Intelligent decision-making method for TBM operating parameters based on dynamic identification of formation information

Abstract

The present invention discloses an intelligent decision-making method for TBM operating parameters based on dynamic identification of stratum information. The method involves obtaining rock-machine parameters during TBM excavation, preprocessing the parameter information and constructing a raw database, establishing three excavation conditions, and establishing a multi-output prediction model for rock mass parameters and a multi-output prediction model for excavation parameters based on a table-based deep learning architecture. A multi-objective optimization algorithm is then constructed, taking into account excavation efficiency, construction cost, equipment safety, and engineering experience. The rock mass parameters are predicted using the multi-output prediction model, the excavation condition is identified, and the excavation parameter multi-output prediction model is used as a fitting function. The method optimizes the TBM excavation state according to the actual excavation condition requirements and generates decision values ​​for operating parameters. The method considers not only excavation rate and tool consumption, but also equipment safety and engineering experience. The resulting decision values ​​for operating parameters can effectively improve construction efficiency, reduce tool costs, and ensure construction safety.

Description

Technical Field

[0001] The present invention belongs to the technical field of tunnel construction and relates to an intelligent decision-making method for TBM operating parameters based on dynamic identification of stratum information. Background Art

[0002] Tunnel boring machines (TBMs) are a key construction method for underground infrastructure, widely used in projects such as water supply systems, urban subway tunnels, and energy infrastructure. During underground construction, the geological environment is often highly complex and uncertain. When operating parameters are poorly matched to the geological environment, tunneling efficiency can be low, leading to machine jams, abnormal tool wear, and increased construction costs.

[0003] Currently, construction personnel primarily rely on experience to determine the geological conditions of the excavation section and the progress of the TBM. They then repeatedly adjust operating parameters to achieve a relatively stable TBM operating state. This is time-consuming and requires extensive experience, and it can be difficult to achieve optimal excavation performance under complex geological conditions. Therefore, developing a method that automatically determines the appropriate TBM operating parameter values ​​for different geological conditions is crucial for ensuring safe and efficient TBM excavation.

[0004] Most current operational parameter decision-making methods are based on geological information. In actual projects, this information must be obtained through coring experiments, which consumes considerable time and effort and hinders timely and effective construction guidance. Furthermore, existing decision-making methods often consider only efficiency and cost, while actual projects also require attention to equipment safety under varying operating conditions and respect for the engineering experience of construction personnel. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent decision-making method for TBM operating parameters based on dynamic identification of stratum information, which solves the problem that the existing TBM operating parameter decision-making method is time-consuming and difficult to guide construction in a timely and effective manner.

[0006] The technical solution adopted by the present invention is a method for intelligent decision-making of TBM operating parameters based on dynamic identification of formation information, comprising the following steps:

[0007] Step 1: Obtain the operating parameters, passive parameters, and rock mass parameters of the TBM excavation process, preprocess the parameter information to construct an original database, and divide the original database into a training set and a test set in a ratio of 8:2. Three excavation conditions are established: soft rock formation, balanced formation, and hard rock formation;

[0008] Step 2: Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters and a multi-output prediction model for tunneling parameters were constructed using the training set. These models were used to identify stratum information and characterize mapping relationships, respectively. The performance of these models was verified using the test set.

[0009] Step 3: Construct a multi-objective optimization algorithm to take into account tunneling efficiency, construction cost, equipment safety, and engineering experience. Use the multi-output prediction model for rock mass parameters to predict rock mass parameters and then identify tunneling conditions. Use the multi-output prediction model for tunneling parameters as the fitting function to optimize the TBM tunneling state and provide decision values ​​for operating parameters based on the actual tunneling conditions.

[0010] The operating parameters include cutterhead speed, cutterhead torque and cutterhead thrust; the passive parameters include support shoe force, top shield cylinder left pressure, top shield cylinder right pressure, side shield cylinder left pressure and side shield cylinder right pressure; the rock mass parameters include rock mass uniaxial compressive strength and rock mass quality indicators.

[0011] The parameter information is preprocessed to construct the database, including using the isolation forest algorithm to identify outliers in the parameter information, then using linear interpolation to replace the outliers, and using the Min-Max method to normalize the data. The calculation formula is as follows;

[0012]

[0013] in, x 、 x * is the characteristic value before and after normalization, min ( x ), max ( x ) are the minimum and maximum values ​​in the feature, respectively.

[0014] Three excavation working conditions are established according to the uniaxial compressive strength of the rock mass, namely, soft rock formations not exceeding 30MPa, balanced formations between 30MPa and 90MPa, and hard rock formations exceeding 90MPa.

[0015] In step 2, based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters is constructed with operating parameters and passive parameters as input and rock mass uniaxial compressive strength and rock mass quality indicators as output. During the TBM tunneling process, operating parameters and passive parameters are directly generated. These parameters serve as input to the multi-output prediction model for rock mass parameters, and the excavation stratum information of the tunnel ring can be identified.

[0016] In step 2, based on the tabular deep learning architecture TabNet, a multi-output prediction model for excavation parameters is constructed with operating parameters and rock parameters as input and excavation rate and tool consumption per linear meter as output. During the excavation process, the newly collected parameter information is dynamically added to the database, and the multi-output prediction model for rock parameters and the multi-output prediction model for excavation parameters are continuously trained to enhance model performance.

[0017] In step 3, by establishing the torque safety factor K T and thrust safety factor K TF Realize equipment safety considerations, namely, to limit excessive torque and excessive thrust in the decision-making process;

[0018]

[0019] in, Actual torque is the actual cutter head torque value of the TBM equipment, TBM maximum torque The maximum cutter head torque allowed for TBM equipment;

[0020]

[0021] in, Actual thust force is the actual cutterhead thrust value of the TBM equipment, TBM maximum thust force The maximum cutterhead thrust allowed for TBM equipment.

[0022] In step 3, by establishing the engineering experience coefficient K e Taking engineering experience into account, i.e. limiting the adjustment range of operating parameters during the decision-making process;

[0023]

[0024] in, T original is the initial value of the cutterhead thrust, T optimized is the decision value of the cutterhead thrust, TF original is the initial value of the cutter head torque, TF optimized is the decision value of the cutter head torque, RPM original is the initial value of the cutter head speed, RPM optimized is the decision value of the cutter head speed.

[0025] In step 3, the Black Kite algorithm is improved into a multi-objective optimization algorithm using a weighted Chebyshev scalarization strategy. The initial population generation strategy of the Black Kite algorithm is then replaced with the Tent chaos mapping strategy to construct a multi-objective Black Kite optimization algorithm. The operating parameters and passive parameters directly generated during the TBM tunneling process are used as inputs to the multi-output prediction model for rock mass parameters to obtain the actual rock mass parameters. The tunneling conditions are identified based on the uniaxial compressive strength of the rock mass in the actual rock mass parameters, and an optimization scheme is then determined. During the optimization process, target values ​​and weights are set according to project requirements, while the rock parameters remain unchanged. The multi-output prediction model for tunneling parameters is used as the fitting function, and the operating parameters and actual rock mass parameters directly generated during the TBM tunneling process are used as inputs to obtain the tunneling rate and tool consumption per linear meter. Weights and target values ​​for the tunneling rate and tool consumption per linear meter are then set based on the excavation plan and construction budget. Taking tunneling efficiency, construction cost, equipment safety, and engineering experience into consideration, different optimization schemes are formulated for different working conditions, ultimately providing corresponding decision values ​​for cutterhead speed, cutterhead torque, and cutterhead thrust.

[0026] The parameter information collected by the model update module is processed using the isolation forest algorithm to establish a database, build and continuously update the model performance, and the operating parameters and passive parameters of the current excavation section are used as input through the excavation stratum information identification module. The rock parameters of the current excavation section are predicted through a tabular deep learning model to realize the identification of the excavation working condition. The multi-objective black kite optimization algorithm is used through the operation parameter optimization decision module, and the multi-objective prediction model of excavation parameters is used as the mapping relationship. The decision values ​​of the cutterhead speed, cutterhead torque and cutterhead thrust are given according to the actual working conditions' requirements for excavation rate and tool consumption.

[0027] The beneficial effects of the present invention are as follows:

[0028] (1) By constructing a multi-output prediction model for rock mass parameters, the parameters obtained in real time by TBM were used to accurately predict the uniaxial compressive strength and rock mass quality indicators, avoiding the coring and testing process and effectively improving construction efficiency.

[0029] (2) The Black Kite Algorithm (BKA) is improved by using the weighted Chebyshev scalarization strategy and the Tent chaos mapping strategy, and a multi-objective Black Kite Optimization Algorithm (MOBK) is constructed. Compared with the multi-objective particle swarm optimization algorithm and the fast elite multi-objective genetic algorithm, the MOBK algorithm has better optimization performance and is more adaptable to optimization problems under multi-constraints.

[0030] (3) The tabular deep learning architecture is applied to the TBM field, which achieves multi-target prediction more accurately than existing machine learning methods such as random forest, support vector regression, and feedforward neural network;

[0031] (4) The present invention not only takes into account the two key performance indicators of TBM, namely, excavation rate and tool consumption, but also considers equipment safety and engineering experience. It achieves the purpose of improving efficiency and reducing costs with different optimization schemes under different working conditions, and provides corresponding operating parameter decision values, which has guiding significance for tunnel construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 1 is a flow chart of an intelligent decision-making method for TBM operating parameters based on dynamic identification of formation information according to the present invention;

[0033] Figure 2 The prediction result of the uniaxial compressive strength of the rock mass by the rock mass parameter multi-output prediction model in Example 6 of the present invention;

[0034] Figure 3 The prediction results of the rock mass quality index by the rock mass parameter multi-output prediction model in Example 6 of the present invention;

[0035] Figure 4 The prediction result of the tunneling rate by the tunneling parameter multi-output prediction model in Example 6 of the present invention;

[0036] Figure 5 The prediction result of the tunneling parameter multi-output prediction model for tool consumption per linear meter in Example 6 of the present invention;

[0037] Figure 6 This is the optimization result of the excavation rate in soft rock formation in Example 6 of the present invention;

[0038] Figure 7 This is the optimization result of the tool consumption per linear meter in the soft rock formation in Example 6 of the present invention;

[0039] Figure 8 The decision result of the cutterhead speed for soft rock formations in Example 6 of the present invention;

[0040] Figure 9 is the decision result of the cutterhead torque for soft rock formations in Example 6 of the present invention;

[0041] Figure 10 is the decision result of the cutterhead thrust in the soft rock formation in Example 6 of the present invention;

[0042] Figure 11 This is the optimization result of the balanced stratum excavation rate in Example 6 of the present invention;

[0043] Figure 12 This is the optimization result of the tool consumption per linear meter of the balanced formation in Example 6 of the present invention;

[0044] Figure 13 The decision result of the cutterhead rotation speed for balancing the formation in Example 6 of the present invention;

[0045] Figure 14 is the decision result of balancing the cutter head torque in Example 6 of the present invention;

[0046] Figure 15 is the decision result of balancing the thrust of the cutterhead in the formation in Example 6 of the present invention;

[0047] Figure 16 This is the optimization result of the hard rock formation excavation rate in Example 6 of the present invention;

[0048] Figure 17 This is the optimization result of the tool consumption per linear meter of the hard rock formation in Example 6 of the present invention;

[0049] Figure 18 is the decision result of the cutterhead rotation speed for hard rock formations in Example 6 of the present invention;

[0050] Figure 19 is the decision result of the cutterhead torque for hard rock formations in Example 6 of the present invention;

[0051] Figure 20 This is the decision result of the cutterhead thrust in the hard rock formation in Example 6 of the present invention. DETAILED DESCRIPTION

[0052] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0053] Example 1

[0054] Intelligent decision-making method for TBM operating parameters based on dynamic identification of formation information, see Figure 1 , including the following steps:

[0055] Step 1: Obtain operating parameters, passive parameters, and rock mass parameters during the TBM excavation process. Operating parameters include cutterhead speed, cutterhead torque, and cutterhead thrust. Passive parameters include gripper support force, top shield cylinder left pressure, top shield cylinder right pressure, side shield cylinder left pressure, and side shield cylinder right pressure. Rock mass parameters include uniaxial compressive strength and rock mass quality indicators.

[0056] The acquired parameter information is pre-processed to construct an original database, including using the isolation forest algorithm to quickly and accurately process outliers in the parameter information. Three tunneling conditions are established based on the uniaxial compressive strength of the rock mass: soft rock formations with a strength of no more than 30 MPa, balanced formations between 30 MPa and 90 MPa, and hard rock formations with a strength of more than 90 MPa.

[0057] The original database is divided into training set and test set in a ratio of 8:2;

[0058] Step 2: Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters and a multi-output prediction model for tunneling parameters were constructed using the training set. These models were used to identify stratum information and characterize mapping relationships, respectively. The performance of these models was verified using the test set.

[0059] Step 3: Based on considerations of tunneling efficiency, construction costs, equipment safety, and engineering experience, a multi-objective optimization algorithm is constructed. The rock mass parameters are predicted using a multi-output prediction model for rock mass parameters to identify the working conditions. The multi-output prediction model for tunneling parameters is used as a fitting function to optimize the TBM tunneling state and determine the operating parameter decision values ​​based on the actual working conditions, thereby achieving the highest tunneling rate while minimizing tool consumption per meter.

[0060] Example 2

[0061] Step 1: Obtain operating parameters, passive parameters, and rock mass parameters during the TBM excavation process. Operating parameters include cutterhead speed, cutterhead torque, and cutterhead thrust. Passive parameters include gripper support force, top shield cylinder left pressure, top shield cylinder right pressure, side shield cylinder left pressure, and side shield cylinder right pressure. Rock mass parameters include uniaxial compressive strength and rock mass quality indicators.

[0062] The acquired parameter information is preprocessed to construct the original database, including using the isolation forest algorithm to quickly identify outliers in the parameter information, then using linear interpolation to replace outliers, and using the Min-Max method to normalize the data. The calculation formula is as follows;

[0063]

[0064] in, x 、 x * is the characteristic value before and after normalization, min ( x ), max ( x ) are the minimum and maximum values ​​in the feature, respectively.

[0065] The original database is divided into training set and test set in a ratio of 8:2;

[0066] Three tunneling conditions were established based on the uniaxial compressive strength of the rock mass: soft rock formations with a compressive strength not exceeding 30 MPa, balanced formations between 30 MPa and 90 MPa, and hard rock formations with a compressive strength exceeding 90 MPa.

[0067] Step 2: Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters and a multi-output prediction model for tunneling parameters are constructed using the training set data. These models are used for identifying stratum information and representing mapping relationships, respectively.

[0068] Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters is constructed, taking operating parameters and passive parameters as input and rock mass uniaxial compressive strength and rock mass quality indicators as output. Operating parameters and passive parameters are directly generated during the TBM excavation of a tunnel ring. These parameters serve as input to the multi-output prediction model for rock mass parameters, and can identify the excavation stratum information of the tunnel ring.

[0069] Based on the tabular deep learning architecture TabNet, a multi-output prediction model for tunneling parameters is constructed with operating parameters and rock mass parameters as input and tunneling rate and tool consumption per linear meter as output.

[0070] Use the test set to verify the performance of the rock mass parameter multi-output prediction model and the tunneling parameter multi-output prediction model;

[0071] During the excavation process, newly collected parameters are dynamically added to the database, and the multi-output prediction model of rock mass parameters and the multi-output prediction model of excavation parameters are continuously trained to enhance model performance.

[0072] Step 3: Based on considerations of tunneling efficiency, construction costs, equipment safety, and engineering experience, a multi-objective optimization algorithm is constructed. The rock mass parameters are predicted using a multi-output prediction model for rock mass parameters to identify the working conditions. The multi-output prediction model for tunneling parameters is used as a fitting function to optimize the TBM tunneling state and determine the operating parameter decision values ​​based on the actual working conditions, thereby achieving the highest tunneling rate while minimizing tool consumption per meter.

[0073] Example 3

[0074] Step 1: Obtain operating parameters, passive parameters, and rock mass parameters during the TBM excavation process. Operating parameters include cutterhead speed, cutterhead torque, and cutterhead thrust. Passive parameters include gripper support force, top shield cylinder left pressure, top shield cylinder right pressure, side shield cylinder left pressure, and side shield cylinder right pressure. Rock mass parameters include uniaxial compressive strength and rock mass quality indicators.

[0075] The acquired parameter information is preprocessed to construct the original database, including using the isolation forest algorithm to quickly identify outliers in the parameter information, then using linear interpolation to replace outliers, and using the Min-Max method for data normalization;

[0076] The original database is divided into training set and test set in a ratio of 8:2;

[0077] Three tunneling conditions were established based on the uniaxial compressive strength of the rock mass: soft rock formations with a compressive strength not exceeding 30 MPa, balanced formations between 30 MPa and 90 MPa, and hard rock formations with a compressive strength exceeding 90 MPa.

[0078] Step 2: Based on the tabular deep learning architecture TabNet, the training set is used to construct a multi-output prediction model for rock mass parameters and a multi-output prediction model for tunneling parameters, which are used for stratum information identification and characterization mapping relationships, respectively.

[0079] Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters is constructed, taking operating parameters and passive parameters as input and rock mass uniaxial compressive strength and rock mass quality indicators as output. Operating parameters and passive parameters are directly generated during the TBM excavation of a tunnel ring. These parameters serve as input to the multi-output prediction model for rock mass parameters, and can identify the excavation stratum information of the tunnel ring.

[0080] Based on the tabular deep learning architecture TabNet, a multi-output prediction model for tunneling parameters is constructed with operating parameters and rock mass parameters as input and tunneling rate and tool consumption per linear meter as output.

[0081] Use the test set to verify the performance of the rock mass parameter multi-output prediction model and the tunneling parameter multi-output prediction model;

[0082] During the excavation process, newly collected parameters are dynamically added to the database, and the multi-output prediction model of rock mass parameters and the multi-output prediction model of excavation parameters are continuously trained to enhance model performance.

[0083] In step 3, a multi-objective optimization algorithm is constructed based on considerations of tunneling efficiency, construction costs, equipment safety, and engineering experience. The rock mass parameters are predicted using a multi-output prediction model for rock mass parameters to identify the working conditions. The tunneling parameter multi-output prediction model is used as a fitting function to optimize the TBM tunneling state and provide decision values ​​for the operating parameters according to the actual working conditions.

[0084] By establishing the torque safety factor K T and thrust safety factor K TF Realize equipment safety considerations, namely, to limit excessive torque and excessive thrust in the decision-making process;

[0085]

[0086] in, Actual torque is the actual cutter head torque value of the TBM equipment, TBM maximum torque The maximum cutter head torque allowed for TBM equipment;

[0087]

[0088] in, Actual thust force is the actual cutterhead thrust value of the TBM equipment, TBM maximum thust force The maximum cutterhead thrust allowed for TBM equipment.

[0089] By establishing the engineering experience coefficient Ke Taking engineering experience into account, i.e. limiting the adjustment range of operating parameters during the decision-making process;

[0090]

[0091] in, T original is the initial value of the cutterhead thrust, T optimized is the decision value of the cutterhead thrust, TF original is the initial value of the cutter head torque, TF optimized is the decision value of the cutter head torque, RPM original is the initial value of the cutter head speed, RPM optimized is the decision value of the cutter head speed.

[0092] Example 4

[0093] Step 1: Obtain the operating parameters, passive parameters, and rock mass parameters of the TBM excavation process. The operating parameters include the cutterhead speed, cutterhead torque, and cutterhead thrust. The passive parameters include the support shoe force, the left pressure of the top shield cylinder, the right pressure of the top shield cylinder, the left pressure of the side shield cylinder, and the right pressure of the side shield cylinder. The rock mass parameters include the uniaxial compressive strength of the rock mass and rock mass quality indicators.

[0094] The acquired parameter information is preprocessed to construct the original database, including using the isolation forest algorithm to identify outliers in the parameter information, then using linear interpolation to replace the outliers, and using the Min-Max method to normalize the data. The calculation formula is as follows;

[0095]

[0096] in, x 、 x * is the characteristic value before and after normalization, min ( x ), max ( x ) are the minimum and maximum values ​​in the feature, respectively.

[0097] The original database is divided into training set and test set in a ratio of 8:2;

[0098] Three tunneling conditions were established based on the uniaxial compressive strength of the rock mass: soft rock formations with a compressive strength not exceeding 30 MPa, balanced formations between 30 MPa and 90 MPa, and hard rock formations with a compressive strength exceeding 90 MPa.

[0099] Step 2: Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters and a multi-output prediction model for tunneling parameters were constructed using the training set. These models were used to identify stratum information and characterize mapping relationships, respectively. The performance of these models was verified using the test set.

[0100] Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters is constructed, taking operating parameters and passive parameters as input and rock mass uniaxial compressive strength and rock mass quality indicators as output. Operating parameters and passive parameters are directly generated during the TBM excavation of a tunnel ring. These parameters serve as input to the multi-output prediction model for rock mass parameters, and can identify the excavation stratum information of the tunnel ring.

[0101] Based on the tabular deep learning architecture TabNet, a multi-output prediction model for tunneling parameters is constructed with operating parameters and rock mass parameters as input and tunneling rate and tool consumption per linear meter as output.

[0102] During the excavation process, newly collected parameters are dynamically added to the database to continuously enhance the model performance.

[0103] In step 3, a multi-objective optimization algorithm is constructed based on considerations of tunneling efficiency, construction costs, equipment safety, and engineering experience. The rock mass parameters are predicted using a multi-output prediction model for rock mass parameters to identify the working conditions. The tunneling parameter multi-output prediction model is used as a fitting function to optimize the TBM tunneling state and provide decision values ​​for the operating parameters according to the actual working conditions.

[0104] By establishing the torque safety factor K T and thrust safety factor K TF Realize equipment safety considerations, namely, to limit excessive torque and excessive thrust in the decision-making process;

[0105]

[0106] in, Actual torque is the actual cutter head torque value of the TBM equipment, TBM maximum torque The maximum cutter head torque allowed for TBM equipment;

[0107]

[0108] in, Actual thust force is the actual cutterhead thrust value of the TBM equipment, TBM maximum thust force The maximum cutterhead thrust allowed for TBM equipment.

[0109] By establishing the engineering experience coefficient K e Taking engineering experience into account, i.e. limiting the adjustment range of operating parameters during the decision-making process;

[0110]

[0111] in, T original is the initial value of the cutterhead thrust, T optimized is the decision value of the cutterhead thrust, TF original is the initial value of the cutter head torque, TF optimized is the decision value of the cutter head torque, RPM original is the initial value of the cutter head speed, RPM optimized is the decision value of the cutter head speed.

[0112] The Black Kite algorithm was improved into a multi-objective optimization algorithm using a weighted Chebyshev scalarization strategy. The initial population generation strategy of the Black Kite algorithm was then replaced with a Tent chaos mapping strategy to construct a multi-objective Black Kite optimization algorithm. The operational and passive parameters directly generated during the TBM tunneling process were used as inputs to a multi-output prediction model for rock mass parameters to obtain the actual rock mass parameters. The tunneling conditions were identified based on the uniaxial compressive strength of the actual rock mass parameters, and an optimization scheme was determined. During the optimization process, target values ​​and weights were set according to project requirements, while the rock parameters remained unchanged. The multi-output prediction model for tunneling parameters was used as a fitting function, and the operational and actual rock mass parameters directly generated during the TBM tunneling process were used as inputs to obtain the tunneling rate and tool consumption per linear meter. Weights and target values ​​for the tunneling rate and tool consumption per linear meter were then set based on the excavation plan and construction budget. Taking tunneling efficiency, construction cost, equipment safety, and engineering experience into consideration, different optimization schemes were developed for different working conditions, ultimately providing corresponding decision values ​​for cutterhead speed, cutterhead torque, and cutterhead thrust.

[0113] Example 5

[0114] Step 1: Obtain operating parameters, passive parameters, and rock mass parameters during the TBM excavation process. Operating parameters include cutterhead speed, cutterhead torque, and cutterhead thrust. Passive parameters include gripper support force, top shield cylinder left pressure, top shield cylinder right pressure, side shield cylinder left pressure, and side shield cylinder right pressure. Rock mass parameters include uniaxial compressive strength and rock mass quality indicators.

[0115] The acquired parameter information is preprocessed to construct the original database, including using the isolation forest algorithm to quickly identify outliers in the parameter information, then using linear interpolation to replace outliers, and using the Min-Max method to normalize the data. The calculation formula is as follows;

[0116]

[0117] in, x 、 x * is the characteristic value before and after normalization, min ( x ), max ( x ) are the minimum and maximum values ​​in the feature, respectively.

[0118] The original database is divided into training set and test set in a ratio of 8:2;

[0119] Three excavation conditions are established based on the uniaxial compressive strength of the rock mass: soft rock formations with a compressive strength not exceeding 30 MPa, balanced formations between 30 MPa and 90 MPa, and hard rock formations with a compressive strength exceeding 90 MPa.

[0120] Step 2: Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters and a multi-output prediction model for tunneling parameters are constructed using the training set data. These models are used for identifying stratum information and representing mapping relationships, respectively.

[0121] Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters is constructed, taking operating parameters and passive parameters as input and rock mass uniaxial compressive strength and rock mass quality indicators as output. Operating parameters and passive parameters are directly generated during the TBM excavation of a tunnel ring. These parameters serve as input to the multi-output prediction model for rock mass parameters, and can identify the excavation stratum information of the tunnel ring.

[0122] Based on the tabular deep learning architecture TabNet, a multi-output prediction model for tunneling parameters is constructed with operating parameters and rock mass parameters as input and tunneling rate and tool consumption per linear meter as output. During the tunneling process, newly collected parameters are dynamically added to the database to continuously enhance model performance.

[0123] Use the test set to verify the performance of the rock mass parameter multi-output prediction model and the tunneling parameter multi-output prediction model;

[0124] During the tunneling process, newly collected parameter information is dynamically added to the database, and the rock mass parameter multi-output prediction model and the tunneling parameter multi-output prediction model are continuously trained to enhance model performance;

[0125] Step 3: Based on considerations of tunneling efficiency, construction costs, equipment safety, and engineering experience, a multi-objective optimization algorithm is constructed to minimize the adjustment of operating parameters, thereby improving tunneling efficiency and reducing construction costs while ensuring equipment safety. The specific process is as follows:

[0126] Step 3.1: Improve the Black Kite algorithm into a multi-objective optimization algorithm through the weighted Chebyshev scalarization strategy, and then use the Tent chaotic mapping strategy to replace the initial population generation strategy of the Black Kite algorithm to construct a multi-objective Black Kite optimization algorithm.

[0127] Step 3.2: Predict rock mass parameters using a multi-output rock mass parameter prediction model to identify the operating condition. This involves using the operational and passive parameters directly generated during the TBM tunneling process as inputs to the multi-output rock mass parameter prediction model to obtain actual rock mass parameters. The tunneling condition is then identified based on the rock mass uniaxial compressive strength, which is part of the actual rock mass parameters.

[0128] In step 3.3, during the optimization process, the rock parameters are kept constant. The multi-output prediction model for tunneling parameters is used as the fitting function, and the operating parameters and actual rock mass parameters directly generated during the TBM tunneling process are used as inputs. The TBM tunneling state is optimized by adjusting the decision variables to determine the final value of the decision variable. This is achieved by comparing the actual tunneling rate and tool consumption per linear meter output by the multi-output prediction model with the target tunneling rate and tool consumption per linear meter required by the project. Taking into account tunneling efficiency, construction costs, equipment safety, and engineering experience, different optimization schemes are formulated for different working conditions. This minimizes the adjustment of operating parameters, improves tunneling efficiency, and reduces construction costs while ensuring equipment safety. Ultimately, the corresponding decision values ​​for cutterhead speed, cutterhead torque, and cutterhead thrust are determined.

[0129] By establishing the torque safety factor K T and thrust safety factor K TF Realize equipment safety considerations, namely, to limit excessive torque and excessive thrust in the decision-making process;

[0130]

[0131] in, Actual torque is the actual cutter head torque value of the TBM equipment, TBM maximum torque The maximum cutter head torque allowed for TBM equipment;

[0132]

[0133] in, Actual thust force is the actual cutterhead thrust value of the TBM equipment, TBM maximum thust force The maximum cutterhead thrust allowed for TBM equipment.

[0134] By establishing the engineering experience coefficient K e Taking engineering experience into account, i.e. limiting the adjustment range of operating parameters during the decision-making process;

[0135]

[0136] in, T original is the initial value of the cutterhead thrust, T optimized is the decision value of the cutterhead thrust, TF original is the initial value of the cutter head torque, TF optimized is the decision value of the cutter head torque, RPM original is the initial value of the cutter head speed, RPM optimized is the decision value of the cutter head speed.

[0137] Develop corresponding optimization plans for different working conditions:

[0138] (1) For soft rock formations, the TBM torque is likely to approach the maximum value allowed by the equipment. It is necessary to limit the increase of TBM torque and ensure that the adjustment of operating parameters is not too large. The optimization solution at this time is to increase K T , reduce K e , increase the excavation rate and reduce tool consumption. K T and K e The target values ​​are 1 and 0, respectively. Set a target value for the advance rate based on the excavation plan, and a target value for tool consumption per linear meter based on the construction budget. The weights are generally the same, i.e., 1 / 4. If a particular target requires special attention during construction, increase its weight.

[0139] (2) For balanced formations, the TBM thrust and torque change amplitudes are similar and moderate, and the equipment as a whole is safe, so K is not considered. T and K TF , but to ensure that the adjustment of operating parameters is not too large, the optimization solution at this time is to reduce K e , increase the excavation rate and reduce tool consumption. K e Set the target value to 0. Set a target value for the advance rate based on the excavation plan, and a target value for tool consumption per linear meter based on the construction budget. These weights are generally equal, i.e., 1 / 3. If a specific target requires special attention during construction, increase its weight.

[0140] (3) For hard rock formations, the TBM thrust is likely to approach the maximum value allowed by the equipment. It is necessary to limit the increase in TBM thrust while also ensuring that the adjustment of operating parameters is not too large. The optimization solution at this time is to increase K TF , reduce K e , increase the excavation rate and reduce tool consumption. K TF and K eThe target values ​​are 1 and 0, respectively. Set a target value for the advance rate based on the excavation plan, and a target value for tool consumption per linear meter based on the construction budget. The weights for these are generally the same, i.e., 1 / 4. If a particular target requires special attention during construction, increase its weight.

[0141] The present invention collects and obtains parameter information through a model updating module, processes the parameter information using an isolation forest algorithm, establishes a database, builds and continuously updates the performance of the model, uses the operating parameters and passive parameters of the current excavation section as input through an excavation stratum information identification module, predicts the rock parameters of the current excavation section through a table deep learning model, and realizes the identification of the excavation working condition, uses a multi-objective black kite optimization algorithm through an operating parameter optimization decision module, and uses a multi-objective prediction model of excavation parameters as a mapping relationship, and gives decision values ​​for cutterhead speed, cutterhead torque and cutterhead thrust according to the actual working conditions' requirements for excavation rate and tool consumption.

[0142] Example 6

[0143] Step 1: Obtain the operating parameters, passive parameters, and rock mass parameters recorded during the TBM excavation process and preprocess these parameter information to construct an original database. The specific steps are as follows:

[0144] Step 1.1: Segment the parameters recorded during the TBM excavation process into each individual excavation ring. Operational parameters include cutterhead speed, cutterhead torque, and cutterhead thrust. Passive parameters include gripper support force, top shield cylinder left pressure, top shield cylinder right pressure, side shield cylinder left pressure, and side shield cylinder right pressure. Rock mass parameters include uniaxial compressive strength and rock mass quality indicators.

[0145] In step 1.2, the obtained parameter information is preprocessed to construct the original database. The isolation forest algorithm is used to quickly identify outliers in the parameter information. Then, linear interpolation is used to replace the outliers, and the Min-Max method is used to normalize the data. The calculation formula is as follows;

[0146]

[0147] in, x 、 x * is the characteristic value before and after normalization, min ( x ), max ( x ) are the minimum and maximum values ​​in the feature, respectively.

[0148] Step 1.3: Divide the original database into training set and test set in a ratio of 8:2;

[0149] Step 1.4: Establish three tunneling conditions based on the uniaxial compressive strength of the rock mass: soft rock formations with a compressive strength not exceeding 30 MPa, balanced formations with a compressive strength between 30 MPa and 90 MPa, and hard rock formations with a compressive strength exceeding 90 MPa.

[0150] Step 2: Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters and a multi-output prediction model for tunneling parameters are constructed using the training data. These models are used for stratum information identification and characterization mapping relationships, respectively. The details are as follows:

[0151] In step 2.1, based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters is constructed, using operating parameters and passive parameters as input and rock mass uniaxial compressive strength and rock mass quality indicators as output. Operating parameters and passive parameters are directly generated during the TBM tunneling process. These parameters serve as input to the multi-output prediction model for rock mass parameters, and can identify the tunneling stratum information of the tunnel ring, i.e., the tunneling conditions.

[0152] Step 2.2: Based on the tabular deep learning architecture TabNet, a multi-output prediction model for tunneling parameters is constructed with operating parameters and rock mass parameters as inputs and tunneling rate and tool consumption per linear meter as outputs.

[0153] Step 2.3: Use the test set to verify the performance of the rock mass parameter multi-output prediction model and the tunneling parameter multi-output prediction model. In this embodiment, the determination coefficient is used. R 2 , root mean square error RMSE and mean absolute percentage error MAPE To verify the performance of the model, the calculation formula is as follows:

[0154]

[0155]

[0156]

[0157] in, 、 and represent the actual value, predicted value and average value respectively. Represents the total number of samples.

[0158] R 2 The closer the value is to 1, RMSE and MAPE The closer the value is to zero, the closer the model prediction is to the actual value of the dataset, that is, the more accurate the model prediction is.

[0159] Draw a scatter plot to visualize the performance of the two multi-output prediction models on the test set, see Figure 2-3 , the rock mass parameter multi-output prediction model is effective in predicting the uniaxial compressive strength of rock mass. R 2 , RMSE, MAPE They are 0.9115, 4.7358, and 0.0496 respectively; in terms of rock mass quality index prediction R 2 , RMSE, MAPE They are 0.8986, 5.6703, and 0.0923 respectively. Figure 4-5 , the tunneling parameter multi-output prediction model is effective in predicting tunneling rate. R 2 , RMSE, MAPE They are 0.9598, 2.0560, and 0.0333 respectively; in terms of the predicted tool consumption per meter R 2 , RMSE, MAPE They are 0.9371, 0.0485, and 0.1180 respectively. The test results show that the performance of these two models meets the actual engineering needs.

[0160] Step 2.4: During the tunneling process, the newly collected parameter information is dynamically added to the database, and the rock mass parameter multi-output prediction model and the tunneling parameter multi-output prediction model are continuously trained to enhance the model performance.

[0161] Step 3: Based on considerations of tunneling efficiency, construction costs, equipment safety, and engineering experience, a multi-objective optimization algorithm is constructed. The rock mass parameters are predicted using a multi-output prediction model to identify the working conditions. Using the multi-output prediction model as a fitting function, the TBM tunneling state is optimized based on the actual working conditions and the operating parameter decision values ​​are determined. The specific steps are as follows:

[0162] Step 3.1: Improve the Black Kite algorithm into a multi-objective optimization algorithm by using a weighted Chebyshev scalarization strategy. Then, replace the initial population generation strategy of the Black Kite algorithm with the Tent chaotic mapping strategy to construct a multi-objective Black Kite optimization algorithm. In this embodiment, the initial population size is defined as 100, and the maximum number of iterations is set to 50.

[0163] In step 3.2, the cutterhead speed, torque, and thrust are used as decision variables, and the advance rate and tool consumption per linear meter are used as target variables. The operational parameters and passive parameters directly generated during the TBM tunneling cycle are used as inputs to a multi-output rock parameter prediction model. The actual rock parameters are obtained. Based on the uniaxial compressive strength of the rock mass, which is included in the actual rock parameters, the tunneling condition is identified, and an optimization solution is determined. During the optimization process, target values ​​and weights for the advance rate and tool consumption per linear meter are set based on project requirements (excavation plan and construction budget). The rock parameters are kept constant. The multi-output tunneling parameter prediction model is used as a fitting function, with the operational parameters and actual rock parameters directly generated during the TBM tunneling cycle as inputs. The decision variables are adjusted to optimize the TBM's tunneling state and determine the final values ​​of the decision variables. In this embodiment, equal weights are assigned, and the optimal value in the database is used as the target value.

[0164] In step 3.3, different optimization plans are formulated for different working conditions, taking into account excavation efficiency, construction costs, equipment safety, and engineering experience. This allows for minimal adjustment of operating parameters, thereby improving excavation efficiency and reducing construction costs while ensuring equipment safety. Ultimately, the corresponding decision values ​​for cutterhead speed, cutterhead torque, and cutterhead thrust are determined.

[0165] By establishing the torque safety factor K T and thrust safety factor K TF Realize equipment safety considerations, namely, to limit excessive torque and excessive thrust in the decision-making process;

[0166]

[0167] in, Actual torque is the actual cutter head torque value of the TBM equipment, TBM maximum torque The maximum cutter head torque allowed for TBM equipment;

[0168]

[0169] in, Actual thust force is the actual cutterhead thrust value of the TBM equipment, TBM maximum thust force The maximum cutterhead thrust allowed for TBM equipment.

[0170] By establishing the engineering experience coefficient K e Taking engineering experience into account, i.e. limiting the adjustment range of operating parameters during the decision-making process;

[0171]

[0172] in, T original is the initial value of the cutterhead thrust,T optimized is the decision value of the cutterhead thrust, TF original is the initial value of the cutter head torque, TF optimized is the decision value of the cutter head torque, RPM original is the initial value of the cutter head speed, RPM optimized is the decision value of the cutter head speed.

[0173] Develop corresponding optimization plans for different working conditions:

[0174] (1) For soft rock formations, the TBM torque is likely to approach the maximum value allowed by the equipment. It is necessary to limit the increase of TBM torque and ensure that the adjustment of operating parameters is not too large. The optimization solution at this time is to increase K T , reduce K e , increase the excavation rate and reduce tool consumption. K T and K e The target values ​​are 1 and 0, respectively. Set a target value for the advance rate based on the excavation plan, and a target value for tool consumption per linear meter based on the construction budget. The weights are generally the same, i.e., 1 / 4. If a particular target requires special attention during construction, increase its weight.

[0175] (2) For balanced formations, the TBM thrust and torque change amplitudes are similar and moderate, and the equipment as a whole is safe, so K is not considered. T and K TF , but to ensure that the adjustment of operating parameters is not too large, the optimization solution at this time is to reduce K e , increase the excavation rate and reduce tool consumption. K e Set the target value to 0. Set a target value for the advance rate based on the excavation plan, and a target value for tool consumption per linear meter based on the construction budget. These weights are generally equal, i.e., 1 / 3. If a specific target requires special attention during construction, increase its weight.

[0176] (3) For hard rock formations, the TBM thrust is likely to approach the maximum value allowed by the equipment. It is necessary to limit the increase in TBM thrust while also ensuring that the adjustment of operating parameters is not too large. The optimization solution at this time is to increase K TF , reduce K e , increase the excavation rate and reduce tool consumption. K TF and K e The target values ​​are 1 and 0, respectively. Set a target value for the advance rate based on the excavation plan, and a target value for tool consumption per linear meter based on the construction budget. The weights for these are generally the same, i.e., 1 / 4. If a particular target requires special attention during construction, increase its weight.

[0177] The final results of tunneling parameter optimization and operation parameter decision-making for the three working conditions are as follows: Figure 6-20 As shown in the figure, using soft rock conditions as an example, the average advance rate in soft rock formations increased from 48.89 mm / min to 56.19 mm / min, and the average cutter consumption decreased from 0.30 cutter / m to 0.21 cutter / m, resulting in improvements of 15.37% and 28.58% in advance rate and cutter consumption, respectively. Furthermore, the corresponding decision values ​​for cutterhead speed, cutterhead torque, and cutterhead thrust are given. It can be seen that the soft rock conditions significantly limit the increase in torque, reducing the average torque from 2384.37 kN·m to 2106.78 kN·m. The adjustment range is small, with an average adjustment of 25.56%. This demonstrates that this method can optimize the TBM excavation state through appropriate adjustment schemes, providing decision values ​​for operating parameters and contributing to efficient TBM construction.

[0178] During actual TBM excavation, incremental learning is used to improve the performance of the optimization method. That is, new data is continuously collected during the TBM excavation process and added to the database. The two multi-output prediction models are retrained and the optimization framework is automatically improved to achieve automatic updating of the entire method.

Claims

1. An intelligent decision-making method for TBM operating parameters based on dynamic identification of formation information, characterized by: The following steps are involved: Step 1: Obtain the operating parameters, passive parameters, and rock mass parameters of the TBM excavation process, preprocess the parameter information to construct an original database, and divide the original database into a training set and a test set in a ratio of 8:

2. Three excavation conditions are established: soft rock formation, balanced formation, and hard rock formation; Step 2: Based on the tabular deep learning architecture TabNet, a multi-output prediction model for rock mass parameters and a multi-output prediction model for tunneling parameters were constructed using the training set. These models were used to identify stratum information and characterize mapping relationships, respectively. The performance of these models was verified using the test set. In step 2, a multi-output prediction model for rock mass parameters is constructed based on the tabular deep learning architecture TabNet, using operating parameters and passive parameters as inputs and rock mass uniaxial compressive strength and rock mass quality indicators as outputs. During the TBM tunneling process, operating parameters and passive parameters are directly generated. These parameters are used as inputs for the multi-output prediction model for rock mass parameters, and information about the excavation stratum of the tunnel ring can be identified. In step 2, based on the tabular deep learning architecture TabNet, a multi-output prediction model for tunneling parameters is constructed with operating parameters and rock mass parameters as inputs and tunneling rate and tool consumption per linear meter as outputs. During the tunneling process, newly collected parameter information is dynamically added to the database, and the multi-output prediction model for rock mass parameters and tunneling parameters is continuously trained to enhance model performance. Step 3: Based on considerations of tunneling efficiency, construction costs, equipment safety, and engineering experience, a multi-objective optimization algorithm is constructed. The rock mass parameters are predicted using a multi-output rock mass parameter prediction model, and tunneling conditions are identified. Using the multi-output rock mass parameter prediction model as a fitting function, the TBM tunneling state is optimized based on the actual tunneling conditions and decision values ​​for operating parameters are determined. In step 3, the Black Kite algorithm is improved into a multi-objective optimization algorithm using a weighted Chebyshev scalarization strategy. Then, the initial population generation strategy of the Black Kite algorithm is replaced with a Tent chaos mapping strategy, thereby constructing a multi-objective Black Kite optimization algorithm. The operating parameters and passive parameters directly generated during the TBM tunneling process are used as inputs to a multi-output prediction model for rock mass parameters to obtain actual rock mass parameters. The tunneling working condition is identified based on the uniaxial compressive strength of the rock mass in the actual rock mass parameters, and an optimization scheme is determined. During the optimization process, target values ​​and weights are set according to engineering requirements, while the rock parameters remain unchanged. The tunneling parameter multi-output prediction model is used as a fitting function, and the operating parameters and actual rock mass parameters directly generated during the TBM tunneling process are used as inputs to obtain the tunneling rate and tool consumption per linear meter. Then, weights and target values ​​for the tunneling rate and tool consumption per linear meter are set in combination with the excavation plan and construction budget. Different optimization schemes are formulated for different working conditions by comprehensively considering tunneling efficiency, construction cost, equipment safety, and engineering experience, and ultimately, corresponding decision values ​​for cutterhead speed, cutterhead torque, and cutterhead thrust are determined.

2. The intelligent decision-making method for TBM operating parameters based on dynamic identification of stratum information according to claim 1 is characterized in that: The operating parameters include the cutterhead speed, cutterhead torque and cutterhead thrust; the passive parameters include the support shoe support force, the left pressure of the top shield cylinder, the right pressure of the top shield cylinder, the left pressure of the side shield cylinder and the right pressure of the side shield cylinder; the rock mass parameters include the uniaxial compressive strength of the rock mass and the rock mass quality index.

3. The intelligent decision-making method for TBM operating parameters based on dynamic identification of formation information according to claim 2 is characterized in that: The parameter information is preprocessed to construct a database, including using the isolation forest algorithm to identify outliers in the parameter information, then using linear interpolation to replace the outliers, and using the Min-Max method to normalize the data. The calculation formula is as follows: in, x 、 x * is the characteristic value before and after normalization, min ( x ), max ( x ) are the minimum and maximum values ​​in the feature, respectively.

4. The intelligent decision-making method for TBM operating parameters based on dynamic identification of formation information according to claim 3 is characterized in that: The three excavation conditions are established according to the uniaxial compressive strength of the rock mass, i.e., a soft rock formation is defined as one not exceeding 30 MPa, a balanced formation is defined as one between 30 MPa and 90 MPa, and a hard rock formation is defined as one exceeding 90 MPa.

5. The intelligent decision-making method for TBM operating parameters based on dynamic identification of formation information according to claim 4 is characterized in that: In step 3, by establishing the torque safety factor K T and thrust safety factor K TF Realize equipment safety considerations, namely, to limit excessive torque and excessive thrust in the decision-making process; in, Actual torque is the actual cutter head torque value of the TBM equipment, TBM maximum torque The maximum cutter head torque allowed for TBM equipment; in, Actual thust force is the actual cutterhead thrust value of the TBM equipment, TBM maximum thrust The maximum cutterhead thrust allowed for TBM equipment.

6. The intelligent decision-making method for TBM operating parameters based on dynamic identification of formation information according to claim 5 is characterized in that: In step 3, by establishing the engineering experience coefficient K e Taking engineering experience into account, i.e. limiting the adjustment range of operating parameters during the decision-making process; in, T original is the initial value of the cutterhead thrust, T optimized is the decision value of the cutterhead thrust, TF original is the initial value of the cutter head torque, TF optimized is the decision value of the cutter head torque, RPM original is the initial value of the cutter head speed, RPM optimized is the decision value of the cutter head speed.

Images