Intelligent control method for tunnel boring by adopting TBM (Tunnel Boring Machine) equipment

Abstract

The invention provides an intelligent control method for tunnel boring through TBM equipment, and the method comprises the steps: obtaining a plurality of pieces of boring data, and determining a plurality of pieces of feature boring data related to rock mass information from the plurality of pieces of boring data; the multiple pieces of feature tunneling data are input into a target TBM surrounding rock sensing algorithm, multiple rock mass mechanical parameters are obtained, and the target TBM surrounding rock sensing algorithm is determined according to multiple pre-trained TBM surrounding rock sensing models; inputting the plurality of rock mass mechanical parameters into a target tunneling parameter decision model to obtain a plurality of tunneling control parameters; according to the multiple tunneling control parameters, tunneling equipment is controlled to conduct intelligent control tunneling work. The tunneling speed of the TBM equipment can be increased, and the service life of a cutter of the TBM equipment can be prolonged.

Description

Technical Field

[0001] The present application relates to the technical field of tunnel construction, and in particular to an intelligent control method for tunnel excavation using TBM equipment. Background Art

[0002] With the continuous development of the economy, the emergence of more and more super tunnels with large crossing ranges and complex environments has brought great technical difficulties and challenges to tunnel construction.

[0003] In the study of the mapping relationship between excavation parameters and surrounding rock, Nelson et al. found that the TBM field excavation index FPI is positively linearly correlated with rock hardness, and proposed to use this index to characterize the excavability of a certain stratum. Rourke et al. derived the relationship between the TBM field excavation index and the total hardness of the rock based on the data of metamorphic rock tunnels, which is similar to the conclusion of Nelson. Hamidi, Hassanpour, Farrokh et al. analyzed the TBM excavation data and found that FPI or ln(FPI) is linearly related to rock strength. Through regression, the relationship model between the two can be obtained, and a bridge for mutual estimation of rock strength and equipment performance parameters is established. Leu et al. and Guan et al. understand the change of geological conditions as a random process of state transfer. The hidden Markov method is used to predict the change of geological conditions, which can effectively overcome the problem of insufficient geological information. The prediction results can be used to support other construction decisions. Liu Bin et al. suggested using machine learning methods to predict rock parameters from machine parameters, and proposed to combine advanced detection and other methods to comprehensively obtain rock parameters, and conducted preliminary exploration using the actual data of TBM in the Jilin Yinsong Water Supply Project. Shi et al. proposed a geological prediction method based on deep neural network, which can provide relatively accurate geological prediction results based on TBM excavation data. This method verifies the applicability and potential of deep learning in rock mass perception. Zhang et al. used big data mining methods and artificial intelligence algorithms to classify the TBM excavation status and its corresponding rock mass status of the Jilin Yinsong Water Supply Project based on four parameters: cutterhead speed, cutterhead torque, thrust and excavation speed.

[0004] Based on the above research results, the correlation between TBM excavation parameters and rock mass mechanical parameters is analyzed, and the perception and evaluation of the rock mass status during excavation are performed using methods such as multivariate regression, pattern recognition or artificial intelligence. With the development of the Internet of Things and big data technology, more comprehensive collection of TBM excavation information and rock mass information, and the establishment of a TBM excavation data big data sample library, play an important role in achieving real-time perception and evaluation of the rock mass during excavation.

[0005] The uniaxial compressive strength of rock mass and the development of joints are key variables that characterize the strength characteristics and integrity of the rock mass at the excavation face. However, the above-mentioned rock mechanical parameters are difficult to measure directly online in real time through existing technical means. They can only be measured offline during TBM shutdown, and intelligent control of TBM equipment cannot be achieved. Summary of the invention

[0006] In view of this, the purpose of this application is to provide an intelligent control method for tunnel excavation using TBM equipment, which can determine the rock mechanical parameters based on multiple excavation data, and then intelligently control the excavation equipment based on multiple models or algorithms, which can not only improve the excavation speed of the TBM equipment, but also improve the tool life of the TBM equipment.

[0007] An embodiment of the present application provides an intelligent control method for tunnel excavation using TBM equipment, characterized in that the method includes: acquiring multiple excavation data, and determining multiple characteristic excavation data related to rock mass information from the multiple excavation data; inputting the multiple characteristic excavation data into a target TBM surrounding rock perception algorithm to obtain multiple rock mass mechanical parameters, wherein the target TBM surrounding rock perception algorithm is determined based on multiple pre-trained TBM surrounding rock perception models; inputting the multiple rock mass mechanical parameters into a target excavation parameter decision model to obtain multiple excavation control parameters; and controlling the excavation equipment to perform intelligent control excavation work based on the multiple excavation control parameters.

[0008] Optionally, multiple characteristic excavation data are determined by the following steps: based on multiple historical excavation data, the correlation between the rock quality grade and multiple on-site excavation indexes is determined, and the on-site excavation index is the ratio of the thrust of a single disc cutter of the TBM equipment to the penetration rate; based on the correlation between the rock quality grade and the multiple on-site excavation indexes, multiple characteristic excavation data are determined, and the characteristic excavation data include single cutter thrust, cutter head torque, cutter head speed and penetration rate.

[0009] Optionally, the TBM surrounding rock perception algorithm includes a least squares regression sub-model, a BP neural network sub-model and a vector machine sub-model, wherein the target TBM surrounding rock perception algorithm is determined by the following formula: ; Among them, w1 is the weight value of the least squares regression sub-model, w2 is the weight value of the BP neural network sub-model, and w3 is the weight value of the vector machine sub-model. are multiple rock mass mechanical parameters calculated by the least squares regression sub-model, Multiple rock mass mechanical parameters calculated by the BP neural network sub-model, Several rock mechanical parameters calculated for the vector machine submodel.

[0010] Optionally, the least squares regression sub-model is trained by the following steps: based on the least squares method model, a least squares regression sub-model is established according to multiple rock mechanical parameters to be measured and multiple characteristic excavation data; multiple historical excavation data are input into the initial least squares regression sub-model to obtain a first loss function value of the initial least squares regression sub-model; according to the obtained first loss function value, the initial least squares regression sub-model is trained until the first loss function value of the initial least squares regression sub-model converges to a minimum value; according to the least squares regression sub-model when the first loss function value of the initial least squares regression sub-model converges to a minimum value, a target least squares regression sub-model is determined.

[0011] Optionally, the multiple rock mass mechanical parameters to be measured include the uniaxial compressive strength of rock and the number of rock mass joints per unit volume, wherein the target least squares regression sub-model includes the following formula: ; ; Among them, F n is the thrust of a single knife, p is the penetration, n is the rotation speed of the cutter head, T is the torque of the cutter head, is the uniaxial compressive strength of rock; is the number of joints per unit volume of rock mass.

[0012] Optionally, the BP neural network sub-model is determined by the following steps: using the learning mechanism of signal forward propagation and error reverse regulation, a BP network model for processing nonlinear information is built through iterative learning; using the GA algorithm to optimize the BP network model hyperparameters, thereby determining the target BP network model.

[0013] Optionally, the vector machine sub-model is determined by the following steps: multiple initial vector machine sub-models are determined based on multiple historical excavation data; the vector machine sub-model is determined based on the least squares support vector machine function, the Lagrangian function and the kernel function; the GA algorithm is used to optimize the hyperparameters of the vector machine sub-model to determine the target vector machine sub-model.

[0014] Optionally, the weight values ​​of the least squares regression sub-model, the BP neural network sub-model and the weight value of the vector machine sub-model are determined by the following steps: multiple evidence sets are generated according to the multiple excavation data; based on the multiple evidence sets, the weight values ​​of the least squares regression sub-model, the BP neural network sub-model and the first intermediate weight value of the vector machine sub-model are calculated with the goal of minimizing the mean square error of the error, and a BPA function is constructed according to the weight values ​​of the least squares regression sub-model, the BP neural network sub-model and the first intermediate weight value of the vector machine sub-model; the credibility of each evidence set is determined according to the distance of the BPA function between different evidence sets; the BPA function is corrected according to the credibility of each evidence set, and the corrected BPA functions on different evidence sets are synthesized according to the Dempster synthesis rule to obtain the weight values ​​of the least squares regression sub-model, the BP neural network sub-model and the weight value of the vector machine sub-model.

[0015] Optionally, the target excavation parameter decision model is determined by the following steps: a TBM excavation parameter optimization model based on a multi-objective particle swarm optimization algorithm is constructed based on an MLP neural network rock-machine parameter mapping model; multiple pre-set constraints are converted into a feasible domain of decision variables to change the constraint processing strategy in the multi-objective particle swarm optimization algorithm; and multiple relevant parameters of the multi-objective particle swarm optimization algorithm are set to obtain a target excavation parameter decision model.

[0016] Optionally, the method also includes: obtaining multiple verification excavation data of excavation equipment under intelligent control excavation work and manual control excavation work in multiple sections with different surrounding rock types, the multiple verification excavation data including multiple excavation efficiency data and tool life data; determining multiple efficiencies of the intelligent control excavation work in multiple sections with different surrounding rock types and multiple efficiencies of the manual control excavation work in multiple sections with different surrounding rock types based on comparison between the multiple verification excavation data of excavation equipment under intelligent control excavation work and manual control excavation work in multiple sections with different surrounding rock types; determining the efficiency improvement of the intelligent control excavation work relative to the manual control excavation work based on the multiple efficiencies of the intelligent control excavation work in multiple sections with different surrounding rock types and the multiple efficiencies of the manual control excavation work in multiple sections with different surrounding rock types.

[0017] The intelligent control method for tunnel excavation using TBM equipment provided in the embodiment of the present application can determine the rock mechanical parameters based on multiple excavation data, and then intelligently control the excavation equipment based on multiple models or algorithms, which can not only improve the excavation speed of the TBM equipment, but also improve the tool life of the TBM equipment.

[0018] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A flow chart of an intelligent control method for tunnel excavation using TBM equipment provided in an embodiment of the present application; Figure 2 A schematic diagram of a technical solution for soft measurement of TBM surrounding rock parameters provided in an embodiment of the present application; Figure 3 A schematic diagram of a tunneling record table provided in an embodiment of the present application; Figure 4 A schematic diagram of excavation data collection and update iteration provided in an embodiment of the present application; Figure 5 A schematic diagram of tunneling data stream transmission, processing, and services provided in an embodiment of the present application; Figure 6 The probability statistics diagram of FPI distribution of granite and limestone strata provided in the embodiments of the present application; Figure 7 A statistical table of cutterhead vibration parameters for different surrounding rock integrity coefficient intervals provided in the embodiment of the present application; Figure 8 A schematic diagram of the correlation between the main excavation parameters provided in the embodiment of the present application; Fig. 9 A schematic diagram of the positive and negative correlation ranking of FPI and main excavation parameters provided in the embodiment of the present application; Fig.10 A schematic diagram of a neural network topology structure provided in an embodiment of the present application; Fig.11 A flowchart of the GA algorithm model provided in the embodiment of the present application; Fig.12 A basic flow chart of multi-model fusion based on DS evidence provided in the embodiment of the present application; Fig.13 Schematic diagram of different mapping function curves between the credibility factor ξ and ui provided in the embodiments of the present application; Fig.14 A decision diagram of the excavation parameter decision model provided in an embodiment of the present application; Fig.15 A schematic diagram of the Pareto frontier provided in an embodiment of the present application; Fig.16 A schematic diagram of a particle position update mechanism provided in an embodiment of the present application; Fig.17 A flow chart of the particle swarm algorithm provided in the embodiment of the present application; Fig.18 A flow chart of a multi-objective particle swarm algorithm provided in an embodiment of the present application; Fig.19 A schematic diagram of a particle position update mechanism provided in an embodiment of the present application; Fig. 20 A table of parameter settings for a multi-objective particle swarm optimization algorithm provided in an embodiment of the present application; Fig.21 A schematic diagram of the surrounding rock conditions of the test section provided in the embodiment of the present application; Fig. 22 The test segment division and basic information diagram provided for the embodiment of the present application; Fig.23 A schematic diagram of a verification scheme provided for an embodiment of the present application; Fig.24 A comparison chart of the average excavation efficiency of the intelligent auxiliary decision-making test section and the manual control test section in various surrounding rock types provided in the embodiments of the present application; Fig.25 A comparison chart of the measured propulsion speeds in the test sections 1-1 and 1-2 provided in the embodiment of the present application; Fig.26 A comparison chart of the measured average single-knife rock breaking amount in the test sections 1-1 and 1-2 provided in the embodiments of the present application; Fig. 27 A comparison chart of the measured propulsion speeds in test sections 2-1 and 2-2 provided in the embodiment of the present application; Fig.28 A comparison chart of the measured average single-knife rock breaking amount in the test sections 2-1 and 2-2 provided in the embodiment of the present application; Fig.29 A comparison chart of the measured propulsion speeds in test sections 3-1 and 3-2 provided in the embodiment of the present application; Fig.30 A comparison chart of the measured average single-knife rock breaking amount in the test sections 3-1 and 3-2 provided in the embodiment of the present application; Fig.31 A comparison chart of the measured propulsion speeds in test sections 4-1 and 4-2 provided in the embodiment of the present application; Fig.32 This is a comparison chart of the measured average single-knife rock breaking amount in the test sections 4-1 and 4-2 provided in the embodiments of the present application. DETAILED DESCRIPTION

[0021] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme 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 only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work belongs to the scope of protection of the present application.

[0022] First, the application scenarios to which this application is applicable are introduced. This application can be applied to the text readability evaluation of the informed consent form.

[0023] An intelligent control method for tunnel excavation using TBM equipment provided in an embodiment of the present application can determine rock mechanical parameters based on multiple excavation data, and then intelligently control the excavation equipment based on multiple models or algorithms, which can not only improve the excavation speed of the TBM equipment, but also improve the tool life of the TBM equipment.

[0024] See also Figure 1 , Figure 1 The flowchart of an intelligent control method for tunnel excavation using TBM equipment provided in the embodiment of the present application. Figure 1 As shown in , the intelligent control method for tunnel excavation using TBM equipment provided in the embodiment of the present application includes: S101, acquiring a plurality of excavation data, and determining a plurality of characteristic excavation data related to rock mass information from the plurality of excavation data.

[0025] Here, please see Figure 2 , Figure 2 Schematic diagram of the technical solution for soft measurement of TBM surrounding rock parameters provided in the embodiment of the present application. Figure 2 As shown in , the technical solution for soft measurement of TBM surrounding rock parameters provided in the embodiment of the present application realizes real-time perception of the mechanical parameters of the rock mass when the excavation face is excavating by selecting equipment operating parameters (measurable and easy-to-measure auxiliary variables) and rock mechanical parameters (variables to be measured and difficult to measure) and establishing a mapping relationship model between the above key variables.

[0026] Specifically, the excavation data may be TBM excavation parameters, which refer to various technical parameters that need to be controlled and adjusted when using TBM for tunnel excavation. These parameters directly affect the efficiency and safety of excavation and the quality of the tunnel.

[0027] As an example, see Figure 3 , Figure 3 This is a schematic diagram of the excavation record table provided in the embodiment of the present application. Figure 3 As shown in Figure 3 The parameters in the tunneling system can be entered into the database in real time. The intelligent data collection module is needed to collect the tunneling parameters generated during construction. The tunneling parameters generated during construction, mainly thrust, cutterhead speed, cutterhead torque, penetration, tunneling speed, etc., can be collected through the intelligent data collection system. MySQL processes the data and connects it to the intelligent tunneling system, and finally displays it in the intelligent tunneling system.

[0028] As an example, see Figure 4 , Figure 4 This is a schematic diagram of the excavation data collection and update iteration provided in the embodiment of the present application. Figure 4 As shown in , the realization of tunnel construction automation and intelligence requires rapid acquisition and effective processing of data, such as Figure 4 As shown, a TBM intelligent tunneling data stream transmission mechanism is established, and the tunneling machine data, geological information, surrounding rock perception data and jamming machine cases collected on site are transmitted to the main server in real time through cables or wireless signals. On this basis, this application proposes a TBM tunneling database, which provides strong data support for construction and scientific research by building a database support platform. After fine classification and stratification, the database eventually forms sub-libraries of seven data types, including a digital resource library, a geological prediction resource library, a tunneling parameter database, a slag image database, a vibration time history database, a support parameter database, and a jamming machine case library.

[0029] As an example, see Figure 5 , Figure 5 A schematic diagram of tunneling data stream transmission, processing, and services provided in an embodiment of the present application.

[0030] Among them, after a long period of automatic input of geological, excavation, rock slag, vibration and support parameters, about 20 million data have been obtained, among which approximately 1.6 million data are generated in each TBM cycle.

[0031] Specifically, multiple characteristic excavation data can be determined through the following steps: based on multiple historical excavation data, determining the correlation between the rock quality grade and multiple on-site excavation indexes, the on-site excavation index is the ratio of the thrust of a single disc cutter of the TBM equipment to the penetration rate; based on the correlation between the rock quality grade and the multiple on-site excavation indexes, determining multiple characteristic excavation data, the characteristic excavation data including single cutter thrust, cutter head torque, cutter head rotation speed and penetration rate.

[0032] It should be noted that due to the large and complex nature of TBM equipment, there are as many as 200 to 600 operating parameters monitored by sensors. The selection and optimization of auxiliary variables plays a decisive role in the success of the soft measurement method. On the one hand, it affects the accuracy of the soft measurement model, and on the other hand, it affects the generalizability of the model.

[0033] Among them, the field advancement index (FPI) is the ratio of the thrust of a single cutter to the penetration rate, which reflects the thrust required for the cutter to penetrate 1 mm into the rock and is closely related to the conditions of the rock mass being excavated. Since Nelson et al. (1983), many scholars have found that the TBM field advancement index (FPI) is strongly correlated with the mechanical properties of the rock mass being excavated.

[0034] Since FPI eliminates the influence of different numbers of TBM cutters, different tunnel diameters and different operating parameters (penetration), it is very suitable for horizontal comparison between different projects. Through excavation tests and a large amount of actual excavation data, the variation between rock mass mechanical parameters and thrust and penetration has been analyzed in detail. As a direct calculation value of thrust and penetration, FPI also has a very obvious correlation with the surrounding rock grade.

[0035] This application counts the mean FPI values ​​of the excavation cycles of two main lithologies with different quality levels in the TBM project, and uses data processing methods to analyze the distribution of the mean FPI values ​​of all 13,000 excavation cycles in the project. For details, please refer to Figure 6 , Figure 6 This is a probability statistical diagram of FPI distribution for granite and limestone formations; please refer to Figure 7 , Figure 7 This is a statistical table of cutterhead vibration parameters in different surrounding rock integrity coefficient intervals.

[0036] In this way, through Figure 6 and Figure 7 The analysis shows that: (1) The higher the surrounding rock quality grade, the greater the roller thrust required for unit penetration. As the surrounding rock changes from Class II to Class V, the peak value of the FPI interval probability distribution gradually moves to the left. The FPI means of granite and limestone formations are 28, 24, 16, 13 and 29, 27, 19, 18, respectively, indicating that the worse the surrounding rock strength and integrity, the smaller the roller thrust required for unit penetration; (2) For surrounding rocks of the same grade, the probability distribution of the granite FPI interval is relatively concentrated, while the probability distribution of the limestone FPI interval is relatively dispersed. Under different surrounding rock grade conditions, the FPI means of granite formations vary greatly, while the FPI means of limestone formations are relatively small, indicating that the surrounding rock grade of granite formations is more correlated with FPI.

[0037] It should be noted that the statistical results of surrounding rock quality grades are derived from the tracking records during TBM excavation and are somewhat subjective; however, the statistical laws of tens of thousands of sets of data in the charts are sufficient to verify their relevance.

[0038] Furthermore, this paper takes FPI as a reference variable related to the surrounding rock quality grade and selects the main TBM excavation parameters for correlation analysis.

[0039] This paper chooses the Pearson correlation coefficient to determine the correlation of auxiliary variables, and its correlation coefficient can be calculated by the following formula: Where cov(X,Y) is the covariance of random variables X and Y, σ(X) and σ(Y) are the standard deviations of the variables.

[0040] If X and Y are independent of each other, then p(X,Y)=0; the correlation coefficient from 0 to 1 indicates the change from unrelated to strongly correlated variables.

[0041] The main applicable scope of the Pearson correlation coefficient is: ① The standard deviation of the two variables is not 0; ② There is a linear relationship between the two variables and they are continuous data; ③ The two variables are normally continuously distributed; ④ The observed values ​​of the two variables are independent of each other.

[0042] This paper once again uses the characteristic value sample set of the stable section of the entire TBM excavation cycle to analyze the correlation results. Figure 8 As shown, the darker the red, the stronger the positive correlation, and the darker the blue, the stronger the negative correlation.

[0043] In order to facilitate the selection of auxiliary variables used in the soft sensor model, the main variables with high positive and negative correlations with the FPI Pearson correlation coefficient are selected and sorted. Fig. 9 .

[0044] from Figure 8-9 It can be seen that there is a strong positive correlation between the total thrust, support shoe pressure, cutter head speed and its setting value, cutter head power and torque and FPI; there is a strong negative correlation between penetration, thrust speed and its setting value and FPI.

[0045] Here, in the actual excavation process, considering the close correlation between the support shoe pressure and the total thrust, between the cutterhead power and the cutterhead torque, between each operating parameter and its set value, and between the excavation speed and the cutterhead speed and penetration, after eliminating redundant variables based on business knowledge, this application selects four groups of parameters: single-knife thrust, cutterhead torque, cutterhead speed and penetration as auxiliary variables in soft measurement modeling of rock excavation.

[0046] S102, inputting a plurality of characteristic excavation data into a target TBM surrounding rock perception algorithm to obtain a plurality of rock mass mechanical parameters, wherein the target TBM surrounding rock perception algorithm is determined based on a plurality of pre-trained TBM surrounding rock perception models.

[0047] Specifically, the TBM surrounding rock perception algorithm includes a least squares regression sub-model, a BP neural network sub-model and a vector machine sub-model.

[0048] Among them, the target TBM surrounding rock perception algorithm is determined by the following formula: ; Among them, w1 is the weight value of the least squares regression sub-model, w2 is the weight value of the BP neural network sub-model, and w3 is the weight value of the vector machine sub-model. are multiple rock mass mechanical parameters calculated by the least squares regression sub-model, Multiple rock mass mechanical parameters calculated by the BP neural network sub-model, Several rock mechanical parameters calculated for the vector machine submodel.

[0049] Specifically, the least squares regression sub-model can be trained through the following steps: based on the least squares method model, a least squares regression sub-model is established according to multiple rock mass mechanical parameters to be measured and multiple characteristic excavation data; multiple historical excavation data are input into the initial least squares regression sub-model to obtain the first loss function value of the initial least squares regression sub-model; according to the obtained first loss function value, the initial least squares regression sub-model is trained until the first loss function value of the initial least squares regression sub-model converges to a minimum value; according to the least squares regression sub-model when the first loss function value of the initial least squares regression sub-model converges to a minimum value, the target least squares regression sub-model is determined.

[0050] Specifically, the multiple rock mass mechanical parameters to be measured include the uniaxial compressive strength of rock and the number of rock mass joints per unit volume.

[0051] Wherein, the target least squares regression sub-model includes the following formula: ; ; Among them, F n is the thrust of a single knife, p is the penetration, n is the rotation speed of the cutter head, T is the torque of the cutter head, is the uniaxial compressive strength of rock; is the number of joints per unit volume of rock mass.

[0052] It should be noted that the least squares method is a form of expression that obtains the "best" or "most likely" result through a large number of observations and measurements of an event. It is one of the most important statistical methods at present. Several major statistical branches such as regression analysis, variance analysis and linear model theory are based on the least squares method.

[0053] Here, the mathematical process of least squares solution is given. First, assume that the regression training set is ,in , , select the linear function f through linear regression to model this training set: Among them: w, x are column vectors, is the inner product of the two. The least squares method defines the square sum loss function as follows: ; Then, the error sum is minimized to find the value of the parameter (w, b). Therefore, the training set vector expression is changed to ,and: The output vector difference then becomes: , the square sum loss function is written in matrix form as: ; Then the parameters Taking the partial derivative and setting the resulting linear expression equal to 0 gives the standard equation.

[0054] Then, based on the idea of ​​least squares method, this paper constructs a regression model between the rock mass mechanical parameters Y to be measured and the known TBM excavation parameter matrix A as shown in the following formula: ; In the formula, the rock mass state parameter , regression coefficient vector The tunneling parameters are derived from the soft sensor auxiliary variables selected in the previous section and are expressed as the matrix Among them, J v is the number of rock joints per unit volume, UCS is the uniaxial compressive strength of rock, F n is the single-knife thrust, p is the penetration rate, N is the cutter head speed, T is the cutter head torque, and Z is the number of samples.

[0055] Since it is an overdetermined system of equations, the least squares method is used to determine the regression coefficient vector B so that the loss function H is minimized: When H is minimum, , thus obtaining the regression coefficient in the sense of minimum mean square error.

[0056] Using the TBM rock machine parameter database, with the help of human experience and comparative verification, we finally constructed and solved The characteristic equation is as follows, and the multivariate nonlinear regression model between the mechanical properties of rock mass and TBM excavation parameters is obtained: ; ; Here, the expression of the regression model needs to combine statistical analysis and human experience. The rock mass mechanical characteristic parameters have a linear relationship with the single-cutter thrust, penetration, and cutterhead speed, and a binary nonlinear relationship with the cutterhead torque. For nonlinear relationship variables, parameter fitting can also use logarithmic or exponential relationships.

[0057] However, due to the weak nonlinear representation ability of the least squares method, the selection of variable parameters and model form depends on human experience. In order to improve the perception accuracy and stability of the soft sensor model, this application also introduces two types of machine learning algorithms, back propagation (BP) neural network and support vector machine regression (SVR), to model the rock mass mechanical parameters.

[0058] Specifically, the BP neural network sub-model can be determined through the following steps: using the learning mechanism of signal forward propagation and error reverse regulation, a BP network model for processing nonlinear information is built through iterative learning; using the GA algorithm to optimize the hyperparameters of the BP network model, thereby determining the target BP network model.

[0059] Specifically, the vector machine sub-model can be determined by the following steps: multiple initial vector machine sub-models are determined based on multiple historical excavation data; the vector machine sub-model is determined based on the least squares support vector machine function, the Lagrangian function and the kernel function; the GA algorithm is used to optimize the hyperparameters of the vector machine sub-model to determine the target vector machine sub-model.

[0060] Among them, the artificial neural network (BP neural network) imitates the response process of brain neurons to external signals, establishes a multi-layer perception simulator, uses the learning mechanism of signal forward propagation and error reverse regulation, and builds a network model for processing nonlinear information through iterative learning. Therefore, it has powerful data recognition and simulation capabilities, and has obvious advantages in solving nonlinear system problems.

[0061] As an example, see Fig.10 , Fig.10 A schematic diagram of a neural network topology structure provided in an embodiment of the present application.

[0062] Among them, the support vector machine is based on a nonlinear mapping algorithm, which maps the input vector to a high-dimensional feature space and constructs an optimal separation hyperplane to achieve intelligent classification.

[0063] The vector machine sub-model proposed in this application is LSSVM. LSSVM transforms the inequality constraints in the SVM method into equality constraints, providing convenient conditions for solving the Lagrange multiplier. The mathematical solution process is as follows.

[0064] First, the data samples in the training set are Represents, i=1,…,n, y i is the target result of the i-th sample, x i is the input value of the i-th sample, and n is the number of training samples. The SVM model can be expressed in the feature space as follows: ; Where: is the nonlinear transformation mapping function, w is the weight vector, and b is the bias.

[0065] Secondly, the objective function of LSSVM can be expressed as follows: ; Where: is the error variable; r>0 is the penalty coefficient.

[0066] Again, by introducing the Lagrangian function, we can get: ; In the formula, a i is the Lagrange multiplier. According to the KKT (Karush-Kuhn-Tucker, referred to as KKT) condition, calculate The following formula can be obtained: ; in, ; ; .

[0067] I is the unit matrix. According to the Mercer condition, the kernel function can be written as: Finally, after jointly solving a and b, we can get the nonlinear function of LSSVM: Neural networks have defects such as long learning time and easy to fall into local minima. The use of genetic algorithms (GA) to train neural networks does not require prior knowledge, is insensitive to initial parameters, has strong global search capabilities, and can quickly find the best parameter combination, with higher stability and stronger global optimization capabilities, greatly improving training speed and prediction accuracy. When the support vector machine uses radial basis RBF as the kernel function, it is necessary to set reasonable penalty parameters and kernel function parameters to improve the classification accuracy of the model. Genetic algorithms can also be used to optimize parameters and search optimization. Therefore, this application uses the GA algorithm to optimize the hyperparameters of the above two machine learning models to obtain the optimal model.

[0068] Specifically, the overall process of the GA algorithm model is as follows Fig.11 As shown, no further description is given here.

[0069] Specifically, the weight values ​​of the least squares regression sub-model, the BP neural network sub-model, and the weight value of the vector machine sub-model can be determined by the following steps: generating multiple evidence sets based on the multiple excavation data; based on the multiple evidence sets, calculating the weight values ​​of the least squares regression sub-model, the BP neural network sub-model, and the first intermediate weight value of the vector machine sub-model with the goal of minimizing the mean square error of the error, and constructing a BPA function based on the weight values ​​of the least squares regression sub-model, the BP neural network sub-model, and the first intermediate weight value of the vector machine sub-model; determining the credibility of each evidence set based on the distance of the BPA function between different evidence sets; correcting the BPA function based on the credibility of each evidence set, and synthesizing the corrected BPA functions on different evidence sets according to the Dempster synthesis rule to obtain the weight values ​​of the least squares regression sub-model, the BP neural network sub-model, and the weight value of the vector machine sub-model.

[0070] Here, the least squares method has the advantages of strong interpretability and simple and convenient application, but due to the many influencing factors and the complex modeling process, the model establishment relies on the professional experience and ability of researchers. The machine learning model has a strong nonlinear representation ability, but it is easily affected by the sample size and data quality, resulting in problems of overfitting and poor generalization ability. In order to solve the above problems, this application proposes a decision-level fusion method based on the improved DS evidence theory to realize the above-mentioned multiple models.

[0071] As an example, see Fig.12 , Fig.12 The basic flow chart of multi-model fusion based on DS evidence provided in the embodiment of the present application.

[0072] Here, the evidence theory was first proposed by Dempster in 1967, and later extended by Shafer in 1976 to form evidence reasoning, so it is called DS evidence theory. DS evidence theory can provide a powerful tool for the expression and synthesis of uncertain information, so it has been widely used in expert systems and information fusion.

[0073] Specifically, the principle of evidence theory is as follows: Assume Θ is a hypothesis space, which is a finite non-empty set. Any proposition A in the space should be contained in 2Θ. Define a mapping function m (mass function) from 2Θ→[0,1] that satisfies: ; m(A) is called the basic probability assignment (BPA) of proposition A. The belief function Bel(A) and the likelihood function Pl(A) are defined as follows: ; ; but The confidence interval of event A indicates the degree of confidence in A.

[0074] Assumptions , evidence collection , BPA is m1, m2, …, m n , then the Dempster synthesis rule is: ; In the formula, It reflects the conflict of evidence, and the coefficient 1 / (1-k) is the regularization factor.

[0075] Reasonable configuration of BPA is an important factor that determines whether DS evidence theory can be effectively applied. First, N subsets are randomly selected from the training data set as evidence sets. Then, BPA is defined on evidence set i according to the prediction errors of different models, giving larger weights to models with smaller prediction errors and smaller weights to models with larger prediction errors.

[0076] The specific definition method of BPA is given. On the evidence set i, it is assumed that the weights assigned to the three soft measurement models of rock mass mechanical parameters, namely, least squares regression, BP neural network and SVM support vector machine, are w i1 、w i2 、w i3 , then the final fusion result of rock mass mechanical parameters is , and the error mean and variance of the fusion results are: ; For the prediction errors of different soft sensor models, it is assumed that they are independent of each other, that is, In order to minimize the error variance of the prediction results, the partial derivative of D(E) is calculated, and ; When satisfied , the following results can be obtained: ; Where w i1 、w i2 、w i3 BPA that can be used as evidence set i.

[0077] However, the traditional DS evidence theory will produce the Zaldeh paradox when synthesizing high-conflict evidence sources. To solve the above problem, this application introduces a credibility factor ξ to measure the reliability of the evidence source, and modifies the evidence source according to ξ to reduce the conflict between different evidence sources.

[0078] Specifically, first, define the distance between the observations of evidence set i and the observations of other evidence sets: The size of ui reflects the consistency between the observations of evidence set i and other evidence sets. The smaller ui is, the higher the credibility is. Fig.13 , Fig.13 The mapping function curves of different types and parameters such as parabolic function, exponential function, trigonometric function, etc. are compared. When ui is small, the credibility of evidence is higher, and ξi decays slowly with ui; when ui is large, the credibility of evidence is low, and ξi decays rapidly to 0 with ui. After comparative analysis, the mapping function selection is as follows: The BPA corrected by the credibility factor is defined as , the calculation formula is as follows: ; Finally, the modified BPA on different evidence sets is synthesized according to the Dempster synthesis rule to obtain the final weight as the weight coefficient of multi-model fusion.

[0079] As an example, after calculation, the weight coefficients of the three soft sensor models of least squares regression, BP+GA, and LSSVR+GA obtained in this application are 0.2470, 0.3452, and 0.4078, respectively.

[0080] According to the above improved fusion method, the prediction results of the three models are fused at the decision level to obtain the final TBM surrounding rock perception model expression as follows: ; Where: Y is the rock mass mechanical parameter obtained after algorithm fusion, w i is the weight value obtained after DS evidence fusion of the three algorithms, , , These are the rock mechanical parameters predicted by the three soft measurement models: least squares regression, BP neural network and support vector machine.

[0081] S103, inputting a plurality of rock mass mechanical parameters into a target excavation parameter decision model to obtain a plurality of excavation control parameters.

[0082] Specifically, the target excavation parameter decision model can be determined through the following steps: constructing a TBM excavation parameter optimization model based on a multi-objective particle swarm optimization algorithm based on an MLP neural network rock-machine parameter mapping model; converting multiple pre-set constraints into a feasible domain of decision variables to change the constraint processing strategy in the multi-objective particle swarm optimization algorithm; setting multiple relevant parameters of the multi-objective particle swarm optimization algorithm to obtain a target excavation parameter decision model.

[0083] Here, considering the efficiency improvement, safe excavation and actual construction needs, this application proposes an excavation parameter decision model. Specifically, the excavation parameter decision model includes three aspects: constraint conditions, parameter optimization, and strategy decision.

[0084] As an example, see Fig.14 , Fig.14 A decision diagram of the tunneling parameter decision model provided in an embodiment of the present application.

[0085] A feasible framework is provided for solving multi-objective optimization problems. In multi-objective optimization problems, there are often conflicts between the optimization objectives, and it is usually impossible to find a solution that can achieve the best of all objectives at the same time. The Pareto optimal solution set is to find a set of solutions that are both non-inferior and feasible in the multi-objective optimization problem. Each solution in the set represents the best possible compromise between the objectives, that is, each solution in the Pareto optimal solution set is not dominated by any other solution in the multi-objective optimization problem. The Pareto frontier is the mapping of the Pareto optimal solution set in the objective function space, showing the best feasible trade-off between the various objectives. The Pareto frontier is shown as follows Fig.15 As shown (taking dual-objective optimization as an example).

[0086] The idea of ​​the particle swarm optimization algorithm is to use the sharing of search information among individuals in a flock of birds to guide the entire flock to move closer to the location with the most known food in the flock while retaining the diversity information of the individuals themselves, and gradually find the optimal solution in space by continuously updating individual information and group information.

[0087] By abstracting the individuals in the flock into “particles” without volume and mass, during the iteration process, the update of the particle position is affected by both the particle itself and the particle group information. The schematic diagram of the position update is shown in the figure. Fig.14 As shown, the particle position update mechanism is as follows: ; Where: , are the velocities of the i-th particle at the d-step and the d-1-step respectively; ω is the inertia weight of the velocity; c1 is the individual learning factor of the particle; is the best position that the ith particle has passed by until the dth iteration; c2 is the social learning factor of the particle; r1 and r2 are two random numbers between 0 and 1 respectively; is the best position that all particles have passed by until the dth iteration; , are the positions of the i-th particle at the d-th iteration and the d-1-th iteration respectively.

[0088] See also Fig.16 , Fig.16 Schematic diagram of particle position update mechanism provided in the embodiment of the present application. Fig.16 As shown in the figure, the speed update of a single particle consists of three parts: speed inertia, self-cognition and social cognition. The speed inertia part is composed of the inertia weight and the particle's previous movement speed, which reflects the particle's trust in its previous movement and is used to balance the particle's ability to globally optimize and locally optimize. The self-cognition part can be understood as the distance and direction between the particle's current position and its own historical best position, which reflects the particle's memory and preference for its own historical best position, allowing the particle to move based on its own experience, increasing the algorithm's exploration ability and ability to jump out of the local optimal solution, and helping to explore the solution space more comprehensively. The social cognition part can be understood as the distance and direction between the particle's current position and the particle group's historical best position, which reflects the information sharing between particles, allowing the entire group to move to the discovered best area, speeding up the search process for the global optimal solution, and avoiding the algorithm from falling into the local field of view of individual particles, thereby increasing the possibility of finding the global optimal solution.

[0089] For details, please refer to Fig.17 , Fig.17Flowchart of the particle swarm algorithm provided in the embodiment of the present application. For multi-objective optimization problems, Cello et al. extended the particle swarm algorithm to multiple objectives and proposed a multi-objective particle swarm optimization algorithm (MOPSO) based on the external archive idea and the Pareto dominance principle in 2002. The multi-objective particle swarm optimization algorithm has the advantages of wide application range, simple optimization structure, high efficiency and high speed, and is widely used. It is one of the most classic multi-objective optimization algorithms.

[0090] The biggest challenge in multi-objective problems is how to select the "leader" who guides the whole situation. The multi-objective particle swarm optimization algorithm inherits the updating method of particle velocity and position in the particle swarm optimization algorithm, and introduces the idea of ​​external archiving on this basis. A certain number of non-dominated solutions will be stored in the external archive unit. During the iteration process, the dominance of the new non-dominated solution is compared with the solution in the archive to determine whether the new non-dominated solution enters the archive. The grid mechanism is used to estimate the grid congestion. By calculating the maximum and minimum values ​​of each objective function on the non-dominated solution, and then dividing the maximum and minimum value intervals into N equal parts, each solution is arranged in a different grid. The number of solutions in the grid represents the congestion of the grid. When the archive overflows, the grid mechanism will re-divide the grid space, find the most crowded section, delete a certain number of solutions to maintain the archive size, and maintain the diversity of solutions. When selecting the "leader particle" that guides the whole situation, first use the number of particles in the grid as the fitness value, take the inverse of the fitness value, and then use the roulette method to select the grid. The flowchart of the multi-objective particle swarm algorithm is as follows Fig.18 shown.

[0091] As an example, this application constructs a TBM excavation parameter optimization model based on the multi-objective particle swarm optimization algorithm based on the above explanation and the MLP neural network rock-machine parameter mapping model. The main framework of the multi-objective particle swarm algorithm is consistent with the basic particle swarm framework, and an external archiving mechanism and a global optimal solution update strategy are added on this basis. Specifically, first, the input parameters of the MLP neural network are divided into two parts: a unique hot encoding and a numerical value. The unique hot encoding part represents the surrounding rock grade, and the uniaxial compressive strength of the rock, the surrounding rock integrity coefficient and the particle swarm position (thrust, speed) together constitute the numerical part. The penetration, torque and single-knife rock breaking volume of the corresponding particle swarm position are predicted through the MLP neural network. Secondly, the feasible domain is processed and the particle position update mechanism is performed.

[0092] Exemplarily, the constraints can be converted into feasible domains of decision variables, which can facilitate changing the constraint processing strategy in the multi-objective particle physics algorithm.

[0093] Specifically, the present application proposes a method for processing constraints under the framework of a multi-objective particle swarm algorithm: first, a polygonal feasible domain range is constructed through the shapely library based on the coordinates of the feasible domain vertices. When the particle swarm position is initialized, the random initialization position of the particle swarm is limited to the maximum value range of the feasible domain boundary. At the same time, the shapely library's built-in function is used to determine whether the position of a single particle is within the polygon (including the boundary line). If the particle position is not within the feasible domain, the particle is reinitialized until the initial positions of all particles are within the feasible domain; when the basic particle swarm method is used to update the particle swarm position, if the updated position of the particle is outside the feasible domain, the following process is used to correct the deviation: first, connect the original position of the particle with the updated position; if the connecting line has only one intersection with the feasible domain boundary, the final particle position is corrected to the intersection; if the connecting line has multiple intersections with the feasible domain boundary, the final particle position is corrected to the intersection closest to the updated position.

[0094] The above correction method ensures that the particle maintains its original direction of movement during the movement, thereby ensuring the effectiveness of the search. Please refer to Fig.19 , Fig.19 Three situations when particle positions are updated are reproduced. Fig.19 The yellow range is the feasible range: Fig.19 Figure a in the figure shows that the particle updates its position in the feasible domain. The blue dot indicates the position of the particle before the update, and the red dot indicates the position of the particle after the particle is updated according to the particle position update method in the basic particle swarm algorithm, which is also the final position of the particle. Fig.19 Figure b and Fig.19 Figure c in the figure indicates that the particle position is outside the feasible domain after being updated. The blue dot indicates the position of the particle before being updated. The green dot is the position of the particle after being updated according to the particle position update method in the basic particle swarm algorithm. The red dot is the position after correction, which is also the final position of the particle.

[0095] As an example, see Fig. 20 , Fig. 20 The multi-objective particle swarm optimization algorithm parameter setting value table provided in the embodiment of the present application can be based on Fig. 20 The setting values ​​in are used to set the relevant parameters of the multi-objective particle swarm optimization algorithm.

[0096] S104. Control the excavation equipment to perform intelligent control excavation work according to the multiple excavation control parameters.

[0097] Optionally, the method also includes: obtaining multiple verification excavation data of excavation equipment under intelligent control excavation work and manual control excavation work in multiple sections with different surrounding rock types, the multiple verification excavation data including multiple excavation efficiency data and tool life data; determining multiple efficiencies of the intelligent control excavation work in multiple sections with different surrounding rock types and multiple efficiencies of the manual control excavation work in multiple sections with different surrounding rock types based on comparison between the multiple verification excavation data of excavation equipment under intelligent control excavation work and manual control excavation work in multiple sections with different surrounding rock types; determining the efficiency improvement of the intelligent control excavation work relative to the manual control excavation work based on the multiple efficiencies of the intelligent control excavation work in multiple sections with different surrounding rock types and the multiple efficiencies of the manual control excavation work in multiple sections with different surrounding rock types.

[0098] Here, in order to verify the effect of the intelligent auxiliary excavation system in this paper, three test sections are set up in the range of K116+442~K117+229 of the XE tunnel VIII section to compare the excavation efficiency of the driver when it is controlled based on subjective experience and when it adopts the suggestions of the auxiliary excavation system, and at the same time test the accuracy of the surrounding rock perception system. The test section is located in the low mountain and hilly area of ​​erosional structure, with slightly undulating and open terrain. The tunnel is buried at a depth of 63~142m. The rock type is mainly granite gneiss and biotite gneiss. The quartz content in the rock is 20~30%. The rock is hard and relatively complete. The surrounding rock category is mainly Ⅲa and Ⅲb, such as Fig.21 shown.

[0099] Three sections of type II, IIIa and IIIb surrounding rock were selected from K116+442 to K117+229, and each section was approximately divided into two subsections of equal length for comparative experiments. Comparative tests of the same type of surrounding rock in continuous sections largely avoid interference with the test results caused by large differences in surrounding rock parameters. The test sections are numbered according to the surrounding rock type and the control method of tunneling parameters. Specific information is as follows: Fig. 22 As shown in Figure 1-1, 2-1, and 3-1, the TBM driver selected the excavation parameters based on his subjective experience, which served as the control sections for the verification of the excavation parameter auxiliary decision-making method. Sections 1-2, 2-2, and 3-2 were used to verify the effect of the auxiliary excavation system. In addition, the surrounding rock parameter perception model was compared and verified in these three sections. The program steps are as follows: Fig.23 shown.

[0100] The final measured advancement speed PR and the measured average single-cut rock breaking volume Hf obtained in each excavation cycle in the test were calculated, and the results were weighted averaged according to the actual advance of each cycle. Then, the results of the two test sections belonging to the same level of surrounding rock were plotted in the same figure. The results are shown in Figures 25 to 32The blue dotted line and thick line on the left side of the figure represent the measured value and average value in the test section using intelligent auxiliary decision-making, respectively, and the yellow part on the right side represents the data of the section using manual control. The average excavation efficiency comparison between the intelligent auxiliary decision-making test section and the manual control test section in various surrounding rock types is shown in Fig.24 .from Fig.24 It can be seen that in the test sections of various levels of surrounding rock, the excavation efficiency achieved by using auxiliary decision-making is generally better than that of manual control. Compared with manual control, the use of auxiliary decision-making increased the advancement speed by 4.14%, 8.75% and 13.92% in Class II, Class IIIa and Class IIIb surrounding rock, respectively. It is not difficult to find that in Class II surrounding rock, the advancement speed is more difficult to increase, while the advancement rate in Class IIIb surrounding rock is more significantly improved. It can be seen that the auxiliary decision-making agent tends to use a higher thrust than the driver decision in Class IIIb surrounding rock, thereby achieving an increase in advancement speed. The average single-cutter rock breaking amount in Class II, Class IIIa and Class IIIb surrounding rock increased by 6.51%, 5.34% and 4.16%, respectively. It can be seen that auxiliary decision-making can play a good role in optimizing tool wear in Class II surrounding rock. The above results show that the thrust and cutter head speed given by the intelligent auxiliary decision-making method in this paper can improve the TBM excavation speed and tool life to a certain extent, and the application effect is good.

[0101] The intelligent control method for tunnel excavation using TBM equipment provided in the present application can determine the rock mechanical parameters based on multiple excavation data, and then intelligently control the excavation equipment based on multiple models or algorithms, which can not only improve the excavation speed of the TBM equipment, but also improve the tool life of the TBM equipment.

[0102] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An intelligent control method for tunnel excavation using TBM equipment, characterized in that: The method comprises: Acquire a plurality of excavation data, and determine a plurality of characteristic excavation data related to rock mass information from the plurality of excavation data; Inputting a plurality of characteristic excavation data into a target TBM surrounding rock perception algorithm to obtain a plurality of rock mass mechanical parameters, wherein the target TBM surrounding rock perception algorithm is determined based on a plurality of pre-trained TBM surrounding rock perception models; Input multiple rock mass mechanical parameters into the target excavation parameter decision model to obtain multiple excavation control parameters; According to the multiple excavation control parameters, the excavation equipment is controlled to perform intelligent control excavation work.

2. The method according to claim 1, characterized in that Determine multiple characteristic excavation data through the following steps: Determine the correlation between the rock mass quality grade and multiple field excavation indexes based on multiple historical excavation data, wherein the field excavation index is the ratio of the thrust of a single cutter of the TBM equipment to the penetration rate; According to the correlation between the rock mass quality grade and a plurality of on-site excavation indexes, a plurality of characteristic excavation data are determined, wherein the characteristic excavation data include single cutter thrust, cutter head torque, cutter head rotation speed and penetration rate.

3. The method according to claim 2, characterized in that The TBM surrounding rock perception algorithm includes a least squares regression sub-model, a BP neural network sub-model and a vector machine sub-model. Among them, the target TBM surrounding rock perception algorithm is determined by the following formula: ； Among them, w1 is the weight value of the least squares regression sub-model, w2 is the weight value of the BP neural network sub-model, and w3 is the weight value of the vector machine sub-model. are multiple rock mass mechanical parameters calculated by the least squares regression sub-model, Multiple rock mass mechanical parameters calculated by the BP neural network sub-model, Several rock mechanical parameters calculated for the vector machine submodel.

4. The method according to claim 3, characterized in that The least squares regression sub-model is trained by the following steps: Based on the least squares model, a least squares regression sub-model is established according to multiple rock mass mechanical parameters to be measured and multiple characteristic excavation data; Inputting a plurality of historical excavation data into an initial least squares regression sub-model to obtain a first loss function value of the initial least squares regression sub-model; According to the obtained first loss function value, the initial least squares regression sub-model is trained until the first loss function value of the initial least squares regression sub-model converges to a minimum value; The target least squares regression submodel is determined according to the least squares regression submodel when the first loss function value of the initial least squares regression submodel converges to the minimum value.

5. The method according to claim 4, characterized in that The multiple rock mass mechanical parameters to be measured include the uniaxial compressive strength of rock and the number of rock mass joints per unit volume. Wherein, the target least squares regression sub-model includes the following formula: ； ； Among them, F n is the thrust of a single knife, p is the penetration, n is the rotation speed of the cutter head, T is the torque of the cutter head, is the uniaxial compressive strength of rock; is the number of joints per unit volume of rock mass.

6. The method according to claim 3, characterized in that Determine the BP neural network sub-model through the following steps: Using the learning mechanism of signal forward propagation and error reverse regulation, a BP network model for processing nonlinear information is built through iterative learning; The GA algorithm is used to optimize the hyperparameters of the BP network model, thereby determining the target BP network model.

7. The method according to claim 3, characterized in that Determine the vector machine submodel by following the steps below: Determine multiple initial vector machine sub-models based on multiple historical excavation data; Determine a vector machine sub-model according to the least squares support vector machine function, the Lagrangian function and the kernel function; The GA algorithm is used to optimize the hyperparameters of the vector machine sub-model to determine the target vector machine sub-model.

8. The method according to claim 3, characterized in that The weight values ​​of the least squares regression sub-model, the BP neural network sub-model, and the vector machine sub-model are determined by the following steps: generating a plurality of evidence sets according to the plurality of excavation data; Based on the multiple evidence sets, the weight values ​​of the least squares regression sub-model, the BP neural network sub-model and the first intermediate weight value of the vector machine sub-model are calculated with the goal of minimizing the mean square error of the error, and a BPA function is constructed according to the weight values ​​of the least squares regression sub-model, the BP neural network sub-model and the first intermediate weight value of the vector machine sub-model; Determining the credibility of each evidence set according to the distance of the BPA function between different evidence sets; The BPA function is modified according to the credibility of each evidence set, and the modified BPA functions on different evidence sets are synthesized according to the Dempster synthesis rule to obtain the weight values ​​of the least squares regression sub-model, the BP neural network sub-model and the vector machine sub-model.

9. The method according to claim 1, characterized in that: The target excavation parameter decision model is determined through the following steps: Based on the MLP neural network rock-machine parameter mapping model, a TBM excavation parameter optimization model based on the multi-objective particle swarm optimization algorithm is constructed; Converting a plurality of pre-set constraints into a feasible domain of decision variables to change the constraint processing strategy in the multi-objective particle swarm optimization algorithm; A plurality of relevant parameters of the multi-objective particle swarm optimization algorithm are set to obtain a target excavation parameter decision model.

10. The method according to claim 1, characterized in that The method further comprises: Acquire multiple verified excavation data of excavation equipment under intelligent control excavation work and manual control excavation work in multiple sections with different surrounding rock types, wherein the multiple verified excavation data include multiple excavation efficiency data and tool life data; By comparing multiple verified excavation data of excavation equipment under intelligent control excavation and manual control excavation in multiple sections with different surrounding rock types, multiple efficiencies of intelligent control excavation in multiple sections with different surrounding rock types and multiple efficiencies of manual control excavation in multiple sections with different surrounding rock types are determined; Based on multiple efficiencies of intelligent-controlled tunneling work in multiple sections with different surrounding rock types and multiple efficiencies of manually-controlled tunneling work in multiple sections with different surrounding rock types, the efficiency improvement extent of the intelligent-controlled tunneling work relative to the manually-controlled tunneling work is determined.

Images