Drilling-planting integrated machine tunneling resistance intelligent collaborative regulation method based on multi-source perception

Abstract

The present application relates to a kind of drilling and planting integrated machine excavation resistance intelligent collaborative regulation method based on multi-source perception, specifically relates to the field of industrial automation and intelligent control, and self-adaptive digital twin model is constructed by multi-source perception data fusion, and based on the system state vector, candidate control instruction sequence is carried out parallel foresight simulation, and by rolling solving multi-objective optimization function, the intelligent advance prediction and self-adaptive pre-regulation of excavation resistance are realized, which effectively inhibits resistance mutation and fluctuation, significantly improves drilling stability and pile quality, while optimizing control reduces energy consumption, and the online parameter updating mechanism based on model prediction error endows the model with continuous self-evolution ability, ensuring the adaptability and reliability of control system in whole life cycle.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and intelligent control, and more specifically, to a method for intelligent collaborative control of tunneling resistance in a drilling and planting machine based on multi-source sensing. Background Technology

[0002] As a key piece of equipment in modern foundation treatment and pile foundation construction, the drilling and pile-forming machine directly affects the efficiency and quality of the project in complex strata. In actual construction, geological conditions are often highly uncertain, frequently encountering complex situations such as alternating layers of soft and hard soil, local boulders or concrete obstacles, and quicksand layers rich in groundwater. At the same time, the equipment's own condition, such as the progressive wear of the drill bit and the real-time posture of the multi-joint actuator, is also dynamically changing. Under such complex working conditions where internal and external factors intertwine, the resistance encountered during the tunneling process exhibits strong nonlinear and abrupt characteristics. When the drill bit enters a hard rock layer from a soft soil layer or encounters an unknown obstacle, the tunneling resistance will rise sharply in a very short time. Traditional construction methods mainly rely on the operator's experience to adjust parameters, which often cannot react in time to such instantaneous changes, resulting in abnormal vibration of the equipment, deviation of the drilling trajectory, or even drill rod jamming or structural overload damage, seriously threatening construction safety, pile quality, and equipment life.

[0003] Currently, control technologies for tunneling resistance have two main limitations. First, they rely on preliminary geological survey reports for macroscopic pre-setting of construction parameters. However, these reports only provide general stratigraphic trends and cannot accurately detect and predict microscopic, localized lithological abrupt changes along the borehole path, such as small boulders or hard interlayers. Second, they commonly employ control strategies based on real-time sensor feedback. This involves monitoring parameters such as drilling torque, feed pressure, and rotational speed, and adjusting accordingly after abnormal fluctuations. This is essentially a "post-event response" control method, which is ineffective at the moment the drill bit contacts the interface of a sudden resistance change. The system has been impacted, and the control actions inevitably have a lag. Even if the system can respond quickly, the impact load caused by the sudden change has already adversely affected the equipment structure and borehole stability. Therefore, the existing technology lacks an intelligent pre-control method that can predict the short-range strata characteristics and impending resistance changes in advance based on online multi-dimensional information before the resistance change occurs, and drive the execution system to adjust the working parameters in advance to achieve a smooth transition. This capability gap is the core technical bottleneck that restricts the drilling and planting machine from achieving efficient, stable, and high-quality automated construction in highly complex strata. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing an intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing, thereby solving the problems mentioned in the background art.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: specifically, it includes the following steps: Step S1: Acquire multi-source sensing data of the drilling and planting machine in real time. The multi-source sensing data includes equipment operating parameters, machine response signals, and prior geological and construction data. Based on the machine response signals and prior geological and construction data, generate an adaptive feature vector that characterizes the comprehensive properties of the contact interface between the drill bit and the formation. Based on the multi-source sensing data and the adaptive feature vector, construct and maintain a digital twin model reflecting the interaction between the drilling tool and the formation online, and define the vector composed of equipment operating parameters as the system state vector. Step S2: In each control cycle, obtain the current system state vector of the drilling and planting machine, and use the system state vector as the initial state of the digital twin model; in the digital twin model, for multiple different candidate control command sequences, simulate the response of the drilling and planting machine in a limited prediction time domain in the future, and obtain the predicted system state sequence and the predicted tunneling resistance sequence corresponding to each candidate control command sequence, wherein each predicted tunneling resistance sequence contains multiple predicted resistance values ​​arranged in the order of future sampling times; Step S3: Construct a multi-objective optimization function that includes the objectives of tunneling resistance stability, energy consumption, and control action smoothness. Using the predicted system state sequence and predicted tunneling resistance sequence obtained in Step S2 as inputs, solve the multi-objective optimization function. Select the sequence that makes the function value of the multi-objective optimization function optimal from multiple candidate control command sequences as the optimal control command sequence. Record the predicted tunneling resistance sequence corresponding to the optimal control command sequence and define it as the optimal predicted resistance sequence. Send the first control command in the optimal control command sequence as the actual control command for the current control cycle to the bottom-level execution mechanism of the drilling and planting machine. Step S4: Collect the actual tunneling resistance of the drilling and planting machine after the actual control command is executed; calculate the model prediction error between the actual tunneling resistance and the first predicted resistance value in the optimal predicted resistance sequence; based on the model prediction error, use the recursive least squares algorithm to adjust and update a set of time-varying parameters in the digital twin model that characterize the dynamic interaction between the drill string and the formation online. In a preferred embodiment, the process of defining the system state vector in step S1 is specifically as follows: The equipment operating parameters are extracted from the real-time multi-source sensing data, including the pressure of the propulsion hydraulic cylinder, the torque of the rotary motor, the speed of the rotary motor, and the feed speed. The hydraulic cylinder pressure, rotary motor torque, rotary motor speed, and feed speed are arranged in a preset order to form a system state vector.

[0006] In a preferred embodiment, the specific process of generating the adaptive feature vector is as follows: The multi-source sensing data also includes body response signals and prior geological and construction data. The body response signals include the raw time-domain vibration signals collected by vibration sensors installed on the power head and drill pipe, and the raw time-domain acoustic emission signals collected by acoustic emission sensors installed on the power head and drill pipe. The prior geological and construction data include the distribution of standard penetration blows along the depth obtained from the exploration and the historical resistance and depth curves of adjacent pile construction records. The original time-domain vibration signal acquired by the vibration sensor and the original time-domain acoustic emission signal acquired by the acoustic emission sensor are processed synchronously. For the original time-domain vibration signal, the wavelet packet energy spectrum entropy within a set time window is calculated to obtain the vibration characteristic value. For the original time-domain acoustic emission signal, the average value of the ratio of rise time to amplitude of all signal events within the same time window is calculated to obtain the acoustic emission characteristic value. Extract the standard penetration test blows corresponding to the current drill bit depth from the distribution of standard penetration test blows along the depth obtained from the exploration and the historical resistance and depth curves of adjacent pile construction records. Finally, the calculated vibration feature values, acoustic emission feature values, and extracted standard penetration blow counts are fused. This fusion operation is performed by a dynamic feature extraction and fusion network. This network uses a sliding time window to extract a feature sequence composed of wavelet packet energy spectrum entropy from the continuous original vibration time-domain signal and a feature sequence composed of the mean of the rise time to amplitude ratio from the continuous original acoustic emission time-domain signal. The real-time acquired vibration feature values, acoustic emission feature values, and standard penetration blow counts corresponding to the current depth are combined as inputs and fed into a nonlinear fusion module. A nonlinear activation function is then applied to output an adaptive feature vector.

[0007] In a preferred embodiment, the process of constructing and maintaining a digital twin model reflecting the interaction between the drill string and the formation online based on multi-source sensing data and adaptive feature vectors specifically involves: A parameterized nonlinear state-space model is constructed as a digital twin model. In this digital twin model, the system state vector at the next moment is obtained by adding the current system state vector through a dynamic transmission link, the control input vector through a control input link, and the contribution of the formation nonlinear reaction force determined by the current adaptive eigenvector. The characteristics of the dynamic transmission link, the parameters involved in the calculation of the contribution of the nonlinear reaction force of the stratum, and the parameters involved in the calculation of the predicted tunneling resistance together constitute a set of time-varying parameters. The digital twin model uses a recursive least squares algorithm to iteratively update a set of time-varying parameters online based on the model prediction error between the predicted tunneling resistance and the actual tunneling resistance.

[0008] In a preferred embodiment, step S2, which involves generating multiple different candidate control instruction sequences, specifically comprises: For a finite future prediction time domain, for multiple consecutive future sampling moments contained within the prediction time domain, control commands including drill bit thrust setting commands and drill bit rotation speed setting commands are configured for each future sampling moment; By combining various control parameter configuration rules, multiple candidate control command sequences with differences in control commands are generated. These various control parameter configuration rules include rules that keep the thrust setting command constant while changing the speed setting command in the prediction time domain, rules that keep the speed setting value constant while changing the thrust setting command, and rules that make the thrust setting command and speed setting command change in synergy according to a preset relationship.

[0009] In a preferred embodiment, the process of performing parallel simulation of multiple different candidate control command sequences in the digital twin model specifically includes: Within the same control cycle, multiple parallel computation processes are initiated in the computational environment of the digital twin model. Each parallel computation process constitutes an independent simulation task. The number of independent simulation tasks is equal to the number of multiple different candidate control command sequences. Each independent simulation task takes the currently acquired system state vector as a common initial state and loads one of the multiple different candidate control command sequences as input. Subsequently, each independent simulation task recursively calculates the predicted system state vector of the drilling and planting machine at the end of each future sampling time in the future prediction time domain based on the digital twin model. Each recursive calculation depends on three inputs: the first is the predicted system state vector at the end of the previous future sampling time; the second is the control command corresponding to the current future sampling time from the candidate control command sequence; and the third is an adaptive feature vector corresponding to the current future sampling time, which is obtained by calculating the drilling depth based on the feed rate contained in the predicted system state vector and combining it with the prior geological and construction data defined in step S1. Through this recursive simulation that integrates forward-looking stratigraphic information, each independent simulation task ultimately outputs a predicted system state sequence corresponding to the loaded candidate control command sequence. The predicted system state sequence consists of multiple predicted system state vectors arranged in the order of future sampling times. Simultaneously, a predicted tunneling resistance sequence is calculated and output at each future sampling time within the same prediction time domain. The predicted tunneling resistance sequence consists of multiple predicted resistance values ​​arranged in the order of future sampling times.

[0010] In a preferred embodiment, step S3, specifically the process of constructing a multi-objective optimization function that includes the tunneling resistance stability objective, energy consumption objective, and control action smoothness objective, is as follows: The multi-objective optimization function is used to calculate a comprehensive performance evaluation value for each candidate control command sequence. This comprehensive performance evaluation value is obtained by weighted summation of three sub-evaluation values ​​at each future sampling time in the prediction time domain. The three sub-evaluation values ​​correspond to the tunneling resistance stability target, energy consumption target, and control action smoothness target, respectively. When summing the sub-item evaluation values ​​at each future sampling time, a future discount factor between zero and one is introduced; at the same time, a variable dynamic weight coefficient is configured for the sub-item evaluation value of the tunneling resistance stability target, and the value of the dynamic weight coefficient is adaptively adjusted according to the magnitude of the predicted resistance value at the current future sampling time.

[0011] In a preferred embodiment, the process of solving the multi-objective optimization function and selecting the optimal control command sequence from multiple candidate control command sequences specifically involves: Using the predicted system state sequence and predicted tunneling resistance sequence corresponding to each candidate control command sequence obtained in step S2 as input data for calculating the multi-objective optimization function, the comprehensive performance evaluation value corresponding to each candidate control command sequence is calculated in parallel and independently. After completing the quantitative evaluation of all candidate control command sequences, a minimization comparison process is used to select the comprehensive performance evaluation value with the smallest value from all the calculated comprehensive performance evaluation values. The candidate control command sequence corresponding to the smallest comprehensive performance evaluation value is determined as the one that makes the function value of the multi-objective optimization function optimal, and it is formally determined as the optimal control command sequence. Meanwhile, from the set of predicted tunneling resistance sequences output in step S2 that correspond to each candidate control command sequence, the specific predicted tunneling resistance sequence associated with the optimal control command sequence is extracted, and this predicted tunneling resistance sequence is recorded separately and defined as the optimal predicted resistance sequence. Finally, the first control command is extracted from the determined optimal control command sequence in chronological order and sent as the actual control command to the hydraulic valve group of the hydraulic propulsion system and the variable frequency drive of the rotary drive system of the drilling and planting machine, which are the two underlying actuators, to drive the drill bit to perform the drilling action of this control cycle.

[0012] In a preferred embodiment, step S4, specifically calculating the model prediction error between the actual tunneling resistance and the first predicted resistance value in the optimal predicted resistance sequence, involves the following steps: After completing a drilling operation of one control cycle, the actual tunneling resistance is obtained and calculated. At the same time, the first predicted resistance value arranged in chronological order is extracted from the optimal predicted resistance sequence recorded and defined in step S3. The actual tunneling resistance is algebraically subtracted from the first predicted resistance value, and the difference is defined as the model prediction error. Before adjusting a set of time-varying parameters of the digital twin model based on the model prediction error, a parameter sensitivity model is constructed. The specific construction process is as follows: Given the current set of time-varying parameter values, the system state vector defined in step S1, the actual control command mentioned in step S3, and the adaptive feature vector generated in step S1, under the model input conditions, the partial derivative of the predicted resistance value with respect to each parameter in the set of time-varying parameters is obtained for the resistance calculation module used to output the predicted resistance value in the digital twin model. All partial derivatives are arranged in parameter order to form a sensitivity row vector. The calculation of the sensitivity row vector is completed online using automatic differentiation technology or parameter perturbation method.

[0013] In a preferred embodiment, the process of using a recursive least squares algorithm to adjust and update a set of time-varying parameters representing the dynamic interaction between the drill string and the formation in the digital twin model online specifically involves: The update is performed using a recursive least squares algorithm with a fading forgetting factor and parameter constraints. This recursive least squares algorithm maintains a parameter estimation covariance matrix corresponding to a set of time-varying parameter dimensions, which is initialized as a diagonal matrix. In each control cycle, the recursive least squares algorithm first uses the calculated model prediction error and the constructed sensitivity row vector to calculate the Kalman gain vector; a fading forgetting factor slightly less than one is introduced during the calculation process. Then, based on the Kalman gain vector, sensitivity row vector, fading forgetting factor and the parameter estimation covariance matrix of the previous time step, the updated parameter estimation covariance matrix of the current time step is calculated by recursive formula. Next, the Kalman gain vector is multiplied by the model prediction error to obtain a set of correction increments for time-varying parameters. These correction increments are then added to the set of time-varying parameters from the previous time step to obtain an unconstrained intermediate update value for a set of time-varying parameters. Apply physical knowledge-based parameter constraints to the unconstrained intermediate update values, and project each time-varying parameter element in the intermediate update values ​​into a preset reasonable value range; Finally, the parameter values ​​processed by projection constraints are formally determined as a set of time-varying parameters updated at the current moment, and written back to the digital twin model.

[0014] The beneficial effects of this invention are as follows: By integrating equipment operating parameters, machine vibration and acoustic emission signals, and prior geological data, a digital twin model reflecting the interaction characteristics of the drilling tool and the formation is constructed and continuously updated. Based on the system's state vector and the model, multiple candidate control command sequences are simulated in parallel with a look-ahead approach. The optimal control command is obtained by using a multi-objective optimization function that integrates resistance stability, energy consumption, and action smoothness. This process achieves intelligent advance prediction and adaptive pre-control for complex geological conditions, effectively suppressing sudden changes and fluctuations in tunneling resistance, significantly improving the stability of the drilling process and the quality of pile formation. At the same time, the energy consumption of the equipment is reduced by optimizing control actions. Furthermore, by using an online parameter update mechanism based on model prediction errors, the digital twin model has the ability to continuously self-evolve, thereby ensuring the adaptability and reliability of the control system throughout its entire life cycle. Attached Figure Description

[0015] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0018] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application. Example

[0019] This embodiment provides, for example Figure 1 The method for intelligent collaborative control of tunneling resistance of a drilling and planting machine based on multi-source sensing includes the following steps: Step S1: Acquire multi-source sensing data of the drilling and planting machine in real time. The multi-source sensing data includes equipment operating parameters, machine response signals, and prior geological and construction data. Based on the machine response signals and prior geological and construction data, generate an adaptive feature vector that characterizes the comprehensive properties of the contact interface between the drill bit and the formation. Based on the multi-source sensing data and the adaptive feature vector, construct and maintain a digital twin model reflecting the interaction between the drilling tool and the formation online, and define the vector composed of equipment operating parameters as the system state vector. Step S2: In each control cycle, obtain the current system state vector of the drilling and planting machine, and use the system state vector as the initial state of the digital twin model; in the digital twin model, for multiple different candidate control command sequences, simulate the response of the drilling and planting machine in a limited prediction time domain in the future, and obtain the predicted system state sequence and the predicted tunneling resistance sequence corresponding to each candidate control command sequence, wherein each predicted tunneling resistance sequence contains multiple predicted resistance values ​​arranged in the order of future sampling times; Step S3: Construct a multi-objective optimization function that includes the objectives of tunneling resistance stability, energy consumption, and control action smoothness. Using the predicted system state sequence and predicted tunneling resistance sequence obtained in Step S2 as inputs, solve the multi-objective optimization function. Select the sequence that makes the function value of the multi-objective optimization function optimal from multiple candidate control command sequences as the optimal control command sequence. Record the predicted tunneling resistance sequence corresponding to the optimal control command sequence and define it as the optimal predicted resistance sequence. Send the first control command in the optimal control command sequence as the actual control command for the current control cycle to the bottom-level execution mechanism of the drilling and planting machine. Step S4: Collect the actual tunneling resistance of the drilling and planting machine after the actual control command is executed; calculate the model prediction error between the actual tunneling resistance and the first predicted resistance value in the optimal predicted resistance sequence; based on the model prediction error, use the recursive least squares algorithm to adjust and update a set of time-varying parameters in the digital twin model that characterize the dynamic interaction between the drill string and the formation online.

[0020] In this embodiment, it is specifically necessary to explain that the process of defining the system state vector in step S1 is as follows: From the real-time acquired multi-source sensing data, the equipment operating parameters are extracted, including the propulsion hydraulic cylinder pressure, the rotary motor torque, the rotary motor speed, and the feed rate. The propulsion hydraulic cylinder pressure reflects the instantaneous force exerted by the drill bit on the formation and is a direct manifestation of the tunneling resistance. The rotary motor torque characterizes the rotational resistance overcome by the drill bit when cutting the formation. The rotary motor speed and the feed rate together determine the instantaneous drilling efficiency of the drill bit. Using the system state vector composed of these four parameters as the basic input of the digital twin model, it can most directly describe the motion and force state of the drill bit. The hydraulic cylinder pressure, rotary motor torque, rotary motor speed, and feed rate are arranged in a preset order to form a system state vector. One possible preset order is [hydraulic cylinder pressure, rotary motor torque, rotary motor speed, feed rate]. This order conforms to the conventional logic of arranging state variables in a control system, i.e., force first and motion later, which facilitates subsequent model matrix calculations. The specific process of generating adaptive feature vectors is as follows: The multi-source sensing data also includes machine response signals and prior geological and construction data. The machine response signals include the raw time-domain vibration signals collected by vibration sensors installed on the power head and drill pipe, and the raw time-domain acoustic emission signals collected by acoustic emission sensors installed on the power head and drill pipe. The prior geological and construction data include the distribution of standard penetration test blows along the depth obtained from the survey and the historical resistance and depth curves of adjacent pile construction records. The vibration sensors are preferably triaxial accelerometers, installed near the flange of the power head, to sense the longitudinal, lateral, and torsional vibrations of the drill string system. The acoustic emission sensors are arranged on the surface of the drill pipe to capture high-frequency stress wave signals generated by formation fracturing and drill bit friction. The distribution of standard penetration test blows along the depth is stored in the form of discrete point sequences or interpolation functions. The historical resistance and depth curves record the average tunneling resistance value of the completed pile holes at the corresponding depth. The raw vibration time-domain signal acquired by the vibration sensor and the raw acoustic emission time-domain signal acquired by the acoustic emission sensor are processed synchronously. For the raw vibration time-domain signal, the wavelet packet energy spectrum entropy within a set time window is calculated to obtain the vibration characteristic value. For the raw acoustic emission time-domain signal, the average of the ratios of rise time to amplitude of all signal events within the same time window is calculated to obtain the acoustic emission characteristic value. The length of the set time window can be set according to the control period and signal stability, for example, 0.1 seconds to 0.5 seconds. When calculating the wavelet packet energy spectrum entropy, the window is first... The original time-domain signal of the vibration inside the mouth is decomposed into 3-level wavelet packet decomposition to obtain the energy of 8 sub-frequency bands. The entropy value of this energy distribution is calculated as the vibration characteristic value. This value can effectively characterize the disorder of the vibration and the heterogeneity of the formation. For the original time-domain signal of acoustic emission, an amplitude threshold (e.g., 50dB) is set to identify valid events. The ratio of the rise time (from 10% to 90% of the threshold) to the amplitude (RA value) of each event is calculated. Then, the RA values ​​of all events in the window are arithmetically averaged to obtain the acoustic emission characteristic value. This average value can reflect the brittle-plastic tendency of the formation. Furthermore, from the distribution of standard penetration test (SPT) blows along the depth obtained from the exploration and the historical resistance and depth curves of adjacent pile construction records, the SPT blows corresponding to the current drill bit depth are extracted; the drill bit depth is obtained in real time through the drill rig depth encoder, and linear interpolation is performed on the distribution data of SPT blows along the depth to obtain the equivalent SPT blows at the current depth, which serves as a scalar feature characterizing the macroscopic hardness of the formation. Finally, the calculated vibration feature values, acoustic emission feature values, and extracted standard penetration blow counts are fused. This fusion operation is performed using a dynamic feature extraction and fusion network. This network utilizes a sliding time window to extract a feature sequence composed of wavelet packet energy spectrum entropy from the continuous raw vibration time-domain signal, and extracts a feature sequence composed of the mean ratio of rise time to amplitude from the continuous raw acoustic emission time-domain signal. The network combines the real-time acquired vibration feature values, acoustic emission feature values, and the standard penetration blow count corresponding to the current depth as input to a nonlinear fusion module. This nonlinear fusion module has weight and bias parameters that can be adjusted according to the model prediction error, and applies a nonlinear activation function to output an adaptive feature vector. The feature extraction and fusion network is essentially a lightweight time series processing and feature mapping structure. The sliding time window can cover the data of the past 2-5 seconds to construct a feature sequence containing temporal context information. The nonlinear fusion module can be a single-layer or multi-layer perceptron. Its weight parameters and bias parameters are initialized to random small values ​​and are adjusted online adaptively using the model prediction error (or cumulative error) calculated in step S4 as the loss signal through the backpropagation algorithm. The nonlinear activation function can be either the ReLU function or the Sigmoid function. This module maps multi-source heterogeneous instantaneous features and short-time series features into a unified, dense adaptive feature vector. This vector comprehensively encodes the dynamic mechanical properties of the current drill bit-formation contact interface. The process of constructing and maintaining an online digital twin model reflecting the interaction between the drill string and the formation based on multi-source sensing data and adaptive feature vectors is as follows: A parameterized nonlinear state-space model is constructed as a digital twin model. In this model, the system state vector at the next time step is obtained by adding three contributions: the first part is the contribution of the current system state vector through a dynamic transfer loop, the characteristics of which are determined by a set of time-varying parameters; the second part is the contribution of a control input vector through a control input loop, the control input vector having the same physical meaning and dimension as the control commands in the candidate control command sequence to be optimized in step S3; and the third part is the contribution of the formation nonlinear reaction force determined by the current adaptive eigenvector. The dynamic transfer loop is mathematically expressed as a state transition matrix multiplied on the left by the current system state vector, the elements of which are part of a set of time-varying parameters used for... To characterize the inherent dynamic characteristics of the drill string system, such as inertia and damping, the mathematical expression of the control input is a control matrix multiplied by the control input vector. The contribution of the formation nonlinear reaction force is calculated through a mathematical mapping module with a fixed structure and an adaptive feature vector as input. This mathematical mapping module can be either a feedforward neural network or a polynomial function. When a feedforward neural network is used, it has a hidden layer, the number of nodes of which is determined according to the dimension of the adaptive feature vector. The connection weights and biases from the input layer to the hidden layer and from the hidden layer to the output layer constitute a part of a set of time-varying parameters used to characterize the formation reaction characteristics. When a polynomial function is used, the coefficients of each term of the polynomial constitute a corresponding part of a set of time-varying parameters. The linear superposition structure of the three parts ensures the interpretability and computational efficiency of the model. The digital twin model receives the system state vector and control input vector from the previous moment, as well as the current adaptive feature vector, as inputs. It outputs the predicted system state vector for the next moment and the predicted tunneling resistance for the current moment. The predicted tunneling resistance is calculated through a nonlinear relationship that takes the adaptive feature vector as input and implicitly incorporates parameters related to formation cutting characteristics and equipment state characteristics. This nonlinear relationship is specifically implemented by a resistance calculation module, which maps the adaptive feature vector to a scalar resistance value. There are two implementation methods: the first is to calculate the weighted Euclidean norm of the adaptive feature vector, i.e., multiplying each component of the adaptive feature vector by a weighting coefficient, taking the square root of the sum of squares, and the weighting coefficients constitute the part of a set of time-varying parameters used for resistance calculation; the second method is a small resistance calculation network with a feedforward neural network at its core. This network takes the adaptive feature vector as input and outputs the predicted tunneling resistance, with its network weights and biases constituting the corresponding part of a set of time-varying parameters. The digital twin model contains a set of... The time-varying parameters are set during the initialization phase based on the equipment's factory calibration data and prior geological exploration knowledge. During subsequent operation, the parameters are iteratively updated online using a recursive least squares algorithm based on the model prediction error between the predicted and actual tunneling resistance. During initialization, the parameters in the state transition matrix can be theoretically estimated based on the physical properties of the drill pipe, such as its size and material, or obtained by consulting the calibration manual. The parameters in the mathematical mapping module (feedforward neural network or polynomial function) and the resistance calculation module (weighted norm coefficient or small resistance calculation network) can be assigned empirical initial values ​​based on the soil layer classification and SPT blow count range provided in the geological exploration report. The recursive least squares algorithm is executed in each control cycle, using the model prediction error between the latest collected actual tunneling resistance and the model's predicted tunneling resistance as information. It recursively updates all elements of a set of time-varying parameters using a regression vector composed of the current system state vector, control input vector, adaptive feature vector, and their possible historical or transformed values, thereby ensuring that the output of the digital twin model continuously approximates the response of the real system.

[0021] In this embodiment, it is specifically necessary to explain that the process of generating multiple different candidate control instruction sequences in step S2 is as follows: For a finite future prediction time domain, control commands including drill bit thrust setting commands and drill bit rotation speed setting commands are configured for each of the multiple consecutive future sampling moments contained within the prediction time domain. The length of the prediction time domain is typically set to cover the time span from when the device receives the control command to when it generates a complete dynamic response, and is sufficient to evaluate the control effect, such as 2 to 5 seconds in the future. The interval between future sampling moments is consistent with the control cycle, such as 0.1 seconds. By combining various control parameter configuration rules, multiple candidate control command sequences with differentiated control instructions are generated. These various control parameter configuration rules include rules that keep the thrust setting command constant while changing the speed setting command in the prediction time domain, rules that keep the speed setting value constant while changing the thrust setting command, and rules that make the thrust setting command and speed setting command change in synergy according to a preset relationship. The "preset relationship synergy rule" can be specifically embodied as a proportional-integral-derivative (PID) tracking relationship, a fuzzy logic relationship, or a pairing relationship based on experience lookup tables. For example, a synergy rule can be defined as "the ratio of the thrust setting command to the speed setting command remains constant in the prediction time domain"; another rule can be defined as "when the predicted resistance tends to increase, the thrust and speed are simultaneously increased slightly to maintain drilling efficiency". By setting different parameters for each rule (such as constant force value, changing speed gradient, and synergy coefficient), and discretizing or combining them within their reasonable value range, dozens to hundreds of representative candidate control command sequences can be systematically generated, forming an instruction space covering the main operating modes. The process of parallel simulation of multiple different candidate control command sequences in a digital twin model is as follows: Within the same control cycle, multiple parallel computation processes are initiated in the computational environment of the digital twin model. Each parallel computation process constitutes an independent simulation task. The number of independent simulation tasks is equal to the number of multiple different candidate control instruction sequences. Each independent simulation task uses the currently acquired system state vector as a common initial state and loads one of multiple different candidate control instruction sequences as input. Subsequently, each independent simulation task independently and recursively calculates the predicted system state vector of the drilling and implantation machine at the end of each future sampling moment in the future prediction time domain, based on the dynamic relationship embedded in the digital twin model that describes how the system state vector evolves from the current moment to the next moment. The parallel computation process can be implemented using a multi-core CPU or GPU parallel computing architecture. Each independent simulation task is assigned to an independent computing thread or computing core for execution, thereby achieving efficient synchronous completion of a large number of simulation tasks and meeting the real-time control requirements for computational timeliness. Each recursive calculation depends on three inputs: the first is the predicted system state vector at the end of the previous future sampling time; the second is the control command corresponding to the current future sampling time from the candidate control command sequence; and the third is an adaptive feature vector corresponding to the current future sampling time, which is obtained by calculating the drilling depth based on the feed rate contained in the predicted system state vector and combining it with the prior geological and construction data defined in step S1. The specific process for calculating the adaptive feature vector corresponding to the future sampling time is as follows: Based on the feed rate contained in the predicted system state vector at the end of the previous future sampling time, the estimated drilling depth at the end of this future sampling time is calculated. Then, this estimated drilling depth is interpolated using the standard penetration test (SPT) blow count distribution data along the depth defined in step S1 to obtain the SPT blow count corresponding to this future sampling time. Subsequently, based on this SPT blow count and referring to the prior geological and construction data defined in step S1 and the pre-established stratigraphic characteristic mapping relationship, the adaptive feature vector corresponding to this future sampling time is generated or corrected. The adaptive feature vector at future sampling times; the "pre-established formation characteristic mapping relationship" can be an empirical formula or a small lookup table, which maps the standard penetration test (SPT) blow count to the adjustment amount of the relevant components in the adaptive feature vector. For example, when the SPT blow count is higher than a certain threshold (such as 30 blows), the component value representing formation hardness in the adaptive feature vector can be increased accordingly; when the SPT blow count is lower than a certain threshold (such as 10 blows), the component value can be decreased accordingly. This mapping relationship is set according to regional geological experience during initialization and can be indirectly optimized in the model update in step S4. The specific calculation process of the predicted system state vector is as follows: The predicted system state vector at the end of the previous future sampling time is multiplied by a matrix multiplication operation with the state transition matrix in the digital twin model, whose elements are determined by a set of time-varying parameters. The control command corresponding to the current future sampling time is multiplied by a matrix multiplication operation with the control input matrix in the digital twin model. The calculated adaptive feature vector corresponding to the current future sampling time is input into the mathematical mapping module in the digital twin model, which characterizes the formation nonlinear reaction force, to obtain the contribution of the formation nonlinear reaction force. Finally, the result obtained from the state transition matrix operation is used by the control input matrix... The three parts of the numerical values ​​are added together: the result obtained from the input matrix operation, the contribution of the nonlinear reaction force of the formation obtained from the mathematical mapping module, and the predicted system state vector at the end of the future sampling time. This calculation process is the realization of the core dynamic equation of the digital twin model at each prediction step. The matrix multiplication operation is performed according to the rules of linear algebra. The dimensions of the state transition matrix and the control input matrix match the dimensions of the system state vector and the control command. The output of the mathematical mapping module is a vector with the same dimension as the system state vector, representing the influence of the formation reaction force on each state variable of the system. The sum of the three contributions is the sum of the corresponding elements of the vector. The specific calculation process for predicting tunneling resistance is as follows: The calculated adaptive feature vector corresponding to the current future sampling time is input into the resistance calculation module contained in the digital twin model. The resistance calculation module calculates the weighted Euclidean norm of the adaptive feature vector, or processes it through a small resistance calculation network with a feedforward neural network as its core, and outputs the predicted resistance value at the future sampling time. If the weighted Euclidean norm method is used, the weighting coefficient is positive, and its initial value can be set according to the empirical influence of different feature components on resistance. For example, vibration features are given higher weights, and acoustic emission features are given medium weights. If a small resistance calculation network is used, its structure can be a single hidden layer network with 3-5 neurons. The network is trained together during the model parameter update process. Through this recursive simulation that integrates forward-looking stratigraphic information, each independent simulation task outputs a predicted system state sequence corresponding to the loaded candidate control command sequence. The predicted system state sequence consists of multiple predicted system state vectors arranged in the order of future sampling times. Simultaneously, a predicted tunneling resistance sequence is calculated and output at each future sampling time within the same prediction time domain. The predicted tunneling resistance sequence consists of multiple predicted resistance values ​​arranged in the order of future sampling times. The predicted tunneling resistance sequence output by each independent simulation task intuitively shows the possible trajectory of tunneling resistance changes over a period of time under the action of the corresponding candidate control command sequence, such as whether it is stable, rising, falling, or oscillating. These sequences are the most direct quantitative basis for multi-objective optimization decision-making in step S3.

[0022] In this embodiment, it is specifically necessary to explain that in step S3, the process of constructing a multi-objective optimization function that includes the tunneling resistance stability objective, energy consumption objective, and control action smoothness objective is as follows: A multi-objective optimization function is used to calculate a comprehensive performance evaluation value for each candidate control command sequence. This comprehensive performance evaluation value is obtained by weighted summation of three sub-evaluation values ​​at each future sampling time in the prediction time domain. The three sub-evaluation values ​​correspond to the tunneling resistance stability objective, the energy consumption objective, and the control action smoothness objective, respectively. The sub-evaluation value for the tunneling resistance stability objective is calculated based on the predicted resistance value at the current future sampling time. Its calculation considers not only the deviation between the predicted resistance value and a preset expected resistance value, but also the degree of change in the predicted resistance value relative to the previous future sampling time. The stability of the resistance is evaluated by combining deviation and rate of change metrics. The specific calculation process is as follows: First, the difference between the predicted resistance value at the current future sampling time and the preset expected resistance value is calculated. Then, the absolute value of this difference is divided by a preset resistance tolerance bandwidth parameter used to measure the allowable fluctuation range of the resistance, resulting in... A standardized resistance deviation is defined, with the expected resistance value set based on 70% of the equipment's rated load and the resistance tolerance bandwidth parameter set based on 10% of the rated load. For example, if the rated tunneling resistance is 100 kN, the expected resistance value can be set to 70 kN, and the resistance tolerance bandwidth parameter can be set to 10 kN. Next, the resistance change between the predicted resistance value at the current future sampling time and the predicted resistance value at the previous future sampling time is calculated, and the absolute value of this resistance change is multiplied by a rate-of-change penalty coefficient used to penalize rapid changes in resistance. The rate-of-change penalty coefficient can be set according to the control cycle and equipment response characteristics, for example, set to 0.5, to impose a significant penalty on cases where the resistance change exceeds 20 kN per second. Then, the standardized resistance deviation is multiplied by the rate-of-change penalty coefficient, and one is subtracted from the product. The natural logarithm is calculated, and the absolute value is taken to obtain the component evaluation value of resistance stability. This calculation process imposes an exponential penalty on the resistance deviation from the reference value and provides additional suppression for drastic changes. The sub-item evaluation values ​​of the energy consumption target are calculated online using a power consumption estimation model that characterizes the power characteristics of the equipment. This is based on the rotary motor torque, rotary motor speed, and propulsion hydraulic cylinder pressure in the predicted system state vector at the current future sampling time, as well as the control commands in the candidate control command sequence that correspond to the drill bit propulsion force setting command and drill bit rotary speed setting command at that future sampling time. The power consumption estimation model can be constructed based on the motor power calculation formula and the hydraulic system power calculation formula. In this model, the motor power is proportional to the product of the rotary motor torque and the rotary motor speed, and the hydraulic system power is proportional to the product of the propulsion hydraulic cylinder pressure and the feed speed. The system's inherent loss coefficient is then added to estimate the instantaneous total power consumption under the current operating condition. The sub-item evaluation value of the control action smoothness objective is calculated based on the change amplitude of the control command corresponding to the current future sampling time relative to the control command corresponding to the previous future sampling time in the candidate control command sequence. This is used to penalize drastic fluctuations in the control command. The specific calculation process is as follows: calculate the sum of squares of the differences between the corresponding elements of the control command at the current future sampling time and the control command at the previous future sampling time to obtain the sub-item evaluation value of control action smoothness. When summing the sub-item evaluation values ​​at each future sampling time, a future discount factor between zero and one is introduced to gradually reduce the weight of the sub-item evaluation values ​​for more distant future times, so as to reflect the emphasis of control decision on near-term performance. The future discount factor can be set to 0.9, so that the weight of the evaluation value at the Nth future step is 0.9 to the power of (N-1), thus focusing more on improving near-term performance during optimization. At the same time, a variable dynamic weight coefficient is configured for the sub-item evaluation value of the tunneling resistance stability objective. The value of this dynamic weight coefficient is adaptively adjusted according to the magnitude of the predicted resistance value at the current future sampling time. The specific adjustment process is as follows: When the predicted resistance value is greater than the preset first resistance threshold or less than the preset second resistance threshold, the dynamic weighting coefficient automatically increases to the first weighting value to prioritize resistance stability and safety. The first resistance threshold can be set to 85% of the equipment's maximum safe operating resistance, and the second resistance threshold can be set to the minimum resistance that may cause drilling instability or low efficiency, such as 20% of the maximum safe operating resistance. The first weighting value can be set to 1.5 or higher to ensure that the resistance stability target dominates under dangerous or abnormal operating conditions. When the predicted resistance value is between the first and second resistance thresholds, the dynamic weighting coefficient adopts the second weighting value, which is relatively lower than the first weighting value. The second weighting value can be set to 1.0, within this normal resistance range, the resistance stability target is balanced and optimized with energy consumption and motion smoothness targets. The process of solving a multi-objective optimization function and selecting the optimal control command sequence from multiple candidate control command sequences is as follows: Using the predicted system state sequence and predicted tunneling resistance sequence corresponding to each candidate control command sequence obtained in step S2 as input data for calculating the multi-objective optimization function, the comprehensive performance evaluation value corresponding to each candidate control command sequence is calculated in parallel and independently. Parallel computing is implemented using a parallel computing architecture of a multi-core central processing unit or a graphics processing unit. The task of calculating the comprehensive performance evaluation value of each candidate control command sequence is allocated to an independent computing thread or computing core for synchronous execution. After completing the quantitative evaluation of all candidate control command sequences, a minimization comparison process is used to select the smallest comprehensive performance evaluation value from all the calculated comprehensive performance evaluation values. The candidate control command sequence corresponding to the smallest comprehensive performance evaluation value is determined as the one that makes the function value of the multi-objective optimization function optimal, and it is formally determined as the optimal control command sequence. The minimization comparison process is implemented by traversing and comparing the values ​​of all the calculated comprehensive performance evaluation values. Meanwhile, from the set of predicted tunneling resistance sequences output in step S2 that correspond to each candidate control command sequence, the specific predicted tunneling resistance sequence associated with the optimal control command sequence is extracted, and this predicted tunneling resistance sequence is recorded separately and defined as the optimal predicted resistance sequence. This optimal predicted resistance sequence represents the optimal expected trajectory of tunneling resistance changes over a future period of time under the selected optimal control strategy. Finally, from the determined optimal control command sequence, the first control command is extracted in chronological order and used as the actual control command. This command is immediately sent to the hydraulic valve group of the hydraulic propulsion system and the variable frequency drive of the rotary drive system of the drilling and planting machine, which are the two underlying actuators, to drive the drill bit to perform the drilling action of this control cycle. The actual control command includes optimized thrust setpoints and speed setpoints. This command is sent to the proportional control terminal of the hydraulic valve group and the speed setpoint of the variable frequency drive through analog signals or fieldbus digital communication protocols, thereby precisely controlling the thrust of the hydraulic cylinder and the speed of the rotary motor.

[0023] In this embodiment, it is specifically necessary to explain the process of calculating the model prediction error between the actual tunneling resistance and the first predicted resistance value in the optimal predicted resistance sequence in step S4 as follows: After completing a drilling operation in one control cycle, the actual tunneling resistance is acquired and calculated. The actual tunneling resistance is obtained by reading the measured values ​​of the propulsion hydraulic cylinder pressure sensor and the rotary motor torque sensor in real time and calculating them based on the mechanical geometric model of the drill bit. It is a comprehensive scalar representing the drill bit overcoming the total reaction force of the formation. At the same time, the first predicted resistance value, arranged in chronological order, is extracted from the optimal predicted resistance sequence recorded and defined in step S3. This first predicted resistance value represents the optimal expectation of the tunneling resistance at the current moment in the previous control cycle. The actual tunneling resistance is algebraically subtracted from this first predicted resistance value, and the difference is defined as the model prediction error. This model prediction error quantifies the instantaneous deviation generated by the digital twin model when making forward predictions in the previous control cycle. The calculation process of the model prediction error is as follows: subtract the first predicted resistance value in the optimal predicted resistance sequence from the actual tunneling resistance value acquired in the current control cycle to obtain the model prediction error. A positive model prediction error value indicates that the actual resistance is higher than the prediction, and a negative value indicates that the actual resistance is lower than the prediction. This model prediction error will serve as the fundamental basis for driving the correction of model parameters. Before adjusting a set of time-varying parameters of the digital twin model based on the model prediction error, a parameter sensitivity model is constructed. The specific construction process is as follows: Given the current set of time-varying parameter values, the system state vector defined in step S1, the actual control command mentioned in step S3, and the adaptive feature vector generated in step S1, under the model input conditions, for the resistance calculation module in the digital twin model used to output the predicted resistance value, the partial derivative of the predicted resistance value with respect to each parameter in the set of time-varying parameters is calculated. All partial derivatives are arranged in parameter order to form a sensitivity row vector. The calculation of the sensitivity row vector is completed online using automatic differentiation technology or parameter perturbation method. When the resistance calculation module uses a feedforward neural network, the calculation of partial derivatives can be efficiently implemented using the backpropagation algorithm. When the parameter perturbation method is used, a small perturbation (such as 1% of the original value) is applied to each element in the set of time-varying parameters in turn, and the change in the predicted resistance value is observed. The change is divided by the perturbation amount to approximate the partial derivative of the parameter, thereby constructing a complete sensitivity row vector. The process of using the recursive least squares algorithm to adjust and update a set of time-varying parameters representing the dynamic interaction between the drill string and the formation in the digital twin model online is as follows: The update is performed using a recursive least squares algorithm with a fading forgetting factor and parameter constraints. This recursive least squares algorithm maintains a parameter estimation covariance matrix corresponding to a set of time-varying parameter dimensions. It is initialized as a diagonal matrix. The parameter estimation covariance matrix is ​​usually initialized as a scalar multiplied by an identity matrix. The scalar is taken as a large positive number (e.g., 1000) to indicate that the parameter estimation has a large uncertainty in the early stage of the algorithm, so that it is willing to accept a larger correction based on new data. In each control cycle, the recursive least squares algorithm first uses the calculated model prediction error and the constructed sensitivity row vector to calculate the Kalman gain vector. The Kalman gain vector determines the weighting of the current model prediction error on the correction of each time-varying parameter. During the calculation, a fading forgetting factor slightly less than one is introduced. The typical value of the fading forgetting factor is between 0.95 and 0.995. For example, it can be set to 0.98. Its specific value needs to be balanced between parameter tracking speed and estimation stability. The closer the value is to 1, the longer the historical memory and the higher the stability, but the weaker the ability to track time-varying parameters. The smaller the value, the more sensitive it is to the latest data and the stronger the tracking ability, but the estimation results may fluctuate more. Then, based on the Kalman gain vector, sensitivity row vector, fading forgetting factor, and the parameter estimation covariance matrix of the previous time step, the updated parameter estimation covariance matrix of the current time step is calculated by a recursive formula. The recursive formula is as follows: first, multiply the Kalman gain vector by the sensitivity row vector, then subtract this product from the identity matrix, then multiply the result by the parameter estimation covariance matrix of the previous time step, and finally divide the product by the fading forgetting factor. Next, the Kalman gain vector is multiplied by the model prediction error to obtain a set of correction increments for time-varying parameters. These correction increments are then added to the set of time-varying parameters from the previous time step to obtain an unconstrained intermediate update value for a set of time-varying parameters. Physically-based parameter constraints are applied to unconstrained intermediate update values. Each time-varying parameter element in the intermediate update values ​​is projected into a preset reasonable value range. The lower and upper limits of this reasonable value range are determined according to the physical meaning of the parameter. For example, for a parameter characterizing formation cutting damping, the lower limit of its reasonable value range is set to 0, and the upper limit can be set according to the empirical value of the hardest rock layer. For a parameter characterizing the response time of the hydraulic system, its range is determined according to the equipment hydraulic schematic diagram and component manual. The projection operation determines whether the parameter value exceeds the range. If it does, it is forcibly set to the nearest range boundary value. This projection operation ensures that the updated parameters do not violate basic physical laws and maintains the physical interpretability and numerical stability of the model. Finally, the parameter values ​​processed by projection constraints are formally determined as a set of time-varying parameters updated at the current moment. Furthermore, to prevent the parameter estimation covariance matrix from losing positive definiteness or becoming ill-conditioned during long-term recursive calculations, the recursive least squares algorithm also includes a triggered reset mechanism. When the trace of the parameter estimation covariance matrix exceeds a preset threshold or the parameter update amount is negligible over several consecutive periods, the parameter estimation covariance matrix is ​​reset to the initial diagonal matrix. The preset threshold can be set to 10 times the trace of the initial covariance matrix. "The parameter update amount is negligible over several consecutive periods" can be quantified as: within five consecutive control periods, the L2 norm of the correction increment is less than a set of... The current value of the time-varying parameters is one ten-thousandth of the L2 norm. The reset operation injects new uncertainty into the algorithm, preventing it from stopping learning due to overconfidence. This re-injects the recursive least squares algorithm with exploratory capabilities, preventing it from getting stuck in local stagnation. After the update is completed, the latest set of time-varying parameters is written back to the digital twin model. "Writing back to the digital twin model" means updating the corresponding elements of the state transition matrix, the connection weights / coefficients of the mathematical mapping module, and the weighting coefficients / network weights of the resistance calculation module in the nonlinear state-space model constructed in step S1 with the new set of time-varying parameter values, so that the model can make predictions based on more accurate parameters in the next control cycle.

[0024] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0025] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0026] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0027] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0028] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0029] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0030] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent collaborative control of tunneling resistance in a drilling and planting machine based on multi-source sensing, characterized in that, Specifically, the steps include the following: Step S1: Acquire multi-source sensing data of the drilling and planting machine in real time. The multi-source sensing data includes equipment operating parameters, machine response signals, and prior geological and construction data. Based on the machine response signals and prior geological and construction data, generate an adaptive feature vector that characterizes the comprehensive properties of the contact interface between the drill bit and the formation. Based on the multi-source sensing data and the adaptive feature vector, construct and maintain a digital twin model reflecting the interaction between the drilling tool and the formation online, and define the vector composed of equipment operating parameters as the system state vector. Step S2: In each control cycle, obtain the current system state vector of the drilling and planting machine, and use the system state vector as the initial state of the digital twin model; In the digital twin model, for multiple different candidate control command sequences, the response of the drilling and planting machine is simulated in parallel within a finite prediction time domain in the future, to obtain the prediction system state sequence and prediction tunneling resistance sequence corresponding to each candidate control command sequence, wherein each prediction tunneling resistance sequence contains multiple prediction resistance values ​​arranged in the order of future sampling times. Step S3: Construct a multi-objective optimization function that includes the objectives of tunneling resistance stability, energy consumption, and control action smoothness. Using the predicted system state sequence and predicted tunneling resistance sequence obtained in Step S2 as inputs, solve the multi-objective optimization function. Select the sequence that makes the function value of the multi-objective optimization function optimal from multiple candidate control command sequences as the optimal control command sequence. Record the predicted tunneling resistance sequence corresponding to the optimal control command sequence and define it as the optimal predicted resistance sequence. Send the first control command in the optimal control command sequence as the actual control command for the current control cycle to the bottom-level execution mechanism of the drilling and planting machine. Step S4: Collect the actual tunneling resistance of the drilling and planting machine after the actual control command is executed; Calculate the model prediction error between the actual tunneling resistance and the first predicted resistance value in the optimal predicted resistance sequence; Based on the model prediction error, a set of time-varying parameters characterizing the dynamic interaction between the drill string and the formation in the digital twin model are adjusted and updated online using the recursive least squares algorithm.

2. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 1, characterized in that: In step S1, the process of defining the system state vector is as follows: The equipment operating parameters are extracted from the real-time multi-source sensing data, including the pressure of the propulsion hydraulic cylinder, the torque of the rotary motor, the speed of the rotary motor, and the feed speed. The hydraulic cylinder pressure, rotary motor torque, rotary motor speed, and feed speed are arranged in a preset order to form a system state vector.

3. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 2, characterized in that: The specific process for generating adaptive feature vectors is as follows: The multi-source sensing data also includes body response signals and prior geological and construction data. The body response signals include the raw time-domain vibration signals collected by vibration sensors installed on the power head and drill pipe, and the raw time-domain acoustic emission signals collected by acoustic emission sensors installed on the power head and drill pipe. The prior geological and construction data include the distribution of standard penetration blows along the depth obtained from the exploration and the historical resistance and depth curves of adjacent pile construction records. The original time-domain vibration signal acquired by the vibration sensor and the original time-domain acoustic emission signal acquired by the acoustic emission sensor are processed synchronously. For the original time-domain vibration signal, the wavelet packet energy spectrum entropy within a set time window is calculated to obtain the vibration characteristic value. For the original time-domain acoustic emission signal, the average value of the ratio of rise time to amplitude of all signal events within the same time window is calculated to obtain the acoustic emission characteristic value. Extract the standard penetration test blows corresponding to the current drill bit depth from the distribution of standard penetration test blows along the depth obtained from the exploration and the historical resistance and depth curves of adjacent pile construction records. Finally, the calculated vibration feature values, acoustic emission feature values, and extracted standard penetration blow counts are fused. This fusion operation is performed by a dynamic feature extraction and fusion network. This network uses a sliding time window to extract a feature sequence composed of wavelet packet energy spectrum entropy from the continuous original vibration time-domain signal and a feature sequence composed of the mean of the rise time to amplitude ratio from the continuous original acoustic emission time-domain signal. The real-time acquired vibration feature values, acoustic emission feature values, and standard penetration blow counts corresponding to the current depth are combined as inputs and fed into a nonlinear fusion module. A nonlinear activation function is then applied to output an adaptive feature vector.

4. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 3, characterized in that: The process of constructing and maintaining a digital twin model reflecting the interaction between the drill string and the formation online based on multi-source sensing data and adaptive feature vectors is as follows: A parameterized nonlinear state-space model is constructed as a digital twin model. In this digital twin model, the system state vector at the next moment is obtained by adding the current system state vector through a dynamic transmission link, the control input vector through a control input link, and the contribution of the formation nonlinear reaction force determined by the current adaptive eigenvector. The characteristics of the dynamic transmission link, the parameters involved in the calculation of the contribution of the nonlinear reaction force of the stratum, and the parameters involved in the calculation of the predicted tunneling resistance together constitute a set of time-varying parameters. The digital twin model uses a recursive least squares algorithm to iteratively update a set of time-varying parameters online based on the model prediction error between the predicted tunneling resistance and the actual tunneling resistance.

5. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 4, characterized in that: In step S2, the process of generating multiple different candidate control instruction sequences is as follows: For a finite future prediction time domain, for multiple consecutive future sampling moments contained within the prediction time domain, control commands including drill bit thrust setting commands and drill bit rotation speed setting commands are configured for each future sampling moment; By combining various control parameter configuration rules, multiple candidate control command sequences with differences in control commands are generated. These various control parameter configuration rules include rules that keep the thrust setting command constant while changing the speed setting command in the prediction time domain, rules that keep the speed setting value constant while changing the thrust setting command, and rules that make the thrust setting command and speed setting command change in synergy according to a preset relationship.

6. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 5, characterized in that: The process of performing parallel simulation of multiple different candidate control command sequences in the digital twin model is as follows: Within the same control cycle, multiple parallel computation processes are initiated in the computational environment of the digital twin model. Each parallel computation process constitutes an independent simulation task. The number of independent simulation tasks is equal to the number of multiple different candidate control command sequences. Each independent simulation task takes the currently acquired system state vector as a common initial state and loads one of the multiple different candidate control command sequences as input. Subsequently, each independent simulation task recursively calculates the predicted system state vector of the drilling and planting machine at the end of each future sampling time in the future prediction time domain based on the digital twin model. Each recursive calculation depends on three inputs: the first is the predicted system state vector at the end of the previous future sampling time; the second is the control command corresponding to the current future sampling time from the candidate control command sequence; and the third is an adaptive feature vector corresponding to the current future sampling time, which is obtained by calculating the drilling depth based on the feed rate contained in the predicted system state vector and combining it with the prior geological and construction data defined in step S1. Through this recursive simulation that integrates forward-looking stratigraphic information, each independent simulation task ultimately outputs a predicted system state sequence corresponding to the loaded candidate control command sequence. The predicted system state sequence consists of multiple predicted system state vectors arranged in the order of future sampling times. Simultaneously, a predicted tunneling resistance sequence is calculated and output at each future sampling time within the same prediction time domain. The predicted tunneling resistance sequence consists of multiple predicted resistance values ​​arranged in the order of future sampling times.

7. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 6, characterized in that: In step S3, the process of constructing a multi-objective optimization function that includes the objectives of tunneling resistance stability, energy consumption, and control motion smoothness is as follows: The multi-objective optimization function is used to calculate a comprehensive performance evaluation value for each candidate control command sequence. This comprehensive performance evaluation value is obtained by weighted summation of three sub-evaluation values ​​at each future sampling time in the prediction time domain. The three sub-evaluation values ​​correspond to the tunneling resistance stability target, energy consumption target, and control action smoothness target, respectively. When summing the sub-item evaluation values ​​at each future sampling time, a future discount factor between zero and one is introduced; at the same time, a variable dynamic weight coefficient is configured for the sub-item evaluation value of the tunneling resistance stability target, and the value of the dynamic weight coefficient is adaptively adjusted according to the magnitude of the predicted resistance value at the current future sampling time.

8. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 7, characterized in that: The process of solving the multi-objective optimization function and selecting the optimal control command sequence from multiple candidate control command sequences is as follows: Using the predicted system state sequence and predicted tunneling resistance sequence corresponding to each candidate control command sequence obtained in step S2 as input data for calculating the multi-objective optimization function, the comprehensive performance evaluation value corresponding to each candidate control command sequence is calculated in parallel and independently. After completing the quantitative evaluation of all candidate control command sequences, a minimization comparison process is used to select the comprehensive performance evaluation value with the smallest value from all the calculated comprehensive performance evaluation values. The candidate control command sequence corresponding to the smallest comprehensive performance evaluation value is determined as the one that makes the function value of the multi-objective optimization function optimal, and it is formally determined as the optimal control command sequence. Meanwhile, from the set of predicted tunneling resistance sequences output in step S2 that correspond to each candidate control command sequence, the specific predicted tunneling resistance sequence associated with the optimal control command sequence is extracted, and this predicted tunneling resistance sequence is recorded separately and defined as the optimal predicted resistance sequence. Finally, the first control command is extracted from the determined optimal control command sequence in chronological order and sent as the actual control command to the hydraulic valve group of the hydraulic propulsion system and the variable frequency drive of the rotary drive system of the drilling and planting machine, which are the two underlying actuators, to drive the drill bit to perform the drilling action of this control cycle.

9. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 8, characterized in that: In step S4, the process of calculating the model prediction error between the actual tunneling resistance and the first predicted resistance value in the optimal predicted resistance sequence is as follows: After completing a drilling operation of one control cycle, the actual tunneling resistance is obtained and calculated. At the same time, the first predicted resistance value arranged in chronological order is extracted from the optimal predicted resistance sequence recorded and defined in step S3. The actual tunneling resistance is algebraically subtracted from the first predicted resistance value, and the difference is defined as the model prediction error. Before adjusting a set of time-varying parameters of the digital twin model based on the model prediction error, a parameter sensitivity model is constructed. The specific construction process is as follows: Given the current set of time-varying parameter values, the system state vector defined in step S1, the actual control command mentioned in step S3, and the adaptive feature vector generated in step S1, under the model input conditions, the partial derivative of the predicted resistance value with respect to each parameter in the set of time-varying parameters is obtained for the resistance calculation module used to output the predicted resistance value in the digital twin model. All partial derivatives are arranged in parameter order to form a sensitivity row vector. The calculation of the sensitivity row vector is completed online using automatic differentiation technology or parameter perturbation method.

10. The intelligent collaborative control method for tunneling resistance of an integrated drilling and planting machine based on multi-source sensing as described in claim 9, characterized in that: The process of using the recursive least squares algorithm to adjust and update a set of time-varying parameters representing the dynamic interaction between the drill string and the formation in the digital twin model online is as follows: The update is performed using a recursive least squares algorithm with a fading forgetting factor and parameter constraints. This recursive least squares algorithm maintains a parameter estimation covariance matrix corresponding to a set of time-varying parameter dimensions, which is initialized as a diagonal matrix. In each control cycle, the recursive least squares algorithm first uses the calculated model prediction error and the constructed sensitivity row vector to calculate the Kalman gain vector; a fading forgetting factor slightly less than one is introduced during the calculation process. Then, based on the Kalman gain vector, sensitivity row vector, fading forgetting factor and the parameter estimation covariance matrix of the previous time step, the updated parameter estimation covariance matrix of the current time step is calculated by recursive formula. Next, the Kalman gain vector is multiplied by the model prediction error to obtain a set of correction increments for time-varying parameters. These correction increments are then added to the set of time-varying parameters from the previous time step to obtain an unconstrained intermediate update value for a set of time-varying parameters. Apply physical knowledge-based parameter constraints to the unconstrained intermediate update values, and project each time-varying parameter element in the intermediate update values ​​into a preset reasonable value range; Finally, the parameter values ​​processed by projection constraints are formally determined as a set of time-varying parameters updated at the current moment, and written back to the digital twin model.

Images