Intelligent shield muck improvement method and system based on machine learning

Abstract

The invention discloses an intelligent shield muck improvement method and system based on machine learning, and the method and system achieve the intellectualization and precision of muck improvement in the tunneling process of a shield tunneling machine through integrating muck image classification, improvement parameter prediction and model optimization. Firstly, a YOLOV8 network is utilized to classify muck images in real time, classification randomness is corrected in combination with an XGBOOST model, and classification accuracy and stability are improved. Secondly, performing time sequence prediction on the muck improvement parameters based on an LSTM network, and introducing a linear regression and genetic algorithm to improve the prediction precision; and finally, optimal solution is carried out on the muck improvement parameters through a particle swarm optimization (PSO) algorithm, and the stability and efficiency of shield tunneling are remarkably improved. The method solves the problems of randomness and hysteresis caused by dependence on manual adjustment in the prior art, has the advantages of high efficiency and accuracy, and is suitable for automatic application scenes of shield engineering.

Description

Technical Field

[0001] The present invention belongs to the technical field of shield tunneling construction, and in particular relates to an intelligent shield slag improvement method and system based on machine learning. Background Art

[0002] Shield construction is a tunneling technology widely used in underground engineering. During shield tunneling, soil improvement is an important part of ensuring stable pressure in the tunneling chamber, smooth slag discharge and efficient operation of the shield machine. Traditional soil improvement methods rely on the operator's experience and judgment, and manually adjust improvement parameters such as foam, water and bentonite flow by observing the tunneling status and slag outlet images. This method has obvious randomness and hysteresis, which not only affects the tunneling efficiency, but may also cause equipment damage or engineering quality problems.

[0003] With the development of artificial intelligence technology, the application of machine learning algorithms in the engineering field has gradually increased, but existing research on the intelligent improvement of shield slag is still limited to a single task, such as slag classification or parameter prediction, and lacks a systematic solution from data processing, prediction to optimization. Therefore, a comprehensive method is urgently needed to realize the intelligent and efficient improvement of shield slag based on machine learning technology. Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes an intelligent shield slag improvement method and system based on machine learning, and establishes a full-process intelligent slag improvement system by integrating image classification, parameter prediction and optimization algorithms.

[0005] The first aspect of the present invention is a method for improving soil in intelligent shield tunneling based on machine learning, comprising the following steps:

[0006] (1) Collecting images of soil and related sensor parameters during shield tunneling;

[0007] (2) The improved YOLOV8 network is used to classify the muck images. The XGBOOST model analyzes the feature importance in the muck image classification and adjusts the model output probability to correct the classification randomness and obtain the accurate muck type probability;

[0008] (3) Using the LSTM network to input the sensor parameter time series during the excavation process, combined with the probability of soil type, and combined with linear regression and genetic algorithm, the soil improvement parameters required in the future excavation process are predicted;

[0009] (4) Using the particle swarm optimization (PSO) algorithm, with the probability of soil type and tunneling stability as the objective function, the optimal solution of bentonite flow, foam liquid flow, and foam air flow is solved;

[0010] (5) Feeding back the optimized parameters in step (4) to the shield machine control system to achieve real-time monitoring and dynamic adjustment during the tunneling process.

[0011] Furthermore, the YOLOV8 network is based on an improved VanillaNet structure, which improves the accuracy of slag classification to more than 98% through feature extraction, enhancement and classification prediction.

[0012] Specifically, the output calculation formula of YOLOV8 in step (2) is:

[0013] P cls (i) = σ(W cls ·f i )+b cls

[0014] Among them, P cls (i) is the classification probability of the i-th class, W cls is the classification weight, f i is the image feature vector, σ is the sigmoid activation function, b cls is the bias term.

[0015] Specifically, the correction process of the XGBOOST model in step (2) adopts the following expression:

[0016]

[0017] Among them, P XGBoost (i) is the corrected classification probability of the i-th class, α k is the weight of each feature, f k (x) is the contribution of the kth feature on sample x, and N is the total number of features.

[0018] Specifically, the formula of the LSTM network in step (3) is:

[0019] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0020] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0021]

[0022]

[0023] ot =σ(W o ·[h t-1 ,x t ]+b o );

[0024] h t =o t tanh(C t );

[0025] Among them, f t 、i t , are the forget gate, input gate and candidate memory state respectively, C t is the memory unit state, h t is the output at the current moment, h t-1 is the hidden state of the previous time step, x t is the input of the current time step, σ is the Sigmoid activation function, W f ,b f are the weight and bias of the forget gate, W i ,b i are the weight and bias of the input gate, W C ,b C are the weight and bias of the candidate memory state, W o ,b o are the weight and bias of the output gate respectively;

[0026] The calculation formula of the linear regression is:

[0027]

[0028] Among them, w is the weight of linear regression, b is the bias term, and X scaled is the standardized feature column, is the target column;

[0029] The fitness function of the genetic algorithm is:

[0030]

[0031] Among them, individual is the individual in the genetic algorithm, representing an eigenvalue vector; w and b are the weight and bias terms of the linear regression model; It is the target value predicted by the current individual (the probability of slag after treatment); penalty is the penalty term, which is used to deal with the penalty when the characteristic value exceeds the predetermined range.

[0032] Specifically, the fitness calculation formula of the PSO algorithm in step (4) is:

[0033]

[0034] Among them, Z is the value of the fitness function, that is, the value of the objective function; c 0 is the bias term; c i is the coefficient of the first-order term, which represents the coefficient of bentonite flow, foam liquid flow and foam air flow; c ij is the coefficient of the quadratic term, which means x i and x j The cross term of x i is the i-th component of the particle position vector, corresponding to the bentonite flow rate, foam liquid flow rate, or foam air flow rate.

[0035] The second aspect of the present invention is an intelligent shield slag improvement system based on machine learning, comprising the following modules:

[0036] Acquisition parameter module: collects the muck images and related sensor parameters during the shield tunneling process; muck image real-time classification module: uses the improved YOLOV8 network to classify the muck images, corrects the classification randomness through the XGBOOST model, and obtains the accurate probability of muck type;

[0037] Muck improvement parameter prediction module: The improved YOLOV8 network is used to classify the muck images. The XGBOOST model analyzes the feature importance in the muck image classification, adjusts the model output probability to correct the classification randomness, and obtains the accurate muck type probability;

[0038] Prediction parameter module: Use LSTM network to input sensor parameter time series during excavation, combine with the probability of soil type, and combine linear regression and genetic algorithm to predict soil improvement parameters required in future excavation process;

[0039] Parameter optimization module: Particle swarm optimization (PSO) algorithm is used to solve the optimal solution of bentonite flow, foam liquid flow and foam air flow with soil type probability and excavation stability as the objective function;

[0040] Real-time adjustment module: Feedback the parameters optimized by the parameter optimization module to the shield machine control system to achieve real-time monitoring and dynamic adjustment during the excavation process.

[0041] The third aspect of the present invention is an electronic device, characterized in that it includes:

[0042] one or more processors;

[0043] A memory for storing one or more programs;

[0044] When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent shield slag improvement method based on machine learning.

[0045] The fourth aspect of the present invention: a computer-readable storage medium having computer instructions stored thereon, characterized in that when the instructions are executed by a processor, the steps of the intelligent shield slag improvement method based on machine learning are implemented.

[0046] The beneficial effects of the present invention are as follows:

[0047] The present invention can significantly improve shield tunneling efficiency, reduce randomness and hysteresis caused by human intervention, and has high promotion value in actual engineering. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 The technical roadmap of the intelligent shield segment slag improvement method based on machine learning of the present invention;

[0050] Figure 2 It is a schematic diagram of the slag improvement process of the present invention;

[0051] Figure 3 A schematic diagram of a parameter prediction route for slag improvement according to the present invention;

[0052] Figure 4 This is a schematic diagram of the optimization route of the slag improvement parameters of the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0054] The present invention discloses an intelligent shield slag improvement method and system based on machine learning, and combines the following Figure 1 , Figure 2 and Figure 3 The embodiments of the present invention are described in further detail.

[0055] Step 1: Real-time classification of soil images

[0056] (1.1) Data collection and preprocessing

[0057] During the shield tunneling process, the slag sent out by the screw conveyor is captured in real time by a high-resolution camera at the slag outlet to ensure that the image is clear and has no obvious noise. The image data is first extracted by the camera, and then manually classified by the shield machine driver according to the actual situation, and the slag type (such as dry, normal, wet, etc.) is marked. Subsequently, the image data is processed using the YOLOv8 network, and the input of the model includes images that have been resolution-normalized and enhanced. In order to improve the inference speed, the network backbone structure adopts a lightweight and optimized version of VanillaNet.

[0058] (1.2) Classification network training

[0059] First, the YOLOV8 network was trained using a dataset containing different soil types (dry, normal, and wet). The ratio of each type of sample was roughly 1:1:1 to ensure balanced data distribution. During the training process, considering the randomness of the classification results, the XGBOOST model was used to analyze the correlation between historical data (such as propulsion pressure, torque, etc.) and the YOLOV8 classification results, so as to correct the possible deviations of the classification network and improve the accuracy of the model.

[0060] (1.3) Classification result output

[0061] After training, the YOLOV8 network classifies each slag image as dry, normal or wet slag, and outputs the corresponding classification results ("0, 1, 2" represent dry, normal and wet slag respectively). The classification results will be used as reference labels for the XGBOOST model, which further calculates the classification probabilities of each category based on historical data (such as propulsion pressure, torque, etc.) and the classification results of YOLOV8. The output calculation formula of the YOLOV8 is:

[0062] P cls (i) = σ(W cls ·f i )+b cls ;

[0063] Among them, P cls (i) is the classification probability of the i-th class, W cls is the classification weight, f i is the image feature vector, σ is the sigmoid activation function, b cls is the bias term.

[0064] The correction process of the XGBOOST model uses the following expression:

[0065]

[0066] Among them, P XGBoost (i) is the corrected classification probability of the i-th class, α k is the weight of each feature, f k (x) is the contribution of the kth feature on sample x, and N is the total number of features.

[0067] In this way, the reliance on absolute classification results is reduced, and the flexibility and accuracy of classification are improved.

[0068] Step 2: Prediction of soil improvement parameters

[0069] (2.1) Historical data input

[0070] The relevant parameters recorded by sensors during shield construction (such as cutter head torque (X1), average propulsion speed (X2), average soil pressure (X3), bentonite flow (X4), foam liquid flow (X5), foam air flow (X6) and excavation stability parameter (Y2, calculated manually)) were input into the LSTM network to form time series data for training the prediction model. The data length was set to 30 time steps (approximately 30 seconds). These input parameters were obtained through communication with the shield machine driver, taking into full consideration the actual requirements such as computing time and performance.

[0071] (2.2) LSTM network prediction

[0072] The LSTM network is used to extract features from historical data and predict the probability of slag type in the next 30 seconds. The slag type probability output by LSTM is used to reflect the probability of occurrence of different types of slag, thus providing a basis for the subsequent optimization of slag improvement parameters.

[0073] The relevant formula of LSTM network is as follows:

[0074] The formula for the LSTM network is:

[0075] f t =σ(W f ·[h t-1 ,x t ]+b f );

[0076] i t =σ(W i ·[h t-1 ,x t ]+b i );

[0077]

[0078]

[0079] o t =σ(W o ·[h t-1 ,x t ]+b o );

[0080] h t =o t tanh(C t );

[0081] Among them, f t 、i t , are the forget gate, input gate and candidate memory state respectively, C t is the memory unit state, h t is the output at the current moment, h t-1 is the hidden state of the previous time step, x t is the input of the current time step, σ is the Sigmoid activation function, W f ,b f are the weight and bias of the forget gate, W i ,b i are the weight and bias of the input gate, W C ,b C are the weight and bias of the candidate memory state, W o ,b o are the weight and bias of the output gate respectively.

[0082] (2.3) Linear regression and genetic algorithm optimization

[0083] Based on the relationship of the input data (X1,2,3,4,5,6) in the past 30 seconds, the weights and intercepts of the features are obtained through the linear regression model. The input feature weights and intercepts of the linear regression are derived from historical data to describe the parameter relationship in the past time period.

[0084] The calculation formula of the linear regression is:

[0085]

[0086] Among them, w is the weight of linear regression, b is the bias term, and X scaled is the standardized feature column, is the target column;

[0087] Then, the probability value of the type of slag in the next 30 seconds predicted by LSTM is combined with the genetic algorithm for reverse optimization. The goal of the genetic algorithm is to find a feature combination so that the predicted value calculated by these features is as close as possible to the probability value of the type of slag predicted by our LSTM in the next 30 seconds, so as to obtain a more accurate (X1,2,3,4,5,6).

[0088] The fitness function of the genetic algorithm is:

[0089]

[0090] Among them, individual is the individual in the genetic algorithm, representing an eigenvalue vector; w and b are the weight and bias terms of the linear regression model; It is the target value predicted by the current individual (the probability of slag after treatment); penalty is the penalty term used to deal with the penalty when the characteristic value exceeds the predetermined range.

[0091] Step 3: Optimization of soil improvement parameters

[0092] (3.1) Optimization target setting

[0093] This implementation plan uses the particle swarm optimization (PSO) algorithm to optimize the soil improvement effect and excavation stability during the shield machine excavation process. The optimization objectives include the following two main objectives:

[0094] 1. Maximize the probability of normal slag type: By adjusting the slag improvement parameters, the slag type faced by the shield machine during operation is made more "normal", thereby improving excavation efficiency and stability.

[0095] 2. Maximize tunneling stability: This indicator is based on the sum of standard deviations over a period of time from X1 to X6, reflecting the stability of the shield machine during tunneling. The smaller the standard deviation, the smaller the fluctuation during tunneling and the better the stability.

[0096] like Figure 4 As shown in the figure, the optimization objective function takes into account both the slag improvement effect and the excavation stability. The goal is to optimize the slag improvement effect and improve the excavation stability by adjusting the bentonite flow rate, foam liquid flow rate and foam air flow rate.

[0097] (3.2) Particle initialization

[0098] In the particle swarm, each particle represents a combination of soil improvement parameters (X4, X5, X6), where:

[0099] X4: Bentonite flow

[0100] X5: Foam liquid flow

[0101] X6: Foam air flow

[0102] The particles are initialized randomly within the set parameter range, which is determined based on historical experience and actual engineering requirements. For example:

[0103] Bentonite flow rate (X4) range: set according to the conveying capacity of the bentonite system interface and the demand for slag improvement;

[0104] The range of the foam liquid flow rate (X5) and the foam air flow rate (X6): Based on the ratio of the foam liquid and the compressed air to form the foam mixture after being mixed through the foam generator, set a reasonable flow rate range.

[0105] These fluids enter the cutterhead injection port through a common pipeline and participate in the soil improvement process. After the particle position and velocity are initialized, they will be updated according to the fitness function in the subsequent optimization process to find the optimal solution.

[0106] (3.3) Particle update and convergence

[0107] During the iteration process of the particle swarm optimization (PSO) algorithm, particles dynamically adjust their positions and velocities according to a predetermined fitness function. After each iteration, the position and velocity of the particles are updated according to the feedback information of the fitness function, thereby calculating a new fitness value.

[0108] The fitness calculation formula of the PSO algorithm is:

[0109]

[0110] Among them, Z is the value of the fitness function, that is, the value of the objective function; c 0 is the bias term; c i is the coefficient of the first-order term, which represents the coefficient of bentonite flow, foam liquid flow and foam air flow; c ij is the coefficient of the quadratic term, which means x i and x j The cross term of x i is the i-th component of the particle position vector, corresponding to the bentonite flow rate, foam liquid flow rate, or foam air flow rate.

[0111] The fitness function takes into account the following two key factors:

[0112] (a) Soil type probability error: This factor measures the deviation between the PSO model prediction result and the LSTM prediction value through the soil type probability predicted by the LSTM model. Error evaluation usually uses the root mean square error (RMSE) or other error measurement methods. The optimization goal is to minimize the prediction error to improve the accurate prediction of soil type.

[0113] (b) Tunneling stability parameter: This index reflects the stability of the shield machine during tunneling by calculating the sum of the standard deviations of related parameters such as bentonite flow (X1), foam liquid flow (X2), and foam air flow (X3). The smaller the standard deviation, the smaller the fluctuation during tunneling and the better the stability of the system. This stability index is used as an evaluation criterion in the PSO optimization process, and the optimization goal is to maximize the stability of the tunneling process.

[0114] During the iteration process of the PSO algorithm, particles continuously adjust their positions and speeds, gradually approaching the optimal solution to optimize the combination of bentonite flow (X4), foam liquid flow (X5) and foam air flow (X6), thereby achieving dual optimization of slag improvement effect and excavation stability.

[0115] In this scheme, the optimization process not only considers the soil improvement effect in the next 30 seconds, but also optimizes the soil improvement parameters in the past 30 seconds. By using the PSO algorithm twice, the best soil improvement parameters in the next 30 seconds and the past 30 seconds are obtained. The two optimization results will be added according to the fitness weight, and the smaller fitness will get a larger weight, thereby ensuring the overall balance and improvement of the optimization target. Through this mechanism, historical data and future predictions can be effectively combined to obtain more accurate and stable soil improvement parameters.

[0116] (c) Convergence: After multiple iterations, the particle swarm can converge quickly within 10 iterations, the RMSE is significantly reduced, and the optimization objective is effectively improved.

[0117] Step 4: System Integration and Feedback

[0118] (4.1) Integrated system

[0119] The above modules are connected to the shield machine control system through the data bus, and the classification results and parameter prediction values ​​are received in real time, and fed back to the tunneling control logic. The system is implemented in C++ to ensure that the reasoning speed meets the real-time requirements of the shield machine.

[0120] (4.2) Dynamic Adjustment

[0121] According to the current slag classification and predicted parameters, the injection flow of bentonite and foaming agent is automatically adjusted, and the adjustment effect is monitored in real time through sensors to form a closed-loop control.

[0122] The present invention also discloses an intelligent shield slag improvement system based on machine learning, which includes the following modules:

[0123] Acquisition parameter module: collects the muck images and related sensor parameters during the shield tunneling process; muck image real-time classification module: uses the improved YOLOV8 network to classify the muck images, corrects the classification randomness through the XGBOOST model, and obtains the accurate probability of muck type;

[0124] Muck improvement parameter prediction module: The improved YOLOV8 network is used to classify the muck images. The XGBOOST model analyzes the feature importance in the muck image classification, adjusts the model output probability to correct the classification randomness, and obtains the accurate muck type probability;

[0125] Prediction parameter module: Use LSTM network to input sensor parameter time series during excavation, combine with the probability of soil type, and combine linear regression and genetic algorithm to predict soil improvement parameters required in future excavation process;

[0126] Parameter optimization module: Particle swarm optimization (PSO) algorithm is used to solve the optimal solution of bentonite flow, foam liquid flow and foam air flow with soil type probability and excavation stability as the objective function;

[0127] Real-time adjustment module: Feedback the parameters optimized by the parameter optimization module to the shield machine control system to achieve real-time monitoring and dynamic adjustment during the excavation process.

[0128] The present invention also discloses an electronic device, comprising:

[0129] one or more processors;

[0130] A memory for storing one or more programs;

[0131] When the one or more programs are executed by the one or more processors, the one or more processors implement the intelligent shield slag improvement method based on machine learning.

[0132] And a computer-readable storage medium having computer instructions stored thereon, characterized in that when the instructions are executed by a processor, the steps of the intelligent shield slag improvement method based on machine learning are implemented.

[0133] It should be noted that, in this application, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0134] The above description is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. An intelligent shield slag improvement method based on machine learning, characterized in that: The following steps are involved: (1) Collecting images of soil and related sensor parameters during shield tunneling; (2) The improved YOLOV8 network is used to classify the muck images. The XGBOOST model analyzes the feature importance in the muck image classification and adjusts the model output probability to correct the classification randomness and obtain the accurate muck type probability; (3) Using the LSTM network to input the sensor parameter time series during the excavation process, combined with the probability of soil type, and combined with linear regression and genetic algorithm, the soil improvement parameters required in the future excavation process are predicted; (4) Using the particle swarm optimization (PSO) algorithm, with the probability of soil type and tunneling stability as the objective function, the optimal solution of bentonite flow, foam liquid flow, and foam air flow is solved; (5) Feeding back the optimized parameters in step (4) to the shield machine control system to achieve real-time monitoring and dynamic adjustment during the tunneling process.

2. The method according to claim 1, characterized in that The YOLOV8 network is based on an improved VanillaNet structure, which improves the accuracy of soil slag classification to more than 98% through feature extraction, enhancement and classification prediction.

3. The method according to claim 1, characterized in that The output calculation formula of YOLOV8 in step (2) is: P cls (i)=σ(W cls ·f i )+b cls Among them, P cls (i) is the classification probability of the i-th class, W cls is the classification weight, f i is the image feature vector, σ is the sigmoid activation function, b cls is the bias term.

4. The method according to claim 1, characterized in that: The correction process of the XGBOOST model in step (2) uses the following expression: Among them, P XGBoost (i) is the corrected classification probability of the i-th class, α k is the weight of each feature, f k (x) is the contribution of the kth feature on sample x, and N is the total number of features.

5. The method according to claim 1, characterized in that The formula of the LSTM network in step (3) is: f t =σ(W f ·[h t-1 ,x t ]+b f ); i t =σ(W i ·[h t-1 ,x t ]+b i ); the t =σ(W o ·[h t-1 ,x t ]+b o ); h t =o t ·tanh(C t ); Among them, f t 、i t , are the forget gate, input gate and candidate memory state respectively, C t is the memory unit state, h t is the output at the current moment, h t-1 is the hidden state of the previous time step, x t is the input of the current time step, σ is the Sigmoid activation function, W f ,b f are the weight and bias of the forget gate, W i ,b i are the weight and bias of the input gate, W C ,b C are the weight and bias of the candidate memory state, W o ,b o are the weight and bias of the output gate respectively; The calculation formula of the linear regression is: Among them, w is the weight of linear regression, b is the bias term, and X scaled is the standardized feature column, is the target column; The fitness function of the genetic algorithm is: Among them, individual is the individual in the genetic algorithm, representing an eigenvalue vector; w and b are the weight and bias terms of the linear regression model; It is the target value predicted by the current individual (the probability of slag after treatment); penalty is the penalty term, which is used to deal with the penalty when the characteristic value exceeds the predetermined range.

6. The method according to claim 1, characterized in that The fitness calculation formula of the PSO algorithm in step (4) is: Among them, Z is the value of the fitness function, that is, the value of the objective function; c0 is the bias term; c i is the coefficient of the first-order term, which represents the coefficient of bentonite flow, foam liquid flow and foam air flow; c ij is the coefficient of the quadratic term, which means x i and x j The cross term of x i is the i-th component of the particle position vector, corresponding to the bentonite flow rate, foam liquid flow rate, or foam air flow rate.

7. An intelligent shield slag improvement system based on machine learning, characterized in that: Includes the following modules: Acquisition parameter module: collects the muck images and related sensor parameters during the shield tunneling process; muck image real-time classification module: uses the improved YOLOV8 network to classify the muck images, corrects the classification randomness through the XGBOOST model, and obtains the accurate probability of muck type; Muck improvement parameter prediction module: The improved YOLOV8 network is used to classify the muck images. The XGBOOST model analyzes the feature importance in the muck image classification, adjusts the model output probability to correct the classification randomness, and obtains the accurate muck type probability; Prediction parameter module: Use LSTM network to input sensor parameter time series during excavation, combine with the probability of soil type, and combine linear regression and genetic algorithm to predict soil improvement parameters required in future excavation process; Parameter optimization module: Particle swarm optimization (PSO) algorithm is used to solve the optimal solution of bentonite flow, foam liquid flow and foam air flow with soil type probability and excavation stability as the objective function; Real-time adjustment module: Feedback the parameters optimized by the parameter optimization module to the shield machine control system to achieve real-time monitoring and dynamic adjustment during the excavation process.

8. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images