Shield construction parameter prediction method and device, medium and equipment

Abstract

The invention discloses a shield construction parameter prediction method and device, a medium and equipment, and relates to the technical field of tunnel engineering. According to the method, the physical information is fused into the neural network model, shield parameters are predicted through the neural network model, the analysis value of the shield parameters is determined through analysis of the physical information, and then the data loss and the physical loss are determined respectively; the reliability of the data information and the reliability of the physical information are respectively determined through the average value of the data loss and the physical loss of the multiple rounds, so that the adaptive weights of the data loss and the physical loss are determined according to the average value of the data loss and the physical loss of the multiple rounds, and the weighted sum of the adaptive weights is determined to obtain the total loss; in the neural network model training process based on the total loss, the influence weight of the data information and the physical information on the prediction result is adaptively adjusted, the influence of the inaccuracy of the physical model and the data noise on the prediction accuracy is limited, and the prediction accuracy of the shield construction parameters is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering, and in particular to a shield construction parameter prediction method, device, medium and equipment. BACKGROUND

[0002] At present, shield construction is a construction method of using a shield machine to excavate a tunnel underground and simultaneously completing lining. The shield machine integrates functions of excavation, soil removal, lining and grouting, and realizes automation and high efficiency of tunnel construction through its propulsion system, cutting device and assembly system. The construction parameters (such as thrust and torque) of shield construction are crucial to the efficiency, quality and energy consumption of tunnel construction.

[0003] In the prior art, there are traditional data-driven methods to predict shield construction parameters through artificial neural networks and convolutional neural networks, but such methods ignore the physical constraints in construction, and the prediction results lack physical interpretation and perform poorly in sparse or noisy data. There are also purely physical methods to determine shield construction parameters through mechanical analysis and numerical simulation. However, under complex geological conditions, it is difficult to accurately obtain physical model parameters, and the calculation cost is high. There are also hybrid methods that integrate physical information into machine learning models, which have improved accuracy and efficiency compared to separate predictions.

[0004] However, the existing hybrid method for predicting shield construction parameters cannot handle the impact of inaccurate physical information and data noise on model performance, and has significant limitations in predicting construction parameters under complex geological conditions, with poor prediction accuracy. SUMMARY

[0005] Therefore, it is necessary to provide a shield construction parameter prediction method, device, medium and equipment to solve the above technical problems.

[0006] The present application adopts the following technical solutions: The present application provides a shield construction parameter prediction method, comprising: Obtaining shield machine parameters and corresponding regional geological data within a preset number of excavation rings before the current excavation ring of shield construction as sample data; and labeling the true value of the shield construction parameters corresponding to the sample data; Inputting the sample data into a neural network model to obtain a shield construction parameter prediction value, and determining a shield construction parameter analytical value according to the sample data through a calculation formula of the shield construction parameter; Determining a data loss according to the shield construction parameter prediction value and the true value of the shield construction parameter, and determining a physical loss according to the shield construction parameter analytical value and the true value of the shield construction parameter; determining an average value of the data loss and an average value of the physical loss according to the data loss and the physical loss of a preset number of training rounds; Adaptive weights of the data loss and the physical loss are determined according to relative sizes of the data loss average and the physical loss average, and the data loss and the physical loss are weighted respectively to determine a total loss, and the neural network model is trained for multiple rounds with the optimization target of minimizing the total loss; The geological data and the shield machine parameters of the current tunneling ring of the shield construction are input into the trained neural network model to obtain the shield construction parameters of the current tunneling ring of the shield construction.

[0007] Optionally, the neural network model is obtained by sequentially connecting an LSTM model and a KAN model, and the KAN model takes the last input hidden state of the LSTM model as input. The LSTM model is used to extract time-dependent features in the input shield machine parameter time sequence, and the KAN model is used to extract the mapping relationship between the time-dependent features and the shield construction parameters.

[0008] Optionally, the shield construction parameters include total thrust and cutterhead torque of the shield machine. The shield construction parameter analytical value is determined according to the calculation formula of the shield construction parameter based on the sample data, and specifically includes: The shield construction parameter analytical value is determined according to the calculation formula of the shield construction parameter based on the sample data by the following formula: ; ; Wherein, is the total thrust analytical value, is the cutterhead torque analytical value, is the elastic modulus of the tunnel surface soil, is the cutterhead opening ratio, is the Poisson's ratio of the excavation interface, is the advancing amount of a single rotation, is the cutterhead radius, is the static soil pressure coefficient, is the unit weight of the soil, is the burial depth of the tunnel, is the average soil pressure in the cabin, is the average ground pressure on the shield machine, is the length of the shield machine, is the weight of the shield machine, is the total weight of the rear supporting equipment, is the shear modulus of the soil, is the internal friction angle of the soil, is the cutterhead side wall thickness, is the number of stirring rods, is the diameter of the stirring rod. is the length of the stirring rod, is the average distance of the stirring rod from the axis, denotes the friction coefficient between the cutterhead system and the soil in the range of the face, denotes the friction coefficient between the shield shell and the soil, denotes the friction between the rear equipment and the track, denotes the friction coefficient between the cutterhead side wall and the soil, and denotes the friction coefficient between the stirring rod and the slurry.

[0009] Optionally, the adaptive weights of the data loss and the physical loss are determined according to the relative size of the data loss average value and the physical loss average value, respectively, specifically including: The adaptive weights of the data loss and the physical loss are determined according to the relative size of the data loss average value and the physical loss average value by the following formula: ; ; wherein, is the adaptive weight of the data loss, is the adaptive weight of the physical loss, is the data loss average value, is the physical loss average value.

[0010] Optionally, the adaptive weights of the data loss and the physical loss are determined according to the relative size of the data loss average value and the physical loss average value, respectively, specifically including: When the data loss average value is less than 0 or greater than 1, the adaptive weight of the data loss is set to 0, or when the physical loss average value is less than 0 or greater than 1, the adaptive weight of the physical loss is set to 0; When the data loss average value and the physical loss average value are both between 0 and 1, the adaptive weights of the data loss and the physical loss are determined according to the relative size of the data loss average value and the physical loss average value, respectively.

[0011] Optionally, the neural network is used to predict the total thrust and the cutterhead torque of the shield machine in the shield construction; The method further includes: According to the total thrust and the cutterhead torque of the shield machine in the current tunneling ring of the shield construction, the specific energy of the shield construction is determined by the following formula: ; wherein, is the specific energy of the shield construction, is the total thrust of the shield machine, is the forward speed, is the cutterhead rotation speed, A cutter head torque of a shield tunneling machine, A diameter of the cutter head system.

[0012] Optionally, the shield tunneling machine parameters and the geological data of the corresponding area within a preset number of tunneling rings before the current tunneling ring of the shield tunneling are obtained as sample data, and specifically include: The shield tunneling machine parameters and the geological data of the corresponding area within a preset number of tunneling rings before the current tunneling ring of the shield tunneling are obtained. The geological data is preprocessed by the following formula to obtain a weighted geological parameter: ; The shield tunneling machine parameters and the weighted geological parameters of the corresponding area within a preset number of tunneling rings before the current tunneling ring of the shield tunneling are taken as sample data. Wherein, The weighted geological parameter of the geological data is the geological data of the soil from the inner to the outer layer of soil involved in the tunneling face, The geological data of the soil from the inner to the outer layer of soil involved in the tunneling face, The central angle corresponding to the soil from the inner to the outer layer of soil, The soil area corresponding to the soil from the inner to the outer layer of soil, The total number of layers of soil involved in the tunneling face. The geological data includes at least one of water content, density, soil particle specific gravity, porosity, saturation, compression modulus, Poisson's ratio, horizontal reaction coefficient, vertical reaction coefficient, horizontal permeability coefficient, vertical permeability coefficient, cohesion, internal friction angle and static earth pressure coefficient.

[0013] The present application provides a kind of shield construction parameter prediction device, including: The acquisition module is used to obtain the shield tunneling machine parameters and the geological data of the corresponding area within a preset number of tunneling rings before the current tunneling ring of the shield tunneling as sample data;The true value of the shield construction parameter corresponding to the sample data is labeled; The prediction analysis module is used to input the sample data into the neural network model, obtain the shield construction parameter prediction value, and determine the shield construction parameter analytical value according to the sample data through the calculation formula of the shield construction parameter; The loss determination module is used to determine the data loss according to the shield construction parameter prediction value and the shield construction parameter true value, and determine the physical loss according to the shield construction parameter analytical value and the shield construction parameter true value;According to the data loss and physical loss of the preset training number, the average value of data loss and the average value of physical loss are determined; The weighting module is configured to determine adaptive weights of the data loss and the physical loss according to relative sizes of the average data loss and the average physical loss, and to weight the data loss and the physical loss respectively to determine a total loss, and to perform multiple rounds of training on the neural network model with the optimization target of minimizing the total loss. The prediction module is configured to input the geological data and the shield machine parameters of the current tunneling ring of the shield construction into the trained neural network model to obtain the shield construction parameters of the current tunneling ring of the shield construction.

[0014] The present application provides a computer readable storage medium, the storage medium stores a computer program, the computer program is executed by a processor to realize the shield construction parameter prediction method.

[0015] The present application provides a computer device, including a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executes the program to realize the shield construction parameter prediction method.

[0016] The above-mentioned at least one technical solution adopted by the present application can achieve the following beneficial effects: The present application integrates physical information into the neural network model, predicts the shield parameters through the neural network model, determines the analytical value of the shield parameters through the physical information, and then determines the data loss and the physical loss, respectively, and determines the reliability of the data information and the physical information through the average values of multiple rounds of data loss and physical loss, so as to determine the adaptive weights of the data loss and the physical loss according to the average values of multiple rounds of data loss and physical loss, and determine the weighted sum of the two to obtain the total loss, and in the process of training the neural network model with the optimization target of minimizing the total loss, the influence weight of the data information and the physical information on the prediction result is adaptively adjusted, so as to limit the influence of the inaccuracy of the physical model and the data noise on the accuracy of the model, and improve the prediction accuracy of the shield construction parameters. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the present application and together with the description serve to explain the present application. In the drawings:

[0018] Figure 1 A shield construction parameter prediction method flowchart is provided for the present application; Figure 2 A LSTM and KAN hybrid network structure schematic diagram is provided for the present application; Figure 3 A parallel PINNs network structure schematic diagram is provided for the present application; Figure 4 A geological layer weighting coefficient calculation schematic diagram provided by the present application; Figure 5 A processed geological data schematic diagram provided by the present application; Figure 6 A processed shield machine parameter schematic diagram provided by the present application; Figure 7 A training process loss value change curve schematic diagram provided by the present application; Figure 8 A weight change curve schematic diagram provided by the present application; Figure 9 A thrust prediction result and error schematic diagram provided by the present application; Figure 10 A torque prediction result and error schematic diagram provided by the present application; Figure 11 A power consumption prediction result and error schematic diagram provided by the present application; Figure 12 A shield construction parameter prediction device schematic diagram provided by the present application; Figure 13 A computer equipment schematic diagram for realizing the shield construction parameter prediction method provided by the present application. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described below in connection with specific embodiments of the present application and corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0020] At present, in the mixed prediction method of shield construction, part of the research integrates physical information into the machine learning model, but fails to effectively deal with the influence of inaccurate physical information and data noise on the model performance. Whether it is a traditional data-driven method, a pure physical model method or a mixed prediction method, there are significant limitations in dealing with complex geological conditions and construction parameter prediction, which cannot fully meet the efficient and accurate prediction demand of construction parameters in shield construction, and cannot meet the efficient and accurate prediction demand of energy consumption in shield construction.

[0021] The present application aims to solve the problem of lack of physical interpretation and sensitivity to data noise in the prior art. The present application proposes a two-stage Physics-Informed Neural Networks (PINNs) framework based on physical model embedding and adaptive weight allocation, which limits the influence of physical model inaccuracy and data noise on model accuracy, and realizes high-precision prediction of shield construction parameters (total thrust of the shield machine and cutter torque) and energy consumption (specific energy).

[0022] The technical solutions of the embodiments of the present application are described in detail below with reference to the drawings.

[0023] Figure 1 The present application is a shield construction parameter prediction method flowchart, which specifically includes the following steps: S101: Obtain the shield machine parameters and the corresponding regional geological data within a preset number of tunneling rings before the current tunneling ring of shield construction as sample data; and label the true value of the shield construction parameters corresponding to the sample data.

[0024] S102: Input the sample data into the neural network model to obtain the predicted value of the shield construction parameters, and determine the analytical value of the shield construction parameters according to the sample data through the calculation formula of the shield construction parameters.

[0025] S103: Determine the data loss according to the predicted value of the shield construction parameters and the true value of the shield construction parameters, and determine the physical loss according to the analytical value of the shield construction parameters and the true value of the shield construction parameters; determine the average value of the data loss and the average value of the physical loss according to the data loss and the physical loss of the preset training number of rounds.

[0026] S104: Determine the adaptive weights of the data loss and the physical loss according to the relative sizes of the average values of the data loss and the physical loss, and weight the data loss and the physical loss respectively to determine the total loss, and perform multi-round training on the neural network model with the optimization goal of minimizing the total loss.

[0027] S105: Input the geological data and the shield machine parameters of the current tunneling ring of shield construction into the trained neural network model to obtain the shield construction parameters of the current tunneling ring of shield construction.

[0028] For convenience of description, only the server is described below as the execution subject. The server mentioned in the present application can be a server arranged in a business platform, or a device such as a desktop computer, a notebook computer, etc. capable of executing the scheme of the present application.

[0029] In one or more embodiments of the present invention, to improve the prediction accuracy of shield tunneling parameters, corresponding data from a preset number of tunneling cycles prior to the current tunneling cycle can be used as sample data to update and train the prediction model. Specifically, for each tunneling cycle, geological data for the corresponding area and shield machine data for the corresponding tunneling cycle can be used as sample data.

[0030] Geological data may include physical and mechanical parameters of the corresponding area, soil layer thickness and location information, etc. For example, based on the soil layers involved in the excavation face, the soil layer area of ​​each soil layer, the tunnel radius, the central angle corresponding to each soil layer, etc. can be obtained. Other address parameters include moisture content (W), density (ρ), soil particle specific gravity (Gs), porosity (e), saturation (Sr), compression modulus (Es), Poisson's ratio (ν), horizontal reaction coefficient (KCH), vertical reaction coefficient (KCV), horizontal permeability coefficient (KH), vertical permeability coefficient (KV), cohesion (c), internal friction angle (φ), static earth pressure coefficient (K) and other one or more geotechnical parameters can be obtained on demand.

[0031] Shield machine data can include thrust speed (AS), cutterhead speed (CRS), grouting pressure (GP), and excavation chamber pressure (ECP). The actual values ​​of shield machine construction parameters corresponding to the annotated sample data can include the observed values ​​of total thrust (TF) and cutterhead torque (CT) during actual shield machine construction. Two models can be used to predict total thrust and cutterhead torque.

[0032] Furthermore, in one or more embodiments of the present invention, in order to reduce the amount of data, the geological data may be first After obtaining the geological data of the corresponding area within the preset tunneling ring number before the current tunneling ring of shield construction, the geological data can be preprocessed using the following formula to obtain weighted geological parameters: Where, For geological data The weighted geological parameters, The tunnel excavation face involves the first geological data of soil layers, From the inside out The central angle of the circle corresponding to the soil layer, is the tunnel radius, For the The area of ​​the soil layer corresponding to the soil layer, is the total number of soil layers involved in the tunnel excavation face.

[0033] Therefore, the shield machine parameters within the preset tunneling ring number before the current tunneling ring of shield construction and the weighted geological parameters of the corresponding area can be used as sample data.

[0034] For the neural network model, in one or more embodiments of the application, a physics-informed neural network (PINN) is proposed, which embeds the physical equation describing the longitudinal structural mechanical behavior of the shield tunnel into the physical neuron of the neural network, and uses the measured data as the input of the information neuron, so as to realize the real-time inversion and updating of the tunnel structure parameters, surrounding stratum parameters and load distribution law through the combination of the two.

[0035] Specifically, the neural network model can be obtained by combining Long Short-Term Memory (LSTM) and Kolmogorov-Arnold Networks (KAN). The LSTM is good at processing time series data and capturing the time dependence of data, so it is used to extract the time dependence features in the input time series of shield machine parameters; the KAN is based on the Kolmogorov-Arnold representation theorem, which is suitable for representing multi-dimensional mapping relationship and can analyze complex nonlinear relationship, so it is used to extract the mapping relationship between the time dependence features and the shield construction parameters. Based on this, the application proposes a hybrid network structure PINN, Figure 2 FIG. 1 is a schematic diagram of an LSTM and KAN hybrid network structure in the application. The combination of LSTM for capturing dynamic changes and KAN for providing analytical ability for complex mapping can effectively reduce the prediction error of a single model in theory.

[0036] For the input sequence , the calculation formula of the output of the LSTM is as follows: According to the KAN theory, for any dimensional continuous function , there are weighted combinations of single variable functions. The last input hidden state of the LSTM is taken as the input of the KAN, and the mathematical representation of the output shield construction parameter prediction value is as follows:

[0037] wherein, is the LSTM hidden state the first component. is a nonlinear activation function used to construct the output. denotes a weight coefficient that controls the input of the nonlinear function . is a function that performs a nonlinear transformation on the input dimension . is a constant that represents a bias.

[0038] The combination of data information and physical information adopts a PI information embedding method, directly embeds a physical model into a network training process, and introduces physical formulas of total thrust ( ) and cutter torque ( ) of a shield machine in a loss function in a PINNs training process. In one or more embodiments of the present application, the analytical values of the shield construction parameters ( ) can be determined according to sample data by the following formula:

[0039] wherein (kPa) represents an elastic modulus of tunnel surface soil, represents a Poisson's ratio of an excavation interface. A cutter radius is represented by R (m), and a propulsion amount per single rotation ( (m / rev) is defined as a ratio of a propulsion speed to a cutter rotation speed. represents a static soil pressure coefficient, (kN / m3) represents a unit weight of soil, is a cutter radius. (m) corresponds to a burial depth of a tunnel, and (bar) and (kPa) respectively represent an average soil pressure in a cabin and an average ground pressure on the shield machine. Other parameters include , a cutter opening ratio; (m), a length of the shield machine; (kN), a weight of the shield machine; (kN), a total weight of a rear supporting device. A shear modulus (kPa) of soil is calculated by . An internal friction angle (°) and cohesion (kPa) of soil also play an important role. A cutter sidewall thickness is represented by (m). m , (m), (m), and (m) respectively represent the number, diameter, length and average distance from the axis of the stirring rods. The friction coefficient , , , and describe the interaction between different interfaces, including the interaction between the cutterhead system, the shield shell and the soil, and the interaction between the rear matching equipment and the track, in particular, represents the friction coefficient between the cutterhead system and the soil in the range of the working face, represents the friction coefficient between the shield shell and the soil, represents the friction between the rear matching equipment and the track, represents the friction coefficient between the cutterhead side wall and the soil, and represents the friction coefficient between the stirring rod and the slurry.

[0040] The PINNs loss function is composed of two parts: data loss and physical loss, both of which can be constructed using mean square error (MSE). Let the parameters and represent the fitting and network model parameters. The loss functions of TF and CT are defined as follows:

[0041] wherein the parameters and respectively represent the data loss term weight and the physical loss term weight in the fitting. Similarly, and respectively represent the data loss term weight and the physical loss term weight in the fitting. represents the data loss term predicted by TF, represents the data loss term predicted by CT, represents the physical loss term analyzed by TF, represents the physical loss term analyzed by CT. is the number of samples, and respectively represent the observed values (i.e. true values) of the total thrust and the cutterhead torque, is the predicted value of the total thrust predicted by the PINNs model, is a predicted cutterhead torque value predicted by the PINNs model. is an analytical total thrust value obtained by an analytical formula, is an analytical cutterhead torque value obtained by an analytical formula.

[0042] To realize adaptive adjustment of weights of data loss and physical loss in the loss function, the application designs a weight updating strategy based on loss value changes in the first n stages.

[0043] Specifically, in one or more embodiments of the application, the average value of data loss and the average value of physical loss can be determined first, and then the adaptive weights of data loss and physical loss can be determined according to the relative sizes of the average value of data loss and the average value of physical loss by the following formula: wherein, is the adaptive weight of data loss, is the adaptive weight of physical loss, is the average value of data loss, is the average value of physical loss. Subsequently, the data loss and the physical loss can be weighted based on the weights to determine the total loss, and the neural network model can be trained for multiple rounds with the optimization goal of minimizing the total loss.

[0044] Taking thrust as an example, the strategy is explained how to adaptively adjust the weights of data loss and physical loss in the PINNs prediction process. Let and respectively represent the data loss and the physical loss in the thrust prediction model at the i-th iteration, and respectively represent the data loss and the physical loss in the inference prediction model at the i-th iteration, and respectively represent the adaptive weights of data loss and physical loss in the inference prediction model at the i-th iteration. At the i-th iteration, to the average values of data loss and physical loss from the 1st stage to the i-th stage are calculated as follows:

[0045] Using the above method, it can be ensured that and ​The sum is 1. The weight of the loss term will be adjusted according to the change of data loss and physical loss, to ensure that the influence of data anomaly or physical inaccuracy on the accuracy of the model is limited.

[0046] In special cases, such as Or The calculation result is not between 0 and 1, indicating that the data anomaly or physical inaccuracy at this time has seriously reduced the accuracy of the model, and the corresponding weight will be set to 0. That is, when the average value of data loss is less than 0 or greater than 1, the adaptive weight of data loss is set to 0, or when the average value of physical loss is less than 0 or greater than 1, the adaptive weight of physical loss is set to 0.

[0047] In one or more embodiments of the present application, taking total thrust and cutter torque as an example, two stages of model training and application can be included. In the first stage, PINNs suitable for predicting total thrust And cutter torque Are constructed, that is, a parallel PINNs network structure is used to train prediction models for predicting total thrust And cutter torque , Figure 3 A parallel PINNs network structure in the present application is shown in the schematic diagram.

[0048] In the second stage, based on the real-time trained parallel PINNs network structure, the total thrust And cutter torque Of the current tunneling ring of the shield construction can be predicted based on the geological data and shield machine parameters of the current tunneling ring, and further, the specific energy of the shield construction can be determined by the following formula based on the total thrust And cutter torque Predicted by PINNs. .

[0049] In the formula, (kN·m) is the cutter torque of the cutting system of the shield machine, (kN) is the total thrust of the propulsion system of the shield machine, is the forward speed (m / min), is the cutter speed (rev / min), is the diameter of the cutter system. Therefore, the rock breaking energy required by the shield machine during tunneling can be estimated when the total thrust of the shield machine and the torque of the cutter system are determined.

[0050] Training strategy considering time and space factors and calculation efficiency: the PINNs model can be updated once every 1 ring of the shield machine, considering time and space factors and calculation efficiency, historical data of a certain length close to the tunneling position are used to and update of the PINNs prediction model. Assuming that the length of the training data is , the current tunneling position is the ring. As the shield machine tunnels, the data of the previous ring can be selected as the training data of the PINNs prediction model when ; the data from the ring to the ring can be selected as the training data of the PINNs prediction model when .

[0051] Based on the shield construction parameter prediction method shown in Figure 1 , the present application integrates physical information into the neural network model, not only predicts the shield parameters through the neural network model, but also determines the analytical value of the shield parameters through physical information analysis, then determines the data loss and physical loss respectively, and determines the reliability of the data information and physical information through the average value of multiple rounds of data loss and physical loss, so as to determine the adaptive weight of data loss and physical loss based on the average value of multiple rounds of data loss and physical loss, and determine the weighted sum of the two to obtain the total loss. In the process of training the neural network model with the optimization goal of minimizing the total loss, the influence weight of data information and physical information on the prediction result is adaptively adjusted, thereby limiting the influence of physical model inaccuracy and data noise on model accuracy, and improving the prediction accuracy of shield construction parameters.

[0052] Specifically, compared with the prior art, the present application has the following advantages: (1) The adaptive weight allocation strategy automatically calculates the weights of the physical loss term and the data loss term during training, limiting the influence of physical model inaccuracy and data noise on model accuracy.

[0053] (2) The developed parallel PINNs framework realizes high-precision prediction of shield construction parameters (thrust and torque) and energy consumption (specific energy).

[0054] (3) The modeling process considers temporal and spatial factors and computational efficiency, improving the update efficiency of the PINNs prediction model while ensuring computational effectiveness.

[0055] When applying the shield construction parameter prediction method provided by the present application, the steps shown in Figure 1 may not be executed in the order shown, and the execution order of each step can be determined as needed, which is not limited by the present application.

[0056] Further, the application also provides an embodiment of the shield construction parameter prediction method, which is based on a certain shield tunnel project, and a total of 959 ring data sets are collected, covering shield parameters and geological parameters. The specific description of the embodiment is as follows:

[0057] Step 1: data processing.

[0058] Geological data processing: according to the geological characteristics of the project area, the stratum of the tunnel face is distributed in layers, and the physical and mechanical properties of each layer of soil are obviously different. In order to accurately represent the geological conditions, the embodiment calculates the geological weighting coefficient in combination with the physical and mechanical parameters, soil thickness and position information, as shown in Figure 4 , Figure 4 is a schematic diagram of the calculation of a geological layer weighting coefficient in the application. Taking a ring as an example, it is assumed that the tunnel face of the ring involves 3 layers of soil, and the geotechnical parameters (such as compression modulus Es, internal friction angle φ , etc.) of each layer of soil from inside to outside are , , , the corresponding soil area is , , , the tunnel radius is , and the central angle of each layer of soil is , , (satisfying ). According to the formula , the weighted geological parameters of the ring are calculated , which comprehensively reflect the geological conditions of the ring. The processing results of the geological data of water content (W), density ρ , soil particle specific gravity (Gs), porosity (e), saturation (Sr), compression modulus (Es), Poisson's ratio ν , horizontal reaction force coefficient (KCH), vertical reaction force coefficient (KCV), horizontal permeability coefficient (KH), vertical permeability coefficient (KV), cohesion (c), internal friction angle (φ), and static soil pressure coefficient (K) are shown in Figure 5 , Figure 5 is a schematic diagram of the processed geological data in the application.

[0059] Shield machine parameter processing: Parameter data is obtained from the shield machine's integrated data acquisition system, with a frequency of 10 seconds per time or once every 20 mm of excavation. The original data contains a large amount of redundancy and outliers, such as erroneous data generated by some sensors due to external interference. Based on expert experience, the propulsion speed (AS), cutterhead speed (CRS), grouting pressure (GP), and excavation chamber pressure (ECP) can be selected as training samples, and the total thrust (TF) and cutterhead torque (CT) can be used as annotations to construct a training data set with these six key parameters. Statistical analysis methods are used to detect outliers. For example, if the normal range of propulsion speed is set to , data points outside this range are considered outliers and are removed to ensure data quality. Figure 6 This is a schematic diagram of shield machine parameters after processing in the present invention.

[0060] Step 2: Train the model and apply AWAS to dynamically adjust the data and physical loss weights.

[0061] Model initialization: Build a parallel PINNs model and set the parameters of the LSTM and KAN networks. In this example, the LSTM network has 3 hidden layers with 50 neurons in each hidden layer, and the KAN network has 3 hidden layers with 50 neurons in each hidden layer. Set the initial training parameters to 10 training samples and 100 training epochs. and Prediction model 、 、 and All are set to 0.5. The training process uses the MSE loss function, and the loss value change curve is as follows Figure 7 As shown, Figure 7 This is a schematic diagram of a loss value change curve during the training process in the present invention.

[0062] Model training: The training is performed by combining the Adam optimizer and the L-BFGS optimizer. During the training process, the weights of data loss and physical loss are adjusted every 10 epochs according to the adaptive weight allocation strategy (AWAS). Taking a certain training stage as an example, assuming that the current iterations, if , then keep the weight unchanged; if , calculate the first 10 stages (from arrive ) of the data loss average and physical loss value. If the loss value is not If the error is within the specified range, it indicates that the data is abnormal or the physical model has a large error, and the corresponding weight is set to 0. Figure 8 This is a schematic diagram of a weight change curve in the present invention.

[0063] Step 3: prediction result and .

[0064] Input data: the geological data (such as weighted rock-soil parameters) and the shield machine parameters (such as the pushing speed, the cutter head rotating speed, etc.) of the current tunneling ring of the shield construction are input into the trained T-PINNs model in time sequence.

[0065] Calculation process: the model first processes the input sequence through the LSTM network, and then inputs the last hidden state of the LSTM to the KAN network to calculate the predicted values of the total thrust and the cutter head torque, as shown in Figure 9 and Figure 10 . Figure 9 is a prediction result and error diagram in the present application . Figure 10 is a prediction result and error diagram in the present application .

[0066] Step 4: prediction result .

[0067] The specific energy is calculated according to the formula , wherein is the total thrust and the cutter head torque predicted in step 3, is the pushing speed, is the cutter head rotating speed, and is the tunnel diameter. The calculation result is shown in . is a prediction result and error diagram in the present application Figure 11 . Figure 11 The above is a shield construction parameter prediction method provided by one or more embodiments of the present application. Based on the same idea, the present application also provides a corresponding shield construction parameter prediction device, as shown in .

[0068] Figure 12

[0069] Figure 12 is a shield construction parameter prediction device provided by the present application, which comprises: an acquisition module 201, configured to acquire the shield machine parameters and the geological data of the corresponding area within a preset number of tunneling rings before the current tunneling ring of the shield construction as sample data, and label the real values of the shield construction parameters corresponding to the sample data; a prediction and analysis module 202, configured to input the sample data into a neural network model to obtain a predicted value of the shield construction parameters, and determine an analytical value of the shield construction parameters according to the sample data through a calculation formula of the shield construction parameters; ​​The loss determination module 203 is configured to determine data loss according to the shield construction parameter predicted value and the shield construction parameter true value, determine physical loss according to the shield construction parameter analytical value and the shield construction parameter true value, and determine data loss average value and physical loss average value according to the data loss and the physical loss of the preset training round number. The weighting module 204 is configured to determine adaptive weights of the data loss and the physical loss according to the relative sizes of the data loss average value and the physical loss average value, weight the data loss and the physical loss respectively, determine total loss, and perform multi-round training on the neural network model with the optimization target of minimizing the total loss. The prediction module 205 is configured to input the geological data of the current tunneling ring of the shield construction and the shield machine parameters into the trained neural network model to obtain the shield construction parameters of the current tunneling ring of the shield construction.

[0070] The specific limitations of the shield construction parameter prediction device can refer to the limitations of the shield construction parameter prediction method in the foregoing, and will not be described here. The various modules in the shield construction parameter prediction device can be realized by software, hardware and combinations thereof in whole or in part. The various modules described above can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to the various modules.

[0071] The present application also provides a computer readable storage medium, which stores a computer program, and the computer program can be used to execute the shield construction parameter prediction method described above. Figure 1 The shield construction parameter prediction method is provided.

[0072] The present application also provides a shield construction parameter prediction device. Figure 13 The structure diagram of the computer device is shown in FIG. 1. Figure 13 As shown in FIG. 1, at the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory and a non-volatile memory, and of course can also include other hardware required by the business. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs to realize the shield construction parameter prediction method provided above. Figure 1 The shield construction parameter prediction method is provided.

[0073] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments of the methods. In the embodiments of the present application, any reference to memory, storage, database or other medium can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0074] The technical features of the above embodiments can be combined in any way. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, but as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

Claims

1. A shield construction parameter prediction method, characterized in that: include: Obtain the shield machine parameters and geological data of the corresponding area within the preset tunneling ring number before the current tunneling ring of the shield construction as sample data; mark the actual values ​​of the shield construction parameters corresponding to the sample data; The sample data is input into the neural network model to obtain the predicted value of the shield construction parameter, and the analytical value of the shield construction parameter is determined according to the calculation formula of the shield construction parameter based on the sample data; Determine data loss based on predicted values ​​of shield construction parameters and true values ​​of shield construction parameters, and determine physical loss based on analyzed values ​​of shield construction parameters and true values ​​of shield construction parameters; determine average data loss and average physical loss based on data loss and physical loss of a preset number of training rounds; According to the relative size of the average value of data loss and the average value of physical loss, the adaptive weights of data loss and physical loss are determined respectively, and the data loss and physical loss are weighted separately to determine the total loss. The neural network model is trained for multiple rounds with minimizing the total loss as the optimization goal; The geological data of the current tunneling ring of the shield construction and the shield machine parameters are input into the trained neural network model to obtain the shield construction parameters of the current tunneling ring of the shield construction.

2. The shield construction parameter prediction method according to claim 1, characterized in that: The neural network model is obtained by sequentially connecting the LSTM model and the KAN model, and the KAN model takes the last input hidden state of the LSTM model as input; The LSTM model is used to extract the time-dependent features in the input shield machine parameter time series, and the KAN model is used to extract the mapping relationship between the time-dependent features and the shield construction parameters.

3. The shield construction parameter prediction method according to claim 1, wherein: The shield construction parameters include the total thrust of the shield machine and the cutter head torque; Determining the shield construction parameter analytical value according to the sample data through the shield construction parameter calculation formula specifically includes: The analytical values ​​of shield construction parameters are determined based on sample data using the following formula: ； ； in, is the analytical value of the total thrust, is the analytical value of the cutter head torque, is the elastic modulus of the tunnel surface soil, is the cutterhead opening ratio, is the Poisson's ratio of the excavation interface, is the amount of propulsion per single rotation, is the cutterhead radius, is the static soil pressure coefficient, is the unit weight of soil, is the depth of the tunnel, is the average soil pressure in the cabin, is the average ground pressure on the shield machine, is the length of the shield machine, is the weight of the shield machine, is the total weight of the supporting equipment. is the shear modulus of the soil, is the internal friction angle of the soil, is the thickness of the cutter head side wall, is the number of stirring rods, is the diameter of the stirring rod, is the length of the stirring rod, is the average distance between the stirring rod and the axis, Indicates the friction coefficient between the cutterhead system and the soil in the tunnel face area, Indicates the friction coefficient between the shield machine shell and the soil, Indicates the friction between the rear supporting equipment and the track, The friction coefficient between the cutter head side wall and the soil and Indicates the coefficient of friction between the stirring rod and the muddy water.

4. The shield construction parameter prediction method according to claim 1, wherein: Determining the adaptive weights of data loss and physical loss respectively based on the relative sizes of the average data loss value and the average physical loss value specifically includes: The adaptive weights of data loss and physical loss are determined according to the relative sizes of the average data loss and the average physical loss using the following formula: ； ； in, is the adaptive weight of data loss, is the adaptive weight of physical loss, is the average value of data loss, is the average value of physical loss.

5. The shield construction parameter prediction method according to claim 1, wherein: The adaptive weights of data loss and physical loss are determined based on the relative sizes of the average data loss and the average physical loss, specifically including: When the average value of data loss is less than 0 or greater than 1, the adaptive weight of data loss is reset to 0; or, when the average value of physical loss is less than 0 or greater than 1, the adaptive weight of physical loss is reset to 0; When the average value of data loss and the average value of physical loss are both between 0 and 1, the adaptive weights of data loss and physical loss are determined respectively according to the relative sizes of the average value of data loss and the average value of physical loss.

6. The shield construction parameter prediction method according to claim 1, wherein: The neural network is used to predict the total thrust and cutter head torque of the shield machine during shield construction; The method further comprises: Based on the total thrust and cutterhead torque of the shield machine during the current tunneling ring, the specific energy of the shield construction can be determined by the following formula: ; in, is the specific energy of shield construction, is the total thrust of the shield machine, is the forward speed, is the cutter head speed, is the cutter head torque of the shield machine, is the diameter of the cutterhead system.

7. The shield construction parameter prediction method according to claim 1, wherein: The step of obtaining shield machine parameters and geological data of the corresponding area within a preset number of tunneling rings before the current tunneling ring of the shield construction as sample data specifically includes: Obtain shield machine parameters and geological data of the corresponding area within the preset tunneling ring number before the current tunneling ring of shield construction; The geological data are preprocessed by the following formula to obtain weighted geological parameters: ； The shield machine parameters within the preset tunneling ring number before the current tunneling ring of shield construction and the weighted geological parameters of the corresponding area are used as sample data; in, For geological data The weighted geological parameters, The tunnel excavation face involves the first geological data of soil layers, From the inside out The central angle of the circle corresponding to the soil layer, is the tunnel radius, For the The area of ​​the soil layer corresponding to the soil layer, is the total number of soil layers involved in the tunneling face; The geological data includes: at least one of: water content, density, soil particle specific gravity, porosity, saturation, compression modulus, Poisson's ratio, horizontal reaction coefficient, vertical reaction coefficient, horizontal permeability coefficient, vertical permeability coefficient, cohesion, internal friction angle and static earth pressure coefficient.

8. A shield construction parameter prediction device, characterized in that: include: An acquisition module is used to obtain shield machine parameters and geological data of the corresponding area within a preset number of tunneling rings before the current tunneling ring of shield construction as sample data; Mark the actual values ​​of shield construction parameters corresponding to the sample data; The prediction and analysis module is used to input sample data into the neural network model to obtain the predicted value of the shield construction parameter, and determine the analytical value of the shield construction parameter according to the calculation formula of the shield construction parameter based on the sample data; The loss determination module is used to determine the data loss based on the predicted value of the shield construction parameter and the actual value of the shield construction parameter, and to determine the physical loss based on the analyzed value of the shield construction parameter and the actual value of the shield construction parameter; and to determine the average value of the data loss and the average value of the physical loss based on the data loss and physical loss of the preset number of training rounds; A weighting module is used to determine the adaptive weights of data loss and physical loss respectively according to the relative size of the average value of data loss and the average value of physical loss, and to weight the data loss and physical loss respectively to determine the total loss. The neural network model is trained for multiple rounds with the optimization goal of minimizing the total loss. The prediction module is used to input the geological data of the current tunneling ring of the shield construction and the shield machine parameters into the trained neural network model to obtain the shield construction parameters of the current tunneling ring of the shield construction.

9. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the program.

Images