Sludge clay pile foundation drilling parameter regulation and control method and system

Abstract

The invention discloses a mucky clay pile foundation drilling parameter regulation and control method and system, and belongs to the technical field of pile foundation drilling construction.The method comprises the steps that historical construction data of mucky clay pile foundation drilling is collected and preprocessed, and a data set is constructed; constructing and training a machine learning model based on the data set to obtain a drilling parameter time sequence prediction model; actual drilling parameters in the drilling process are collected in real time and compared with predicted drilling parameters output by the model, and drilling parameter errors are determined; and dynamically adjusting the drilling parameters of the drilling machine based on the drilling parameter error. According to the method, key parameters such as the drilling pressure, the rotating speed and the slurry concentration are intelligently regulated and controlled, so that the pile foundation construction efficiency and the hole forming quality in the complex soft soil stratum are improved, the construction risk is reduced, and the engineering safety is ensured.

Description

Technical Field

[0001] The invention belongs to the technical field of pile foundation drilling construction, and particularly relates to a method and system for controlling drilling parameters of a muddy clay pile foundation. Background Art

[0002] With the rapid development of urbanization, pile foundations, as an important form of deep foundation, are widely used in various engineering projects, including high-rise buildings, bridges, and ports. However, pile foundation construction in complex soft soils, such as silty clay strata, often faces a series of problems such as low drilling efficiency, poor hole quality, difficult sediment control, and hole wall collapse, due to the strata's high moisture content, low bearing capacity, poor permeability, and loose structure. These problems not only affect construction progress but also directly threaten the bearing capacity and safety of the pile foundation.

[0003] Existing pile foundation drilling operations primarily rely on the experience of construction personnel, adjusting parameters such as drilling pressure, rotational speed, and mud ratio to adapt to varying ground conditions. However, due to the complex and volatile distribution of the ground, and the limited and delayed availability of information from the construction site, drilling adjustments often lag behind ground changes, preventing precise, dynamic control. Furthermore, existing monitoring methods often focus on a single parameter, such as drilling pressure or mud flow rate, failing to fully utilize multi-source, multi-dimensional data to comprehensively perceive and model ground characteristics, resulting in limited control effectiveness.

[0004] In recent years, with the advancement of sensing and data analysis technologies, multimodal sensors can simultaneously acquire a variety of information during the drilling process, including mechanical, fluid, vibration, and electrical information. Machine learning has also demonstrated superior performance in pattern recognition, prediction, and optimal control. However, the research and application of the deep integration of multimodal sensing and machine learning in pile foundation drilling construction is still in its infancy, and a mature and efficient dynamic control system has yet to be established.

[0005] Therefore, there is an urgent need for a method that can fully utilize multimodal sensors to obtain multidimensional parameters in the drilling process in real time, use machine learning methods to accurately model and dynamically predict formation characteristics and drilling status, and realize intelligent control of key parameters such as drilling pressure, rotation speed, mud ratio and sediment thickness, so as to improve the efficiency and hole quality of pile foundation construction in complex soft soil strata, reduce construction risks and ensure project safety. Summary of the Invention

[0006] In response to the above-mentioned deficiencies in the prior art, the present invention provides a method and system for controlling the parameters of pile foundation drilling in muddy clay, which solves the problems of low utilization of collected data, delayed response of parameter control, and insufficient intelligent control level in the existing methods, which in turn lead to low efficiency and poor hole quality in pile foundation drilling construction.

[0007] In order to achieve the above-mentioned purposes, the technical scheme adopted by the present application is as follows: a method for adjusting drilling parameters of a silt clay pile foundation, comprising the following steps: S100, collecting and preprocessing historical construction data of silt clay pile foundation drilling, and constructing a data set; S200, constructing and training a machine learning model based on the data set to obtain a drilling parameter time series prediction model; S300, collecting actual drilling parameters in real time during the drilling process and comparing them with the predicted drilling parameters output by the model to determine the drilling parameter error; S400, dynamically adjusting the drilling parameters of the drilling rig based on the drilling parameter error.

[0008] Further, the step S100 comprises the following sub-steps: S101, collecting historical construction data of silt clay pile foundation drilling, including the drilling pressure, drilling speed and mud bentonite concentration of each construction; S102, normalizing and standardizing the collected data; S103, performing feature fusion on the normalized and standardized data to obtain a principal component data set; S104, performing feature selection and transformation on the data in the principal component data set to construct a data set.

[0009] Further, the step S200 comprises: The machine learning model is an LSTM model; Each LSTM model learns the long-short dependency relationship of drilling pressure, drilling speed and mud bentonite concentration in the time series in the data set, and the time series features learned by each LSTM model are fused to obtain a drilling parameter time series prediction model.

[0010] Further, in the step S300, the drilling parameter error includes drilling speed error, drilling pressure error and mud bentonite concentration error; The drilling speed error is: In the formula, represents the drilling speed predicted by the drilling parameter time series prediction model, represents the actual drilling speed collected, represents the number of data collection groups; The drilling pressure error is: In the formula, represents the drilling pressure predicted by the model, represents the actual drilling pressure collected, Indicates the number of data collection groups; The mud bentonite concentration error for: Where, represents the bentonite concentration in the mud predicted by the model, It represents the actual collected mud bentonite concentration, and N represents the number of data collection groups.

[0011] Furthermore, in step S400, the dynamic adjustment includes short-term adjustment and mid-term adjustment; The short-term adjustment is to compare the drilling pressure and drilling speed at the next moment predicted by the drilling parameter time series prediction model with the current drilling pressure and drilling speed actually collected, and then adjust the drilling pressure and drilling speed; The mid-term adjustment is to construct a reward function of the reinforcement learning model according to the drilling parameter error, and then adjust the mud bentonite concentration based on the reinforcement learning model.

[0012] Furthermore, the formula for adjusting the bit pressure is: Where, Indicates the target adjustment drilling pressure at the next moment. Indicates the actual current drilling pressure, Indicates the predicted output of the next moment’s WOB. Indicates the allowable WOB prediction error threshold; The formula for drilling speed adjustment is: Where, Indicates the target drilling speed adjustment at the next moment. Indicates the drilling speed adjustment value. Indicates the actual current drilling speed, Indicates the drilling speed at the next moment of the predicted output, Indicates the allowable drilling rate prediction error threshold.

[0013] Furthermore, the reward function is: Where, 、 and They represent the prediction error of bit weight, drilling speed and mud bentonite concentration respectively. Respectively 、 and The weight coefficient of The method for adjusting the concentration of the slurry bentonite is as follows: at each moment t, according to the action output by the reinforcement learning model based on the reward function Adjust the bentonite concentration, the adjustment formula is: = Where, Indicates the current error in the slurry bentonite concentration, Indicates the target bentonite concentration to be adjusted at the next moment. Indicates the bentonite concentration at the next moment of the predicted output, Indicates the actual current collected mud bentonite concentration, Indicates the allowed bentonite concentration prediction error threshold.

[0014] A muddy clay pile foundation drilling parameter control system, comprising: Model building module: used to build a time series prediction model for drilling parameters based on historical construction data of muddy clay pile foundation drilling; the drilling parameters include bit pressure, drilling speed and mud bentonite concentration; Data acquisition module: used to collect actual drilling parameters; Data comparison module: used to compare the actual drilling parameters with the predicted drilling parameters output by the model to determine the drilling parameter error; Parameter adjustment module: used to dynamically adjust the drilling parameters of the drilling rig according to the determined drilling parameter errors.

[0015] Furthermore, the model building module includes a data processing unit and a model building unit; The data processing unit is used to perform standardization / normalization processing, feature fusion processing, feature selection and transformation processing on the historical construction data in sequence to construct a data set; The model building unit is used to learn the long-short dependency relationship of drilling pressure, drilling speed and mud bentonite concentration in the time series of the dataset through the LSTM model, and to fuse the time series features learned by each LSTM model to obtain a drilling parameter time series prediction model.

[0016] Furthermore, the parameter adjustment module includes a short-term adjustment unit and a medium-term adjustment unit; The short-term adjustment unit is used to compare the drilling pressure and drilling speed at the next moment predicted and output by the drilling parameter time series prediction model with the current drilling pressure and drilling speed actually collected, and then adjust the drilling pressure and drilling speed at the next moment; The mid-term adjustment unit is used to construct a reward function of the reinforcement learning model according to the drilling parameter error, and adjust the mud bentonite concentration based on the reinforcement learning model.

[0017] The beneficial effects of the present invention are: (1) Realize intelligent dynamic control of pile foundation drilling parameters: By integrating multimodal sensors, real-time acquisition of multi-dimensional key parameters such as drilling pressure and drilling speed during the drilling process, construct a high-dimensional "configuration-time-space feature matrix", and combine it with a machine learning model to realize the prediction and closed-loop adjustment of the drilling rig's drilling pressure, speed, and mud bentonite concentration, thereby significantly improving drilling efficiency and hole quality.

[0018] (2) Enhance the adaptability to complex soft soil strata: Train the deep learning model through historical construction data and multi-physics field simulation data, use the graph convolutional network to capture the relationship between stratum depths, combine the LSTM model and incremental learning and other technologies to improve the model's ability to understand the stratum structure and real-time adaptability, and overcome the construction challenges brought about by complex stratum changes.

[0019] (3) Constructing a closed-loop control and continuous optimization mechanism: The present invention realizes closed-loop control integrating prediction, feedback, and fine-tuning during the construction process, combines short-term (seconds) and medium-term (minutes) dynamic adjustment mechanisms, and uses reinforcement learning and incremental learning to continuously optimize the control strategy, thereby achieving adaptive evolution of the model and long-term stable improvement of construction quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of the method for controlling drilling parameters of muddy clay pile foundations provided by the present invention. DETAILED DESCRIPTION

[0021] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0022] Example 1: The embodiment of the present invention provides a method for controlling drilling parameters of muddy clay pile foundation; See also Figure 1 , the method comprises the following steps: S100, collect and pre-process historical construction data of muddy clay pile foundation drilling to construct a data set; S200, constructing and training a machine learning model based on the data set to obtain a drilling parameter time series prediction model; S300, collect the actual drilling parameters of the drilling process in real time, and compare the actual drilling parameters with the predicted drilling parameters output by the model to determine drilling parameter errors; S400, dynamically adjust the drilling parameters of the drilling rig based on the drilling parameter errors.

[0023] The step S100 of the embodiment of the application comprises the following sub-steps: S101, collect historical construction data of the mucky clay pile foundation drilling, including the drilling pressure, drilling speed and bentonite concentration of the mud of each construction; Specifically, more than 100,000 sets of historical construction data are collected, covering mucky clay (N value 10-25), including drilling pressure (P), drilling speed (V) and bentonite concentration (Q) in the mud and other data; S102, normalize and standardize the collected data; Specifically, the data with large numerical variation range of drilling pressure and drilling speed are subjected to Z-score standardization processing; the data with small variation range of bentonite concentration in the mud is subjected to Min-Max normalization processing; wherein, the formula of Z-score standardization is: In the formula, X represents the original data, X represents the mean value of the data, X represents the standard deviation of the data, X represents the standardization result; The formula of Min-Max normalization is: In the formula, X represents the original data, X represents the maximum value of the original data, X represents the minimum value of the original data, X represents the normalization result; S103, perform feature fusion on the normalized and standardized data to obtain a principal component data set; Specifically, through principal component analysis (PCA), the covariance matrix of the data matrix of drilling pressure, drilling speed and bentonite concentration in the mud is calculated, and then eigenvalue decomposition is performed to obtain eigenvectors and eigenvalues, and the first k principal components with the largest eigenvalues are selected to retain most of the variances, and the original data is projected onto these principal components to obtain the reduced data, realizing the conversion of multi-dimensional features into a principal component set, reducing redundant information while retaining main features. Wherein, the covariance matrix of the data matrix is: In the formula, X represents the covariance matrix, represents the total number of samples, Represents the feature vector of the i-th sample; S104, performing feature selection and transformation on the data in the principal component data set to construct a data set; Specifically, the Pearson correlation coefficient is used to select features with higher correlation as model inputs and remove redundant or noisy data; the Pearson correlation coefficient is: if A value close to 1 or -1 indicates that there is a strong linear relationship between the two features.

[0024] Step S200 of the embodiment of the present invention includes: Build a machine learning model as an LSTM model; The long-short dependence relationship of drilling pressure, drilling speed and mud bentonite concentration in the time series of the dataset is learned through a LSTM model respectively, and the time series features learned by each LSTM model are integrated to obtain the drilling parameter time series prediction model.

[0025] Specifically, in this embodiment, the process of the LSTM model learning the time series pattern of bit weight, drilling speed, and mud bentonite concentration during the drilling process is as follows: Input Gate: Forget Gate: Candidate memory cells: Memory Update: Output gate: Final output: in, is the input gate control signal, is the input of parameters such as bit pressure and mud flow rate at the current time step, is the hidden state at the previous moment, is the forget gate control signal, The range of the Sigmoid activation function is 0~1. is the current candidate memory (via tanh mapping), is the memory unit state at the current moment, is the output gate control signal, is the weight matrix input to the gate unit, is the weight matrix from the previous hidden state to the gate unit, is the bias term of the gate control unit, is the hyperbolic tangent function with an output range of -1~.

[0026] In this embodiment, the time series features learned by each LSTM model are integrated to obtain the prediction output of the model: Where, represents the fused features, represents the weight matrix, represents the bias term, Represents the predicted output of the model.

[0027] In step S300 of the embodiment of the present invention, during the actual drilling process, the drilling pressure and drilling speed collected by the drilling rig and the bentonite concentration data in the mud collected from the mud pump outlet are compared with the prediction results of the machine learning model to calculate the drilling parameter errors, including the drilling parameter errors including the drilling speed error, the drilling pressure error and the mud bentonite concentration error; Among them, the drilling speed error for: Where, represents the drilling rate predicted by the drilling parameter time series prediction model, Indicates the actual drilling speed, Indicates the number of data collection groups; WOB error for: Where, represents the weight on bit predicted by the model, Indicates the actual acquired drilling pressure, Indicates the number of data collection groups; Mud bentonite concentration error for: Where, represents the bentonite concentration in the mud predicted by the model, It represents the actual collected mud bentonite concentration, and N represents the number of data collection groups.

[0028] In step S400 of the embodiment of the present invention, the parameters and models predicted by machine learning are compared with the actual collected data, and the drilling parameters of the drilling rig are dynamically adjusted to achieve real-time adjustment and closed-loop control.

[0029] The dynamic adjustment in this embodiment includes short-term adjustment and medium-term adjustment; In the formula, the short-term adjustment is that the predicted output of the next moment of the WOB and the ROP is compared with the actually collected current WOB and ROP, and then the WOB and the ROP are adjusted; the medium-term adjustment is that a reward function of a reinforcement learning model is constructed according to the drilling parameter error, and then the bentonite concentration of the mud is adjusted based on the reinforcement learning model.

[0030] In the formula, the WOB adjustment formula is: In the formula, the WOB adjustment formula is: represents the target adjustment WOB of the next moment, represents the actually collected current WOB, represents the predicted output of the next moment of the WOB, represents the allowed WOB prediction error threshold; The ROP adjustment formula is: In the formula, the ROP adjustment formula is: represents the target adjustment ROP of the next moment, represents the ROP adjustment value, represents the actually collected current ROP, represents the predicted output of the next moment of the ROP, represents the allowed ROP prediction error threshold.

[0031] In the embodiment, in the medium-term adjustment process, the bentonite concentration of the mud is adjusted, the adjustment range is ±0.5%, the reinforcement learning (RL) method is used to dynamically adjust the bentonite concentration of the mud, and a reward function is designed by comprehensively considering the WOB error, the ROP error and the bentonite concentration error in the mud, so as to minimize the loss function and ensure the stability of the drilling process; wherein the reward function is: In the formula, the reward function is: , and respectively represent the WOB prediction error, the ROP prediction error and the bentonite concentration prediction error in the mud, respectively represent the weight coefficients of , and , according to the importance of the WOB, the ROP and the bentonite ratio in the mud in the overall optimization target, α is 1, β is 2, and γ is 0.5.

[0032] In the embodiment, the method for adjusting the bentonite concentration of the mud is that at each moment t, the action a t is output based on the reward function of the reinforcement learning model Adjust the bentonite concentration, the adjustment formula is: = Where, Indicates the current error in the slurry bentonite concentration, Indicates the target bentonite concentration to be adjusted at the next moment. Indicates the bentonite concentration at the next moment of the predicted output, Indicates the actual current collected mud bentonite concentration, Indicates the allowed bentonite concentration prediction error threshold.

[0033] In an embodiment of the present invention, based on the above-mentioned adjustment method, a feedback channel is further provided to compare the predicted and measured drilling parameters to realize error feedback correction and model fine-tuning. The sediment thickness is fed back in real time by the ultrasonic boremeter / lidar, and the hole wall deformation is determined by the geological radar system. When the parameter error is greater than the threshold (such as the predicted deviation of the bit pressure is greater than 10%), the model fine-tuning is automatically triggered. In addition, an incremental learning mechanism is adopted to continuously optimize the model performance during the drilling process and enhance the adaptability to complex formations.

[0034] Example 2: This embodiment is a further limitation made on the basis of Example 1, and its purpose is to provide a silty clay pile foundation drilling parameter control system, which is implemented based on the silty clay pile foundation drilling parameter control method in Example 1. For other parts not mentioned, refer to Example 1 or the prior art.

[0035] A muddy clay pile foundation drilling parameter control system in this embodiment includes: Model building module: used to build a time series prediction model for drilling parameters based on historical construction data of muddy clay pile foundation drilling; drilling parameters include bit pressure, drilling speed, and mud bentonite concentration; Data acquisition module: used to collect actual drilling parameters; Data comparison module: used to compare the actual drilling parameters with the predicted drilling parameters output by the model to determine the drilling parameter error; Parameter adjustment module: used to dynamically adjust the drilling parameters of the drilling rig according to the determined drilling parameter errors.

[0036] In an embodiment of the present invention, the model building module includes a data processing unit and a model building unit; The data processing unit is used to perform standardization / normalization processing, feature fusion processing, feature selection and transformation processing on the historical construction data in sequence to construct a data set; The model construction unit is configured to learn long-short dependency relationships of the WOB, the ROP and the mud bentonite concentration in the time sequence in the data set by using the LSTM model, and fuse the time sequence features learned by the LSTM model to obtain a drilling parameter time sequence prediction model.

[0037] In the embodiment of the present application, the parameter adjustment module comprises a short-term adjustment unit and a medium-term adjustment unit. The short-term adjustment unit is configured to compare the WOB and the ROP at the next moment predicted by the drilling parameter time sequence prediction model with the actually collected current WOB and ROP, and then adjust the WOB and the ROP at the next moment. The medium-term adjustment unit is configured to construct a reward function of a reinforcement learning model according to the drilling parameter error, and adjust the mud bentonite concentration based on the reinforcement learning model.

[0038] Specifically, in the short-term adjustment unit, the formula for adjusting the WOB is as follows: In the formula, Wtarget represents the target adjusted WOB at the next moment, Wtarget represents the target adjusted WOB at the next moment, Wcurrent represents the actually collected current WOB, Wnext represents the WOB at the next moment predicted by the drilling parameter time sequence prediction model, Werror represents a permissible WOB prediction error threshold value. The formula for adjusting the ROP is as follows: In the formula, Vtarget represents the target adjusted ROP at the next moment, Vtarget represents the target adjusted ROP at the next moment, Vtarget represents the target adjusted ROP at the next moment, Vcurrent represents the actually collected current ROP, Vnext represents the ROP at the next moment predicted by the drilling parameter time sequence prediction model, Verror represents a permissible ROP prediction error threshold value.

[0039] In the embodiment, in the medium-term adjustment process, the mud bentonite concentration is adjusted, the adjustment range is ±0.5%, the reinforcement learning (RL) method is used to dynamically adjust the concentration of the mud bentonite, and a reward function is designed by comprehensively considering the WOB error, the ROP error and the mud bentonite concentration error to minimize the loss function and ensure the stability of the drilling process; wherein the reward function is as follows: In the formula, Werror, Verror and Cerror represent the WOB prediction error, the ROP prediction error and the mud bentonite concentration prediction error, respectively, , and ​Respectively 、 and The weight coefficients are: α is 1, β is 2, and γ is 0.5, according to the importance of drilling pressure, drilling speed and bentonite ratio in the overall optimization target.

[0040] In this embodiment, the method for adjusting the concentration of bentonite in the slurry is as follows: at each time t, the action outputted by the reinforcement learning model based on the reward function Adjust the bentonite concentration, the adjustment formula is: .

[0041] = Where, Indicates the current error in the slurry bentonite concentration, Indicates the target bentonite concentration to be adjusted at the next moment. Indicates the bentonite concentration at the next moment of the predicted output, Indicates the actual current collected mud bentonite concentration, Indicates the allowed bentonite concentration prediction error threshold.

[0042] In an embodiment of the present invention, based on the above-mentioned adjustment method, a feedback channel is further provided to compare the predicted and measured drilling parameters to realize error feedback correction and model fine-tuning. The sediment thickness is fed back in real time by the ultrasonic boremeter / lidar, and the hole wall deformation is determined by the geological radar system. When the parameter error is greater than the threshold (such as the predicted deviation of the bit pressure is greater than 10%), the model fine-tuning is automatically triggered. In addition, an incremental learning mechanism is adopted to continuously optimize the model performance during the drilling process and enhance the adaptability to complex formations.

[0043] Specific embodiments are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas. At the same time, for those skilled in the art, according to the ideas of the present invention, there may be changes in the specific implementation methods and application scopes. In summary, the contents of this specification should not be understood as limiting the present invention.

[0044] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A method for controlling drilling parameters of muddy clay pile foundation, characterized in that: The following steps are involved: S100, collect and pre-process historical construction data of muddy clay pile foundation drilling to construct a data set; S200, constructing and training a machine learning model based on the data set to obtain a drilling parameter time series prediction model; S300, collecting actual drilling parameters during the drilling process in real time, and comparing them with the predicted drilling parameters output by the model to determine the drilling parameter error; S400: Dynamically adjust the drilling parameters of the drilling rig based on the drilling parameter errors.

2. The method for controlling drilling parameters of muddy clay pile foundation according to claim 1, characterized in that: The step S100 includes the following sub-steps: S101. Collect historical construction data of muddy clay pile foundation drilling, including drilling pressure, drilling speed, and mud bentonite concentration for each construction; S102, normalizing and standardizing the collected data; S103, performing feature fusion on the normalized and standardized data to obtain a principal component data set; S104: Perform feature selection and transformation on the data in the principal component data set to construct a data set.

3. The method for controlling drilling parameters of muddy clay pile foundation according to claim 1, characterized in that: The step S200 includes: Build a machine learning model as an LSTM model; The long-short dependence relationship of drilling pressure, drilling speed and mud bentonite concentration in the time series of the dataset is learned through a LSTM model respectively, and the time series features learned by each LSTM model are integrated to obtain the drilling parameter time series prediction model.

4. The method for controlling drilling parameters of muddy clay pile foundation according to claim 1, characterized in that: In step S300, the drilling parameter errors include drilling speed error, bit pressure error and mud bentonite concentration error; The drilling speed error for: Where, represents the drilling rate predicted by the drilling parameter time series prediction model, Indicates the actual drilling speed, Indicates the number of data collection groups; The weight-on-bit error for: Where, represents the weight on bit predicted by the model, Indicates the actual acquired bit weight, Indicates the number of data collection groups; The mud bentonite concentration error for: Where, represents the bentonite concentration in the mud predicted by the model, It represents the actual collected mud bentonite concentration, and N represents the number of data collection groups.

5. The method for controlling drilling parameters of muddy clay pile foundation according to claim 1, characterized in that: In step S400, the dynamic adjustment includes short-term adjustment and mid-term adjustment; The short-term adjustment is to compare the drilling pressure and drilling speed at the next moment predicted by the drilling parameter time series prediction model with the current drilling pressure and drilling speed actually collected, and then adjust the drilling pressure and drilling speed; The mid-term adjustment is to construct a reward function of the reinforcement learning model according to the drilling parameter error, and then adjust the mud bentonite concentration based on the reinforcement learning model.

6. The method for controlling drilling parameters of muddy clay pile foundation according to claim 5, characterized in that: The formula for adjusting the drilling pressure is: Where, Indicates the target adjustment weight on bit at the next moment. Indicates the actual current drilling pressure, Indicates the predicted output of the next moment’s WOB. Indicates the allowable bit weight prediction error threshold; The formula for drilling speed adjustment is: Where, Indicates the target drilling speed adjustment at the next moment. Indicates the drilling speed adjustment value. Indicates the actual current drilling speed, Indicates the drilling speed at the next moment of the predicted output, Indicates the allowable drilling rate prediction error threshold.

7. The method for controlling drilling parameters of muddy clay pile foundation according to claim 5, characterized in that: The reward function is: Where, 、 and They represent the prediction error of bit weight, drilling speed and mud bentonite concentration respectively. Respectively 、 and The weight coefficient of The method for adjusting the concentration of the slurry bentonite is as follows: at each moment t, according to the action output by the reinforcement learning model based on the reward function Adjust the bentonite concentration, the adjustment formula is: = Where, Indicates the current error in the slurry bentonite concentration, Indicates the target bentonite concentration to be adjusted at the next moment. Indicates the bentonite concentration at the next moment of the predicted output, Indicates the actual current collected mud bentonite concentration, Indicates the allowed bentonite concentration prediction error threshold.

8. A muddy clay pile foundation drilling parameter control system, implemented based on the muddy clay pile foundation drilling parameter control method according to any one of claims 1 to 7, characterized in that: include: Model building module: used to build a drilling parameter time series prediction model based on historical construction data of muddy clay pile foundation drilling; The drilling parameters include weight on bit, drilling speed and mud bentonite concentration; Data acquisition module: used to collect actual drilling parameters; Data comparison module: used to compare the actual drilling parameters with the predicted drilling parameters output by the model to determine the drilling parameter error; Parameter adjustment module: used to dynamically adjust the drilling parameters of the drilling rig according to the determined drilling parameter errors.

9. The muddy clay pile foundation drilling parameter control system according to claim 8, characterized in that: The model building module includes a data processing unit and a model building unit; The data processing unit is used to perform standardization / normalization processing, feature fusion processing, feature selection and transformation processing on the historical construction data in sequence to construct a data set; The model building unit is used to learn the long-short dependency relationship of drilling pressure, drilling speed and mud bentonite concentration in the time series of the dataset through the LSTM model, and to fuse the time series features learned by each LSTM model to obtain a drilling parameter time series prediction model.

10. The muddy clay pile foundation drilling parameter control system according to claim 8, characterized in that: The parameter adjustment module includes a short-term adjustment unit and a medium-term adjustment unit; The short-term adjustment unit is used to compare the drilling pressure and drilling speed at the next moment predicted and output by the drilling parameter time series prediction model with the current drilling pressure and drilling speed actually collected, and then adjust the drilling pressure and drilling speed at the next moment; The mid-term adjustment unit is used to construct a reward function of the reinforcement learning model according to the drilling parameter error, and adjust the mud bentonite concentration based on the reinforcement learning model.

Images