Shield tunneling machine axis prediction control method and system, electronic equipment and storage medium

Abstract

The invention discloses a shield tunneling machine axis predictive control method and system, electronic equipment and a storage medium, the shield tunneling machine axis predictive control method is based on data driving modeling, system dynamic characteristics contained in shield tunneling axis control related data are fully excavated, the predictive control precision is high, and the predictive control precision is high. A GRU (Gated Recurrent Neural Network) model and an ARX (Autoregressive Regressive) model with exogenous variables are combined to construct a guide deviation dynamic characteristic model to predict the guide deviation, the advantage that the GRU is good at processing a long-interval and long-delay time sequence is fully utilized, and the correlation characteristics before and after the shield guide deviation data time sequence are fully excavated; the model not only has the global nonlinear description capability, but also has the capability of describing the time sequence correlation characteristics of the system, the dynamic characteristics of the system are more comprehensively described, and the model is a nonlinear model which is more in line with the guiding actual working characteristics of the shield tunneling machine, and is higher in generalization capability and higher in precision.

Description

Technical Field

[0001] The present invention relates to the technical field of shield machines, and in particular, to a method and system for predicting and controlling the axis of a shield machine, an electronic device, and a computer-readable storage medium. Background Art

[0002] A shield machine is a special construction equipment for tunnel boring projects, with functions such as cutter head excavation, hydraulic propulsion, segment assembly, muck transportation, and guiding measurement. It is a complex system integrating advanced technologies such as machinery, electricity, hydraulics, and optics. Among them, the consistency between the actual formed axis of the tunnel and the designed axis of the tunnel is the key point of the construction quality control of the shield tunnel. During the construction process of the shield machine, due to the influence of uncertain factors such as the construction environment and the operating conditions of the shield machine, it is inevitable that the actual operating route of the shield machine deviates from the designed axis. If the axis control is improper, it is easy to cause the shield tail to squeeze the segments, resulting in problems such as segment misalignment, damage, and leakage, and may even cause ground collapse or uplift, leading to major accidents. Therefore, high-precision control of the shield machine axis is of great significance for ensuring the safety and efficiency of tunnel construction.

[0003] At present, the axis control of shield machines mostly adopts manual control. The shield driver adjusts the parameters of the shield propulsion system based on the guiding axis deviation measured by the guiding system and relying on manual experience, so as to realize the shield tunneling following the designed axis of the tunnel. However, due to the characteristics of the shield machine such as large inertia, time lag, non-linearity, and strong coupling, there is a large lag in manual control, which is easy to cause the snake-like movement of the shield. At the same time, the stratum characteristics in different sections or different segments in the same section may change greatly, which requires a high level of operation for the driver. In addition, the Chinese patent application with the publication number CN116066123A discloses an automatic tracking control method for the shield tunneling trajectory based on model predictive control. It uses a state space model based on the shield tunneling dynamics model as the prediction model, takes the cylinder thrust of the shield propulsion system as the control quantity, and realizes the automatic tracking control of the trajectory by calculating the optimal control sequence with the smallest error between the predicted shield pose state and the reference shield pose state based on the designed axis of the tunnel, and applying the first component of the sequence as the optimal input cylinder thrust control quantity at the current moment to the shield propulsion system; however, this algorithm uses a mechanism model based on dynamics, which is generally established under ideal conditions. Due to the complexity and diversity of shield construction conditions, as well as the differences in construction requirements for different tunnels, the accuracy of the mechanism model is limited and the limitations are great. Summary of the Invention

[0004] The present invention provides a method and system for predicting and controlling the axis of a shield machine, an electronic device, and a computer-readable storage medium, which improve the prediction control accuracy and can control the shield machine to move forward along the designed axis of the tunnel.

[0005] According to an aspect of the present invention, a shield machine axis prediction control method is provided, including the following:

[0006] Obtain tunneling data and perform normalization processing on it. Among them, the tunneling data includes the guiding deviation between the shield head and the shield tail, the guiding pressure difference of the propulsion cylinders, the average propulsion speed, and the total propulsion force value;

[0007] Construct a guiding deviation dynamic characteristic model based on the GRU network model and the ARX model, and use the obtained tunneling data to iteratively train the guiding deviation dynamic characteristic model until the iteration termination condition is met, and save the trained guiding deviation dynamic characteristic model;

[0008] Construct a guiding deviation prediction controller based on the trained guiding deviation dynamic characteristic model, collect real-time tunneling data, perform normalization processing on it, and then input it into the guiding deviation prediction controller. Optimize and solve the prediction control objective function of the guiding deviation prediction controller to obtain the optimal solution of the guiding pressure difference of the propulsion cylinders.

[0009] Further, the expression of the guiding deviation dynamic characteristic model is:

[0010]

[0011] Among them, y(t) represents the output of the model, that is, the guiding deviation vector between the shield head and the shield tail, t represents the current moment, p represents the order of the model output, q represents the order of the model input; s represents the order of the measurable disturbance of the model, f represents the delay of the model input, g represents the delay of the measurable disturbance of the model, u represents the guiding pressure difference of the propulsion cylinders, d represents the measurable disturbance vector of the model, that is, the average propulsion speed and the total propulsion force, ε(t) represents the Gaussian white noise signal, a 0,t-1 represents the coefficient matrix of the model bias, a 11，0,t-1 , a 21，0,t-1 is an element in a 0,t-1 , a i,t-1 represents the coefficient matrix of the model output, a 11,i,t-1 , a 12,i,t-1 , a 21,i,t-1 , a 22,i,t-1 is an element in a i,t-1 , b j,t-1 represents the coefficient matrix of the model input, b 11,j,t-1 , b 21,j,t-1 is an element in b j,t-1 , c k,t-1 represents the coefficient matrix of the measurable disturbance of the model, c 11,k,t-1 , c 12,k,t-1 , c 21,k,t-1 , c 22,k,t-1 is an element in c k,t-1 .

[0012] Furthermore, the guiding deviation dynamic characteristic model includes a GRU layer, a fully connected layer, and an ARX layer. Among them, the output result of each time step of the last layer in the GRU layer is input into the ARX layer after constructing a state-dependent vector through the fully connected layer, and the ARX layer fits the regression coefficient matrix of the ARX layer through the state-dependent vector.

[0013] Furthermore, the state space model of the guiding deviation prediction controller is as follows:

[0014]

[0015] Among them, y(t) represents the output variable, x(t) represents the state variable, A t , B t , Φ t , C are all coefficient matrices, a 1,t represents the coefficient matrix when the model output order at time t is 1, a 2,t represents the coefficient matrix when the model output order at time t is 2, a kn-1,t represents the coefficient matrix when the model output order at time t is k n - 1, a kn,t represents the coefficient matrix when the model output order at time t is k n , b 1,t represents the coefficient matrix when the model input order at time t is 1, b 2,t represents the coefficient matrix when the model input order at time t is 2, represents the coefficient matrix when the model input order at time t is k n - 1, represents the coefficient matrix when the model input order at time t is k n , φ(t) represents the sum of the bias of the model at time t and the disturbances of each order of the model, Ξ(t + 1) represents the matrix of white noise at time t + 1, ε(t + 1) represents the white noise at time t + 1, T represents the transpose of the vector, and I represents the identity matrix.

[0016] Furthermore, the predictive control objective function is:

[0017]

[0018] Among them, minJ represents the predictive control objective function, represents 's 2-norm, represents 's 2-norm, represents the future N of the guiding deviation y obtained based on the state space model yStep forward prediction vector, y r Represents the desired vector of steering deviation, y min Represents the lower limit of steering deviation, y max Represents the upper limit of steering deviation, u m i n Represents the lower limit of the steering pressure difference of the propulsion cylinder, u max Represents the upper limit of the steering pressure difference of the propulsion cylinder Represents the control vector obtained by solving at time t Increment vector of, Δu min Represents the lower limit of the increment of the steering pressure difference of the propulsion cylinder, Δu max Represents the upper limit of the increment of the steering pressure difference of the propulsion cylinder Is the control vector obtained by solving at time t, R 1 (t) represents the weighting coefficient

[0019] Furthermore, the weighting coefficient is adaptively adjusted based on the following formula:

[0020]

[0021] Where, β represents the adjustment coefficient, N g Represents the observation window length, R min Represents the minimum value of the weighting coefficient, R max Represents the maximum value of the weighting coefficient, y(t - i 3 ) represents the steering deviation at time t - i 3 At time t - i 3 -1) represents the steering deviation at time t - i 3 -1

[0022] Furthermore, it also includes the following content:

[0023] Judge whether to update the model based on the mean square error of the steering deviation of the steering deviation dynamic characteristic model

[0024] In addition, the present invention also provides a shield machine axis prediction control system, including:

[0025] Tunneling data acquisition module, used to acquire tunneling data and perform normalization processing on it, where the tunneling data includes the steering deviation of the shield head and the shield tail, the steering pressure difference of the propulsion cylinder, the average propulsion speed, and the total propulsion force value;

[0026] Model construction and training module, used to construct a steering deviation dynamic characteristic model based on the GRU network model and the ARX model, and use the acquired tunneling data to iteratively train the steering deviation dynamic characteristic model until the iteration termination condition is met, and save the trained steering deviation dynamic characteristic model;

[0027] The guiding deviation prediction control module is used to construct a guiding deviation prediction controller based on the trained dynamic characteristics model of the guiding deviation, collect real-time tunneling data, normalize it, and then input it into the guiding deviation prediction controller. The prediction control objective function of the guiding deviation prediction controller is optimized to obtain the optimal solution of the guiding pressure difference of the propulsion cylinder.

[0028] In addition, the present invention also provides an electronic device, including a processor and a memory. A computer program is stored in the memory. The processor is used to execute the steps of the method described above by calling the computer program stored in the memory.

[0029] In addition, the present invention also provides a computer-readable storage medium for storing a computer program for predicting and controlling the axis of a shield machine. When the computer program runs on a computer, it executes the steps of the method described above.

[0030] The present invention has the following beneficial effects:

[0031] In the axis prediction control method of the shield machine of the present invention, first, tunneling data related to axis control during shield tunneling is collected. Then, a dynamic characteristics model of the guiding deviation is constructed based on the GRU network model and the ARX model, and the dynamic characteristics model of the guiding deviation is iteratively trained using the tunneling data. Then, a guiding deviation prediction controller is constructed based on the trained dynamic characteristics model of the guiding deviation. After collecting real-time tunneling data, it is input into the guiding deviation prediction controller for optimization to obtain the optimal solution of the guiding pressure difference of the propulsion cylinder, so as to control the shield machine to move forward along the tunnel design axis. The present invention is based on data-driven modeling, fully excavates the system dynamic characteristics contained in the data related to the shield tunneling axis control, has high prediction control accuracy, and combines the gated recurrent neural network GRU model and the autoregressive model with exogenous variables ARX model to construct a dynamic characteristics model of the guiding deviation to predict the guiding deviation. It makes full use of the advantage of the gated recurrent neural network GRU in dealing with long-interval and long-delay time series, and fully excavates the correlation characteristics before and after the time series of the shield guiding deviation data, so that the model not only has the ability to describe global non-linearity, but also has the ability to describe the time-series correlation characteristics of the system. The system dynamic characteristics are described more comprehensively, and it is a non-linear model that better fits the actual working characteristics of the shield machine guiding. The model has stronger generalization ability and higher accuracy.

[0032] In addition, the axis prediction control system of the shield machine of the present invention also has the above advantages.

[0033] In addition to the purposes, features and advantages described above, the present invention has other purposes, features and advantages. The present invention will be further described in detail below with reference to the drawings. Description of the Drawings

[0034] The accompanying drawings, which form a part of this application, are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0035] Figure 1 is a schematic flowchart of the shield machine axis prediction control method according to a preferred embodiment of this application;

[0036] Figure 2 is a partial network structure schematic diagram of the guiding deviation dynamic characteristic model according to a preferred embodiment of this application;

[0037] Figure 3 is Figure 1 a sub - flowchart of step S2 in

[0038] Figure 4 is Figure 1 another sub - flowchart of step S2 in

[0039] Figure 5 is another schematic flowchart of the shield machine axis prediction control method according to a preferred embodiment of this application;

[0040] Figure 6 is a module structure schematic diagram of the shield machine axis prediction control system according to another embodiment of this application. Specific Embodiments

[0041] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the accompanying drawings and combine with the embodiments to detail this application.

[0042] Referring to Figure 1 , a preferred embodiment of this application provides a shield machine axis prediction control method, including the following:

[0043] Step S1: Obtain tunneling data and perform normalization processing on it. Among them, the tunneling data includes the guiding deviations of the shield head and the shield tail, the guiding pressure difference of the propulsion cylinders, the average propulsion speed, and the total propulsion force value;

[0044] Step S2: Construct a guiding deviation dynamic characteristic model based on the GRU network model and the ARX model, and use the obtained tunneling data to iteratively train the guiding deviation dynamic characteristic model until the iterative termination condition is met, and save the trained guiding deviation dynamic characteristic model;

[0045] Step S3: Based on the trained dynamic characteristics model of the guiding deviation, construct a guiding deviation prediction controller. After collecting the real-time tunneling data and normalizing it, input it into the guiding deviation prediction controller, and optimize and solve the prediction control objective function of the guiding deviation prediction controller to obtain the optimal solution of the guiding pressure difference of the propulsion cylinder.

[0046] It can be understood that for the axis prediction control method of the shield machine in this embodiment, first collect the tunneling data related to the axis control during the shield tunneling process, then construct a dynamic characteristics model of the guiding deviation based on the GRU network model and the ARX model, and use the tunneling data to iteratively train the dynamic characteristics model of the guiding deviation. Then, based on the trained dynamic characteristics model of the guiding deviation, construct a guiding deviation prediction controller. After collecting the real-time tunneling data, input it into the guiding deviation prediction controller for optimization and solution, and the optimal solution of the guiding pressure difference of the propulsion cylinder can be obtained, so as to control the shield machine to move forward along the tunnel design axis. The present invention is based on data-driven modeling, fully excavates the system dynamic characteristics contained in the data related to the shield tunneling axis control, has high prediction control accuracy, and combines the gated recurrent neural network GRU model and the autoregressive with exogenous inputs ARX model to construct a dynamic characteristics model of the guiding deviation to predict the guiding deviation. It makes full use of the advantage of the gated recurrent neural network GRU in dealing with long intervals and long-delay time series, and fully excavates the correlation characteristics before and after the time series of the shield guiding deviation data, so that the model not only has the ability to describe global non-linearity, but also has the ability to describe the time-series correlation characteristics of the system. The system dynamic characteristics are described more comprehensively, and it is a non-linear model that better fits the actual working characteristics of the shield machine guiding. The model has stronger generalization ability and higher accuracy.

[0047] It can be understood that in step S1, after importing the tunnel design axis data into the shield machine guidance system, when the shield machine is tunneling, the shield machine guidance system can measure and calculate the horizontal guidance coordinates of the shield head and shield tail of the shield machine, and the vertical guidance coordinates of the shield head and shield tail, so as to output the horizontal guidance deviation and vertical guidance deviation of the shield head and shield tail. During the tunneling process, the guidance pressure difference, average propulsion speed, and total propulsion force value of the propulsion cylinders can also be measured and calculated in real time. Among them, the guidance pressure difference of the propulsion cylinders includes the horizontal guidance pressure difference and the vertical guidance pressure difference. The horizontal guidance pressure difference is related to the horizontal guidance deviation of the shield head and shield tail, and the vertical guidance pressure difference is related to the vertical guidance deviation of the shield head and shield tail. Set the sampling period of the tunneling data to 1 second, and continuously collect the tunneling data until the preset quantity requirement is met. In addition, since the actual tunneling process of the shield machine is intermittent, and there is a large amount of construction data with tunneling stopped in the tunneling data, taking the shield machine propulsion working state signal as the judgment criterion, the data segments in the tunneling stopped state can be excluded. Among them, the specific normalization processing method is the maximum-minimum value method, which converts all the tunneling data to between [0, 1]. The conversion formula is as follows:

[0048]

[0049] Among them, Y represents the normalized tunneling data, X represents the tunneling data before normalization processing, such as the horizontal guidance deviation and vertical guidance deviation of the shield head and shield tail, the horizontal guidance pressure difference of the propulsion cylinders, the vertical guidance pressure difference of the propulsion cylinders, the average propulsion speed, and the total propulsion force value. X max represents the maximum value of the tunneling data, and X min represents the minimum value of the tunneling data. Of course, in other embodiments of the present invention, other existing normalization processing methods can also be used, such as the Z-score normalization method, the maximum value normalization method, etc.

[0050] It can be understood that in step S2, a dynamic characteristic model of the guidance deviation is constructed based on the GRU network model and the ARX model, and the parameters of the dynamic characteristic model of the guidance deviation are iteratively trained using the obtained tunneling data until the iteration termination condition is met, and then the trained dynamic characteristic model of the guidance deviation is saved. Among them, the tunneling data collection in step S1 is performed at the edge side, the construction and training of the model in step S2 are performed in the cloud, the data collected at the edge side will be uploaded to the cloud, the model is trained in the cloud, and after the model training is completed, the cloud will transmit the trained model parameters to the edge side to facilitate edge computing.

[0051] Among them, the present invention constructs a dynamic characteristic model of the steering deviation of the shield machine, taking the horizontal or vertical guiding pressure difference of the propulsion cylinder as the input variable, the average propulsion speed and the total propulsion force value as measurable disturbances, and the horizontal or vertical guiding deviation of the shield head and the shield tail as the output variable and the state variable of the model. Among them, a partial network structure of the dynamic characteristic model of the steering deviation is as shown in Figure 2 shown. It can be understood that since the horizontal guiding pressure difference of the propulsion cylinder is related to the horizontal guiding deviation of the shield head and the shield tail, and the vertical guiding pressure difference is related to the vertical guiding deviation of the shield head and the shield tail, it is necessary to construct two dynamic characteristic models of the steering deviation, namely the dynamic characteristic model of the horizontal steering deviation and the dynamic characteristic model of the vertical steering deviation. The dynamic characteristic model of the horizontal steering deviation takes the horizontal guiding pressure difference of the propulsion cylinder as the input variable, the average propulsion speed and the total propulsion force value as measurable disturbances, and the horizontal guiding deviation of the shield head and the shield tail as the output variable and the state variable of the model, while the dynamic characteristic model of the vertical steering deviation takes the vertical guiding pressure difference of the propulsion cylinder as the input variable, the average propulsion speed and the total propulsion force value as measurable disturbances, and the vertical guiding deviation of the shield head and the shield tail as the output variable and the state variable of the model.

[0052] Among them, the expression of the dynamic characteristic model of the steering deviation is:

[0053]

[0054] Among them, y(t) represents the output of the model, that is, the horizontal or vertical guiding deviation vector of the shield head and the shield tail, t represents the current moment, p represents the order of the model output, q represents the order of the model input; s represents the order of the measurable disturbance of the model, f represents the delay of the model input, g represents the delay of the measurable disturbance of the model, u represents the horizontal or vertical guiding pressure difference of the propulsion cylinder, d represents the measurable disturbance vector of the model, that is, the average propulsion speed and the total propulsion force, ε(t) represents the Gaussian white noise signal, a 0,t-1 represents the coefficient matrix of the model bias, a 11，0,t-1 , a 21，0,t-1 are the elements in a 0,t-1 , a i,t-1 represents the coefficient matrix of the model output, a 11,i,t-1 , a 12,i,t-1 , a 21,i,t-1 , a 22,i,t-1 are the elements in a i,t-1 , b j,t-1 represents the coefficient matrix of the model input, b 11,j,t-1 , b 21,j,t-1 are the elements in b j,t-1 , c k,t-1 represents the coefficient matrix of the measurable disturbance of the model, c 11,k,t-1 , c 12,k,t-1 , c21,k,t-1 、c 22,k,t-1 c k,t-1 The elements in 0,t-1 、a i,t-1 , b j,t-1 and c k,t-1 It is calculated by the state vector through the GRU network.

[0055] Wherein, the guiding deviation dynamic characteristic model includes a GRU layer, a fully connected layer and an ARX layer, wherein the output result of each time step of the last layer in the GRU layer is input into the ARX layer after constructing a state dependency vector through the fully connected layer, and the ARX layer fits the regression coefficient matrix of the ARX layer through the state dependency vector. It can be understood that the present invention selects the horizontal or vertical guiding deviation of the shield head and the shield tail as the state variable, and extracts the nonlinear characteristics and time series related characteristics in the shield machine guiding deviation time series data through the GRU network. At the same time, in order to ensure more adequate use of the information of each time step, the model no longer only collects the output of the last time step of the last layer of the GRU as the feature vector, but collects the output result of each time step of the last layer of the GRU as the feature vector, and then obtains the state dependency vector through the fully connected layer, and finally fits the regression coefficient matrix of the ARX model through the state dependency vector. In addition, considering that the horizontal guidance deviation dynamic characteristic model and the vertical guidance deviation dynamic characteristic model have two outputs, for example, the output of the horizontal guidance deviation dynamic characteristic model is the horizontal guidance deviation of the shield head and the horizontal guidance deviation of the shield tail, and the output of the vertical guidance deviation dynamic characteristic model is the vertical guidance deviation of the shield head and the vertical guidance deviation of the shield tail, so it is necessary to follow the Figure 2 The network structures are constructed, which are represented as Model1 and Model2 respectively. Among them, Model1 takes the horizontal guidance deviation of the shield head as the output, the horizontal guidance deviation of the shield head and the shield tail as the state variable, the horizontal guidance pressure difference of the propulsion cylinder as the input, and the average propulsion speed and the total propulsion force as the measurable interference; while Model2 takes the horizontal guidance deviation of the shield tail as the output, the horizontal guidance deviation of the shield head and the shield tail as the state variable, the horizontal guidance pressure difference of the propulsion cylinder as the input, and the average propulsion speed and the total propulsion force as the measurable interference.

[0056] Next, we will take Model 1 as an example to explain the calculation process of the state-dependent vector, as follows:

[0057]

[0058] Among them, m represents the number of GRU network layers, l represents the sequence number of the GRU network layer, and T w Indicates the time step number, n w Represents the order of the state variable, which is also the time step parameter in the GRU network. σ represents the Sigmoid activation function. Represents the calculated value of the reset gate at the \(T\) -th time step of the \(l\) -th layer GRU, w where \( Represents the input state vector at the \(T\) -th time step of the \(l\) -th layer GRU, w where \( Represents the hidden state at the \((T - 1)\) -th time step of the \(l\) -th layer GRU, w where \( Represents the weight matrix of the reset gate in the \(l\) -th layer GRU, Represents the bias matrix of the reset gate in the \(l\) -th layer GRU, Represents the calculated value of the update gate at the \(T\) -th time step of the \(l\) -th layer GRU, w where \( Represents the weight matrix of the update gate in the \(l\) -th layer GRU, Represents the bias matrix of the update gate in the \(l\) -th layer GRU, Represents the important state information to be saved at the \(T\) -th time step of the \(l\) -th layer GRU. \(\odot\) represents element - wise multiplication, w where \( Represents the set of candidate hidden states at the \(T\) -th time step of the \(l\) -th layer GRU, w where \( Represents the weight matrix of the set of candidate hidden states in the \(l\) -th layer GRU, Represents the bias matrix of the set of candidate hidden states in the \(l\) -th layer GRU, Represents the hidden state at the \(T\) -th time step of the \(l\) -th layer GRU, w where \( Represents the hidden state at the \(T\) -th time step of the \((l - 1)\) -th layer GRU, \(x(t - 1)\) represents the state vector at time \(t - 1\), \(h\) w where \( m Represents the feature vector composed of the hidden states of all time steps of the \(m\) -th layer GRU, Represents the hidden state at the 1 -st time step of the \(m\) -th layer GRU, Represents the hidden state at the 2 -nd time step of the \(m\) -th layer GRU, Represents the hidden state at the \(n\) -th time step of the \(m\) -th layer GRU, w where \( fc Represents the weight of the fully - connected layer, \(b\) fc Represents the bias of the fully - connected layer, Represents the activation function of the fully - connected layer. Here, tanh is selected, \(\theta\) 1,t-1 Represents the state - dependent vector, corresponding to the coefficients \(a\) 11，0,t-1 、\(a\) 11,i,t-1 、\(a\) 12,i,t-1 、\(b\) 11,j,t-1 、\(c\) 11,k,t-1 and \(c\) 12,k,t-1 。

[0059] Similarly, Model2 takes the horizontal guiding deviation of the shield tail as the output, the horizontal guiding deviations of the shield head and the shield tail as state variables, the horizontal guiding pressure difference of the propulsion cylinders as the input, and the average propulsion speed and the total propulsion force value as measurable disturbances. Model2 has the same structure and working principle as Model1. Through Model2, the state-dependent vector coefficients a 21，0,t-1 、a 21,i,t-1 、a 22,i,t-1 、b 21,j,t-1 、c 21,k,t-1 and c 22,k,t-1 of the ARX model can be calculated.

[0060] It can be understood that the guiding deviation dynamic characteristic model of the present invention takes the horizontal or vertical guiding pressure difference of the propulsion cylinders as the input variable, the average propulsion speed and the total propulsion force value as measurable disturbances, and the horizontal or vertical guiding deviations of the shield head and the shield tail as the output variable and the state vector of the model. The gated recurrent neural network GRU is used to fit the regression coefficient matrix of the autoregressive model with exogenous variables ARX, so that the model not only has the ability of global non-linear description, but also has the ability to describe the time series related characteristics of the system. The system dynamic characteristics are described more comprehensively and accurately, more in line with the actual working characteristics of the shield machine guiding, and the model has stronger generalization ability and higher accuracy.

[0061] It can be understood that as Figure 3 shown, in the step S2, the process of iteratively training the guiding deviation dynamic characteristic model by using the obtained tunneling data includes the following contents:

[0062] Step S21: Initialize the model structure parameters;

[0063] Step S22: Take the guiding pressure difference of the propulsion cylinders as the input variable, the average propulsion speed and the total propulsion force value as measurable disturbances, and the guiding deviations of the shield head and the shield tail as the output variable and the state variable, and construct the input layer data of the GRU network model and the ARX model;

[0064] Step S23: Forward calculation;

[0065] Step S24: Backpropagation to update the model parameters;

[0066] Step S25: After changing the model structure parameters, repeat steps S22, S23, and S24, and select the model structure parameters and model parameters with the smallest loss function value as the optimal model.

[0067] Specifically, taking model1 as an example for explanation. The input layer data structure of the GRU network model can be defined as follows:

[0068] Input_GRU = [y(t - n w) T , y(t - n w + 1) T ,..., y(t - 2) T , y(t - 1) T T

[0069] Among them, y represents the horizontal guiding deviation vector of the shield head and shield tail, and n w represents the order of Input_GRU;

[0070] The input layer data structure of the ARX model can be defined as follows:

[0071] Input_ARX = [1, y(t - 1) T ,..., y(t - p) T , u(t - f),..., u(t - f - q + 1),

[0072] d(t - g) T ,..., d(t - g - s + 1) T T ;

[0073] Initialize the model structure parameters p, q, s, f, g, m, nw, and the output dimension n of each layer of GRU.

[0074] Then, construct the input layer data of the GRU network model and the ARX model according to the above two data structures with the horizontally guided deviation of the shield head and shield tail, the horizontal guided pressure difference of the propulsion cylinder, the average propulsion speed, and the total propulsion force value after normalization processing.

[0075] Next, use the data of the Input_GRU structure as the input of the GRU network model, and through calculation, obtain the state-dependent coefficient vector θ 1,t-1 , that is, the coefficient of the ARX model, then the predicted output of the model can be obtained as:

[0076]

[0077] Among them, represents the predicted output value of the horizontal guiding deviation of the shield head.

[0078] Then, construct the loss function e, and continuously update the model parameters through the backpropagation algorithm until the loss function is minimized to obtain a model with higher accuracy. Among them, the expression of the loss function e is:

[0079]

[0080] Among them, e represents the mean square error between the predicted value and the actual value of the horizontal guiding deviation of the shield head, N represents the number of training samples, and y​​1 Represents the actual value of the horizontal guiding deviation of the shield tunneling machine.

[0081] Finally, change the model structure parameters p, q, s, f, g, m, n w and the output dimension n of each layer of GRU, repeat steps S22, S23, S24, after traversing all the model structure parameters, compare the loss function values e under different model structure parameters, and select the model structure and model parameters with the smallest loss function value as the optimal model.

[0082] Optionally, as Figure 4 shown, it also includes the following content:

[0083] Step S26: Optimize the model parameters of the fully connected layer by using the particle swarm optimization algorithm.

[0084] Specifically, the process of optimizing the model parameters of the fully connected layer by using the particle swarm optimization algorithm includes the following content:

[0085] 1), Initialize the particle swarm, randomly assign an initial position and velocity to each particle, and assign the model parameters of the fully connected layer saved in step S25 to one of the particles;

[0086] 2), Calculate the cost function E cost of each particle, where the specific cost function is the prior art and the specific expression is not elaborated here;

[0087] 3), Update the velocity and position of each particle, and the update formula is:

[0088]

[0089]

[0090] where, represents the velocity of the rth particle at the (l d + 1)th iteration, represents the velocity of the rth particle at the lth d iteration, α represents the inertia weight, β 1 and β 2 represent the acceleration constants, rand 1 and rand 2 represent two random numbers between 0 and 1, pbest r represents the best position found by the rth particle so far, gbest represents the best position found by the entire particle swarm so far, represents the position of the rth particle at the (l d + 1)th iteration, is the rth particle at the lthd Position at the next iteration;

[0091] 4), Update the individual and the global optimal solution. For each particle, if the current position is better than the best position encountered before, update its individual optimal value, and at the same time find the minimum cost function E among all the particles cost of the position, and update this position as the global optimal solution;

[0092] 5), Repeat steps 2) to 4) until the maximum number of iterations set value or the cost function predetermined threshold is reached, then the iteration terminates. When the iteration terminates, output the global optimal solution, and save this solution as the optimal solution of the model parameters.

[0093] It can be understood that the guiding deviation dynamic characteristic model of the present invention collects the output results of each time step of the last layer of the GRU as the feature vector, and uses the particle swarm optimization algorithm PSO to optimize the model parameters of the fully connected layer between the feature vector and the state-dependent vector when constructing the model. This not only ensures that the information of the time series correlation of the shield guiding deviation data is fully utilized, but also has higher modeling accuracy.

[0094] It can be understood that in step S3, a guiding deviation prediction controller is constructed based on the trained guiding deviation dynamic characteristic model. After collecting the real-time tunneling data and performing normalization processing, it is input into the guiding deviation prediction controller, and the prediction control objective function of the guiding deviation prediction controller is optimized to obtain the optimal solution of the guiding pressure difference of the propulsion cylinder. Among them, the real-time tunneling data refers to the guiding deviation of the shield head and the shield tail, the guiding pressure difference of the propulsion cylinder, the average propulsion speed and the total propulsion force value at the current moment. The guiding pressure difference of the propulsion cylinder at the next moment is predicted and output through the guiding deviation prediction controller, and at the next moment, the shield tunneling is controlled according to the prediction output, so as to control the shield machine to move forward along the tunnel design axis.

[0095] Specifically, first convert the guiding deviation dynamic characteristic model into an intermediate model structure, which can be expressed as:

[0096]

[0097] where k n represents the maximum value of the model input and output orders, φ(t - 1) represents the sum of the bias of the model at time t - 1 and the interferences of each order of the model, ε(t) represents the white noise at time t, represents the coefficient matrix of the model output, represents the coefficient matrix of the model input, L represents the row of the coefficient matrix, and this value is equal to the number of model output variables; N a represents the number of model output variables, N bdenotes the number of input variables. In addition, the meanings of the subscripts a and b are the same as those in the expression of the dynamic characteristics model of the guiding deviation. However, considering the actual change in the subscript range, i 1 and j 1 are used to represent.

[0098] Then, the state variables are defined as follows:

[0099]

[0100] where k 1 denotes the order number, x(t) represents the state variable at time t, and x 1,t represents the state vector of order 1, represents the state vector of order k 1 and, represents the output coefficient matrix, represents the input coefficient matrix.

[0101] Based on the above-defined state variables and the intermediate model structure, the state-space model of the guiding deviation prediction controller can be obtained as follows:

[0102]

[0103] where, y(t) represents the output variable, x(t) represents the state variable, and A t , B t , Φ t , and C are all coefficient matrices. a 1,t represents the coefficient matrix when the model output order is 1 at time t, and a 2,t represents the coefficient matrix when the model output order is 2 at time t. represents the coefficient matrix when the model output order is k n -1 at time t, represents the coefficient matrix when the model output order is k n at time t. b 1,t represents the coefficient matrix when the model input order is 1 at time t, and b 2,t represents the coefficient matrix when the model input order is 2 at time t. represents the coefficient matrix when the model input order is k n -1 at time t, represents the coefficient matrix when the model input order is k n at time t. φ(t) represents the sum of the bias of the model and the interferences of each order model, Ξ(t + 1) represents the matrix of white noise at time t + 1, ε(t + 1) represents the white noise at time t + 1, T represents the transpose of the vector, and I represents the identity matrix.

[0104] Then, determine the coefficient matrices A t , B t , Φ t in the state space model according to the tunneling data collected at time t, to obtain a locally linearized model, and then adopt a quadratic programming optimization algorithm to online optimize the predictive control objective function of the steering deviation prediction controller. Among them, the specific quadratic programming optimization algorithm belongs to the prior art and will not be elaborated here. The predictive control objective function is:

[0105]

[0106] where, minJ represents the predictive control objective function, represents the 2-norm of represents the 2-norm of represents the future N y -step forward prediction vector of the steering deviation y obtained based on the state space model, y r represents the steering deviation desired vector, y m i n represents the steering deviation lower limit, y max represents the steering deviation upper limit, u m i n represents the lower limit of the horizontal or vertical steering pressure difference of the propulsion cylinder, u max represents the upper limit of the horizontal or vertical steering pressure difference of the propulsion cylinder, represents the control vector solved at time t the increment vector of min represents the lower limit of the increment of the horizontal or vertical steering pressure difference of the propulsion cylinder, Δu max represents the upper limit of the increment of the horizontal or vertical steering pressure difference of the propulsion cylinder, is the control vector solved at time t, represents the control vector solved at time t - 1, R 1 (t) represents the weighting coefficient. It can be understood that after solving the control vector at time t, the first item u * (t) of

[0107] is the optimal solution of the horizontal or vertical steering pressure difference of the propulsion cylinder. 1 In addition, the weighting coefficient R

[0108]

[0109] where, β represents the adjustment coefficient, Ng represents the observation window length, R min represents the minimum value of the weighting coefficient, R max represents the maximum value of the weighting coefficient, y(t - i 3 ) represents the guidance deviation at time t - i 3 y(t - i 3 -1) represents the guidance deviation at time t - i - 1 3 - 1.

[0110] It can be understood that the present invention adopts a weighting coefficient R 1 (t) adaptive adjustment mechanism to adjust the control intensity, adjusts the weighting coefficient according to the change of the actual output, and increases the weighting coefficient R 1 (t) when the output change speed is large, and conversely decreases the weighting coefficient R 1 (t).

[0111] It can be understood that the guidance deviation prediction controller of the present invention has an online rolling optimization function, not only has strong anti-interference ability, can accurately control the shield machine to move forward along the tunnel design axis, but also can add constraints and adaptively adjust the weighting coefficient through the prediction control objective function, making the axial control of the shield machine more stable and the control cost smaller.

[0112] Optionally, as Figure 5 shown, the shield machine axis prediction control method further includes the following content:

[0113] Step S4: Determine whether to update the model based on the mean square error of the guidance deviation of the guidance deviation dynamic characteristic model.

[0114] It can be understood that in order to adapt to the formation changes in different sections of shield tunneling or different segments in the same section, the present invention also sets a dynamic update mechanism for the model. Specifically, the tunneling data collected in real time at the edge end is transmitted to the cloud and stored in the historical data collection module after being transmitted to the cloud, and is used for continuously training the guidance deviation dynamic characteristic model. Assume that the guidance deviation dynamic characteristic model trained by the cloud based on the real-time collected tunneling data is the latest model, and the guidance deviation dynamic characteristic model trained based on the historically collected tunneling data is the current model, and the current model is the model currently used at the edge end.

[0115] After the tunneling data collected in real time at the edge end is transmitted to the cloud, the tunneling data collected in real time is input into the current model, the predicted value of the guidance deviation of the current model is output, and then the mean square error of the guidance deviation of the current model is calculated. The calculation formula is: where e now represents the mean square error of the guidance deviation of the current model, N s represents the number of samples, y now represents the actual value of the guidance deviation, Represents the predicted value of the guiding deviation of the current model.

[0116] Similarly, input the real-time collected tunneling data into the latest model, output the predicted value of the guiding deviation of the latest model, and then calculate the mean square error of the guiding deviation of the latest model. The calculation formula is: Where, e new Represents the mean square error of the guiding deviation of the latest model, Represents the predicted value of the guiding deviation of the latest model.

[0117] Then, compare the mean square errors of the guiding deviations of the current model and the latest model, and judge whether to update the model at the edge end according to the comparison result. For example, if the following conditions are met, the model is updated:

[0118]

[0119] Where, E c Represents an adjustable threshold, and the specific value can be adjusted according to needs. If the above conditions are met, the cloud will transmit the model parameters of the latest model to the edge end, so as to update the model at the edge end.

[0120] It can be understood that the present invention adopts a model update method of cloud-edge collaboration, and judges whether to update the model structure parameters and model parameters at the edge end through the mean square error of the guiding deviation, so that the model at the edge end can adapt to the formation changes in different sections or different segments of the same section during shield tunneling. Moreover, the model update algorithm is designed in the cloud and does not affect the computing power of the edge end.

[0121] As Figure 6 shown, another embodiment of the present invention also provides a shield machine axis prediction control system, preferably adopting the shield machine axis prediction control method as described above, including:

[0122] A tunneling data acquisition module, configured to acquire tunneling data and perform normalization processing on it. Among them, the tunneling data includes the guiding deviations of the shield head and the shield tail, the guiding pressure difference of the propulsion cylinders, the average propulsion speed, and the total propulsion force value;

[0123] A model construction and training module, configured to construct a guiding deviation dynamic characteristic model based on the GRU network model and the ARX model, and use the acquired tunneling data to iteratively train the guiding deviation dynamic characteristic model until the iteration termination condition is met, and save the trained guiding deviation dynamic characteristic model;

[0124] The guiding deviation prediction control module is used to construct a guiding deviation prediction controller based on the trained dynamic characteristics model of the guiding deviation, collect real-time tunneling data, normalize it, and then input it into the guiding deviation prediction controller to optimize and solve the prediction control objective function of the guiding deviation prediction controller, so as to obtain the optimal solution of the guiding pressure difference of the propulsion cylinder.

[0125] It can be understood that for the axis prediction control system of the shield machine in this embodiment, it first collects the tunneling data related to axis control during shield tunneling, then constructs a dynamic characteristics model of the guiding deviation based on the GRU network model and the ARX model, and iteratively trains the dynamic characteristics model of the guiding deviation using the tunneling data. Then, it constructs a guiding deviation prediction controller based on the trained dynamic characteristics model of the guiding deviation, collects real-time tunneling data and inputs it into the guiding deviation prediction controller for optimization and solution, and then the optimal solution of the guiding pressure difference of the propulsion cylinder can be obtained, so as to control the shield machine to move forward along the tunnel design axis. The present invention is based on data-driven modeling, fully excavates the system dynamic characteristics contained in the data related to the control of the shield tunneling axis, has high prediction control accuracy, and combines the gated recurrent neural network GRU model and the autoregressive model with exogenous variables ARX model to construct a dynamic characteristics model of the guiding deviation to predict the guiding deviation, making full use of the advantage of the gated recurrent neural network GRU in dealing with long interval and long delay time series, and fully excavating the correlation characteristics before and after the time series of the shield guiding deviation data, so that the model not only has the ability of global nonlinear description, but also has the ability to describe the system time series related characteristics, the system dynamic characteristics are described more comprehensively, it is a nonlinear model that more conforms to the actual working characteristics of the shield machine guiding, and the model has stronger generalization ability and higher accuracy.

[0126] In addition, the axis prediction control system of the shield machine further includes:

[0127] The model update module is used to judge whether to update the model based on the mean square error of the guiding deviation of the dynamic characteristics model of the guiding deviation.

[0128] It can be understood that each module of the system embodiment of the present invention corresponds to each step of the above method embodiment, so the specific working process of each module will not be elaborated here, and reference can be made to the above method embodiment correspondingly.

[0129] In addition, another embodiment of the present invention further provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and the processor is used to execute the steps of the above-mentioned method by calling the computer program stored in the memory.

[0130] In addition, another embodiment of the present invention further provides a computer-readable storage medium for storing a computer program for shield machine axis prediction control, and the computer program executes the steps of the above-mentioned method when running on a computer.

[0131] The forms of common computer-readable storage media include: floppy disks, flexible disks, hard disks, magnetic tapes, any other magnetic media, CD-ROMs, any other optical media, punch cards, paper tapes, any other physical media with a pattern of holes, random access memories (RAMs), programmable read-only memories (PROMs), erasable programmable read-only memories (EPROMs), flash erasable programmable read-only memories (FLASH-EPROMs), any other memory chips or cartridges, or any other media readable by a computer. Instructions can be further transmitted or received by a transmission medium. The term transmission medium can include any tangible or intangible medium that can be used to store, encode, or carry instructions for execution by a machine, and includes digital or analog communication signals or the intangible medium that facilitates the communication of the above instructions. The transmission medium includes coaxial cables, copper wires, and optical fibers, which include the wires of a bus used to transmit a computer data signal.

[0132] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript.

[0133] The present application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to generate a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more flows or multiple flows and / or blocks Figure 1 one or more blocks or multiple blocks.

[0134] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to operate in a particular manner, such that the instructions stored in the computer-readable memory produce a manufacture including an instruction device that implements the functions specified in the flowchart(s) Figure 1 one or more flowcharts and / or block diagrams Figure 1 specified in one or more blocks or a plurality of blocks.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, so that the instructions executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart(s) Figure 1 one or more flowcharts and / or block diagrams Figure 1 specified in one or more blocks or a plurality of blocks.

[0136] Although the preferred embodiments of the present application have been described, additional changes and modifications can be made by those skilled in the art once they learn of the basic inventive concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present application.

[0137] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

[0138] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A shield machine axis predictive control method, characterized in that: Includes the following: Acquire excavation data and normalize them, wherein the excavation data includes the guide deviation of the shield head and the shield tail, the guide pressure difference of the propulsion cylinder, the average propulsion speed and the total propulsion force value; A guidance deviation dynamic characteristic model is constructed based on the GRU network model and the ARX model, and the guidance deviation dynamic characteristic model is iteratively trained using the acquired tunneling data until the iteration termination condition is met, and the trained guidance deviation dynamic characteristic model is saved; A steering deviation prediction controller is constructed based on the trained steering deviation dynamic characteristic model. Real-time excavation data is collected, normalized and input into the steering deviation prediction controller. The predictive control objective function of the steering deviation prediction controller is optimized and solved to obtain the optimal solution for the steering pressure difference of the thrust cylinder.

2. The shield machine axis prediction control method according to claim 1, characterized in that: The expression of the guiding deviation dynamic characteristic model is: Among them, y(t) represents the output of the model, that is, the guidance deviation vector of the shield head and the shield tail, t represents the current moment, p represents the order of the model output, q represents the order of the model input; s represents the order of the model measurable interference, f represents the delay of the model input, g represents the delay of the model measurable interference, u represents the guidance pressure difference of the propulsion cylinder, d represents the measurable interference vector of the model, that is, the average propulsion speed and the total propulsion force, ε(t) represents the Gaussian white noise signal, a 0,t-1 The coefficient matrix representing the model bias, a 11，0,t-1 、a 21，0,t-1 for a 0,t-1 The element in a i,t-1 Represents the coefficient matrix of the model output, a 11,i,t-1 、a 12,i,t-1 、a 21,i,t-1 、a 22,i,t-1 for a i,t-1 The elements in b j,t-1 represents the coefficient matrix of the model input, b 11,j,t-1 , b 21,j,t-1 for b j,t-1 The elements in c k,t-1 The coefficient matrix representing the measurable disturbance of the model, c 11,k,t-1 、c 12,k,t-1 、c 21,k,t-1 、c 22,k,t-1 c k,t-1 The elements in .

3. The shield machine axis prediction control method according to claim 1, characterized in that: The guidance deviation dynamic characteristic model includes a GRU layer, a fully connected layer and an ARX layer, wherein the output result of each time step of the last layer in the GRU layer is input into the ARX layer after constructing a state dependency vector through the fully connected layer, and the ARX layer fits the regression coefficient matrix of the ARX layer through the state dependency vector.

4. The shield machine axis prediction control method according to claim 2, characterized in that: The state space model of the steering deviation predictive controller is: in, y(t) represents the output variable, x(t) represents the state variable, A t , B t , Φ t , C are coefficient matrices, a 1,t represents the coefficient matrix when the model output order is 1 at time t, a 2,t represents the coefficient matrix when the model output order is 2 at time t, a kn-1,t Indicates that the model output order at time t is k n The coefficient matrix when -1, a kn,t Indicates that the model output order at time t is k n The coefficient matrix when b 1,t represents the coefficient matrix when the model input order is 1 at time t, b 2,t represents the coefficient matrix when the model input order is 2 at time t, b kn-1,t Indicates that the model input order at time t is k n The coefficient matrix when -1, b kn,t Indicates that the model input order at time t is k n where φ(t) represents the coefficient matrix at time t and the sum of the bias of the model and the interference of each order model at time t, Ξ(t+1) represents the matrix of white noise at time t+1, ε(t+1) represents the white noise at time t+1, T represents the transpose of the vector, and I represents the unit matrix.

5. The shield machine axis prediction control method according to claim 4, characterized in that: The predictive control objective function is: Among them, minJ represents the predictive control objective function, express The 2-norm of express The 2-norm of represents the future guidance deviation y obtained based on the state space model y One step ahead prediction vector, y r represents the expected vector of the steering deviation, y min Indicates the lower limit of the guidance deviation, y max Indicates the upper limit of the guidance deviation, u min Indicates the lower limit of the guide pressure difference of the thrust cylinder, u max Indicates the upper limit of the pilot pressure difference of the thrust cylinder. represents the control vector obtained at time t The increment vector, Δu min Indicates the incremental lower limit of the pilot pressure difference of the thrust cylinder, Δu max Indicates the incremental upper limit of the pilot pressure difference of the thrust cylinder. is the control vector solved at time t, and R1(t) represents the weighting coefficient.

6. The shield machine axis prediction control method according to claim 5, characterized in that: The weighting coefficient is adaptively adjusted based on the following formula: Among them, β represents the adjustment coefficient, N g represents the observation window length, R min Represents the minimum value of the weighting coefficient, R max represents the maximum value of the weighting coefficient, y(t-i3) represents the guidance deviation at time t-i3, and y(t-i3-1) represents the guidance deviation at time t-i3-1.

7. The shield machine axis prediction control method according to claim 1, characterized in that: Also included are the following: The mean square error of the guidance deviation based on the guidance deviation dynamic characteristic model is used to determine whether to update the model.

8. A shield machine axis prediction control system, characterized in that: include: A tunneling data acquisition module is used to acquire tunneling data and normalize it, wherein the tunneling data includes the guide deviation of the shield head and the shield tail, the guide pressure difference of the propulsion cylinder, the average propulsion speed and the total propulsion force value; The model building and training module is used to build a guidance deviation dynamic characteristic model based on the GRU network model and the ARX model, and iteratively train the guidance deviation dynamic characteristic model using the acquired tunneling data until the iteration termination condition is met, and save the trained guidance deviation dynamic characteristic model; The steering deviation prediction control module is used to build a steering deviation prediction controller based on the trained steering deviation dynamic characteristic model, collect real-time excavation data, normalize it and input it into the steering deviation prediction controller, optimize and solve the prediction control objective function of the steering deviation prediction controller, and obtain the optimal solution for the steering pressure difference of the thrust cylinder.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the method according to any one of claims 1 to 7 by calling the computer program stored in the memory.

10. A computer-readable storage medium for storing a computer program for predictive control of a shield machine axis, characterized in that: When the computer program is run on a computer, the steps of the method according to any one of claims 1 to 7 are executed.

Images