loop Control Principles This is the "hand and foot execution end" of this application. Its core is to transform the optimization instructions into specific actions of the equipment, and to achieve closed-loop adaptive control through full-link collaborative frequency conversion regulation, thereby solving the problem of "independent operation of equipment and lack of coordination in regulation" in traditional systems.

[0082] Command execution: The multi-device collaborative execution module converts the control command matrix into control signals, which drive the mud pump, vibrating screen, and centrifuge (the centrifuge adopts a dual-frequency conversion common DC bus system for precise differential speed control) through the frequency converter driver, and adjust the opening of the flow regulating valve through the valve driver to achieve coordinated adjustment of the parameters of each device.

[0083] Closed-loop adaptive control: After the equipment executes the command, the sensing layer will collect new mud conditions and equipment operating parameters again, repeating the "sensing-modeling-optimization-execution" process to form a closed-loop feedback. For example, when the mud viscosity suddenly increases, the system will automatically reduce the mud pump speed and increase the centrifuge differential speed within 10 seconds, so that the outlet quality returns to the normal range without manual intervention.

[0084] During operation, the entire system can automatically respond to sudden changes in mud properties. For example, when the drill bit enters a high-clay layer, causing a sudden increase in inlet viscosity, the multi-dimensional real-time state sensing module immediately captures this change. The dynamic correlation model predicts that the outlet solids content will exceed the standard, and the optimization function then increases the w2 weight. The optimization algorithm outputs instructions to reduce the mud pump speed and increase the centrifuge differential speed. The system completes the adjustment within ten seconds, bringing the outlet parameters back to the normal range. This process requires no manual intervention, achieving true adaptive control.

[0085] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0086] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An energy-saving method for treating mud circulation in underwater pile foundations, characterized in that, include: Real-time data collection is performed on the inlet status parameters of the mud to be treated during the construction of underwater pile foundations, the operating status parameters of each key treatment device in the mud circulation treatment system, and the outlet quality parameters of the treated mud. Based on the collected parameters, a multi-dimensional real-time state feature matrix is ​​constructed through timestamp alignment and data normalization. Using the multidimensional real-time state feature matrix, a dynamic correlation model is established to characterize the nonlinear mapping relationship between inlet state parameters, operating state parameters and outlet quality parameters; A weighted multi-objective optimization function is established with the objectives of minimizing the total system power consumption and minimizing the deviation of the mud purification index. The total system power consumption is obtained by summing the electrical parameters in the operating status parameters of each key treatment device, and the deviation of the mud purification index is calculated by comparing the real-time value of the outlet quality parameter with the preset target value. The multidimensional real-time state feature matrix is ​​used as the input of the dynamic correlation model, and combined with the weighted multi-objective optimization function, the optimal collaborative control instruction matrix is ​​generated by iteratively solving through a preset optimization algorithm. The collaborative control instruction matrix includes the target speed of the mud pump, the target frequency of the vibrating screen motor, the target differential speed of the centrifuge, and the target opening degree of the flow regulating valve between each level of processing unit. The coordinated control command matrix is ​​transmitted to the corresponding equipment execution units within the mud circulation treatment system. The equipment execution units drive each key processing device to operate according to the parameters set by the coordinated control command matrix, forming a closed-loop adaptive control of the mud treatment process.

2. The energy-saving underwater pile foundation mud circulation treatment method according to claim 1, characterized in that, The inlet state parameters include inlet flow rate, inlet density, inlet viscosity, and inlet sediment particle size distribution. The specific methods for real-time collection of entry status parameters include: A flow meter is installed on the main inlet pipe of the mud circulation treatment system to simultaneously obtain the inlet flow rate and inlet density of the mud to be treated; A viscometer is installed on the main inlet pipe to obtain the inlet viscosity of the slurry to be treated; An ultrasonic particle size analyzer is installed on the main inlet pipe to obtain particle size distribution data of the mud and sand in the slurry to be treated.

3. The energy-saving underwater pile foundation mud circulation treatment method according to claim 1, characterized in that, The operating status parameters include the input current and speed of the mud pump, the input current and vibration frequency of the vibrating motor of the vibrating screen, the inlet pressure of the desander, the inlet pressure of the desliming device, and the differential speed and main motor input current of the centrifuge. Real-time collection of operational status parameters specifically includes: Power monitoring modules are installed on the power supply circuits of the vibrating motors of mud pumps, vibrating screens, and centrifuge main motors to collect input current and voltage in real time and calculate instantaneous power. The real-time speed of the mud pump, the real-time frequency of the vibrating screen motor, and the real-time differential speed of the centrifuge can be directly read through the encoder interface or speed sensor built into the frequency converter driver of each device. Install pressure transmitters on the hydrocyclone inlet pipes of the desander and deslimer to obtain their inlet pressure.

4. The energy-saving underwater pile foundation mud circulation treatment method according to claim 1, characterized in that, The dynamic correlation model is specifically a recurrent neural network model based on a long short-term memory network. This model learns by using a large amount of multidimensional real-time state feature matrix data collected in history through an offline training phase, thereby accurately capturing the dynamic impact of mud state changes on treatment effect and system energy consumption. The input layer of the recurrent neural network model of the Long Short-Term Memory network receives the multidimensional real-time state feature matrix, and the output layer outputs the predicted total system power consumption value and the output quality parameter value.

5. The energy-saving underwater pile foundation mud circulation treatment method according to claim 1, characterized in that, After establishing the weighted multi-objective optimization function, a deviation penalty function for mud purification indicators is also included: The mud purification index deviation penalty function compares the real-time density, viscosity and solids content of the treated mud with a preset target value range. When the detected value exceeds the target value range, a penalty term value is calculated by a preset piecewise nonlinear function based on the extent of the exceedance. The weight coefficients in the weighted multi-objective optimization function are adaptively adjusted according to the different working conditions of the underwater pile foundation construction, such as the excavation stage, the hole cleaning stage, or the pouring stage, so as to prioritize the protection of the core technical indicators of the specific stage.

6. The energy-saving method for treating mud circulation in underwater pile foundations according to claim 1, characterized in that, The optimization algorithm is used to iteratively solve the problem. Specifically, a sequential quadratic programming algorithm is adopted. In each control cycle, the quadratic programming algorithm performs a linear approximation of the dynamic correlation model based on the real-time state of the mud. Combined with the weighted multi-objective optimization function, the coordinated control instruction matrix that minimizes the objective function value calculated by the weighted multi-objective optimization function is obtained.

7. An energy-saving underwater pile foundation mud circulation treatment system, characterized in that, include: The multi-dimensional real-time state perception module is used to collect in real time the inlet state parameters of the mud to be treated, the operating state parameters of each key treatment equipment in the mud circulation treatment system, and the outlet quality parameters of the treated mud during the construction of underwater pile foundations. It also performs timestamp alignment and data normalization on all collected parameters to construct a multi-dimensional real-time state feature matrix. The dynamic correlation modeling and optimization decision-making module, which is connected to the multi-dimensional real-time state perception module, is used to establish a dynamic correlation model that represents the nonlinear mapping relationship between inlet state parameters, operating state parameters and outlet quality parameters, and to establish a weighted multi-objective optimization function with the objectives of minimizing the total power consumption of the system and minimizing the deviation of the mud purification index. Finally, based on the real-time input multi-dimensional real-time state feature matrix, the optimization function is solved by the optimization algorithm to generate a set of optimal collaborative control command matrices. The multi-device collaborative execution module receives the collaborative control instruction matrix from the dynamic correlation modeling and optimization decision-making module, converts it into specific control signals, and sends them to the equipment execution units such as mud pumps, vibrating screens, centrifuges, and flow regulating valves in the mud circulation treatment system to drive their collaborative operation.

8. The energy-saving underwater pile foundation mud circulation treatment system according to claim 7, characterized in that, The multi-dimensional real-time state perception module specifically includes: Coriolis mass flow meter, online rotary viscometer and ultrasonic particle size analyzer are installed on the main inlet pipe of the mud circulation treatment system. Power monitoring modules are installed on the power supply circuits of mud pumps, vibrating screen motors, and centrifuge main motors. Speed ​​and frequency sensors associated with the mud pump, vibrating screen, and centrifuge, respectively; And pressure transmitters installed on the inlet pipes of the desander and desilter respectively; All sensors and monitoring modules communicate with the dynamic correlation modeling and optimization decision-making module via an industrial fieldbus.

Images