Shield tunneling mud cake blockage prediction method and system, terminal and storage medium

Abstract

The invention discloses a shield tunneling mud cake blockage prediction method and system, a terminal and a storage medium, and the method comprises the steps: obtaining low-fidelity data of a target tunnel, and training a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model; acquiring high-fidelity data, and inputting the high-fidelity data into a low-fidelity model to output a low-fidelity predicted value; fusing the high-fidelity data and the low-fidelity predicted value to obtain extended input data, training a high-fidelity residual network to obtain a high-fidelity residual model, and obtaining a target multi-fidelity model according to the low-fidelity model and the high-fidelity residual model; and obtaining shield tunneling data of the target tunnel, and inputting the shield tunneling data to the target multi-fidelity model to output a mud cake blockage risk prediction result. According to the invention, preliminary modeling is carried out by using low-fidelity data, the dependence on high-fidelity data is reduced, the data processing cost is reduced, and the model prediction precision is improved through the active learning strategy, so that the mud cake blockage risk prediction precision is improved.

Description

Technical Field

[0001] The present invention relates to the field of geographic information technology, and in particular to a shield tunneling mud cake blockage prediction method, system, terminal and computer-readable storage medium. Background Art

[0002] Shield tunneling technology has become the mainstream construction method in modern tunnel construction due to its high efficiency and continuous construction capabilities. However, in the shield tunneling process, mud cake blockage has always been the main bottleneck restricting the progress of the project and safe construction. Mud cake blockage will not only significantly reduce construction efficiency and prolong the project cycle, but also may cause construction interruption, equipment failure and serious safety accidents.

[0003] At present, there are two main methods for predicting the risk of mud cake blockage: one is a model trained based on high-fidelity data (referring to the actual shield excavation data where the "mud cake" phenomenon occurs). This type of method can more accurately reflect the actual working conditions on site, but due to the high cost and difficulty of collecting high-fidelity data and the limited number of samples, the model is not stable enough when facing a complex and dynamic construction environment; the other is a simulation model built based on low-fidelity data. Although this type of method can quickly establish a model using low-fidelity data and has low data collection costs, due to the large difference between low-fidelity data and actual working conditions, its prediction results are often not accurate enough to meet the actual needs of the project.

[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0005] The main purpose of the present invention is to provide a method, system, terminal and storage medium for predicting mud cake blockage in shield tunneling, aiming to solve the problems that the existing technology based on high-fidelity data training models are not stable enough when facing complex and dynamic construction environments; the prediction results of models based on low-fidelity data are often not accurate enough to meet the actual needs of the project.

[0006] To achieve the above object, the present invention provides a method for predicting mud cake blockage in shield tunneling, and the method comprises the following steps: Acquire low-fidelity data of the target tunnel, and train a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model; Acquire high-fidelity data of the target tunnel, input the high-fidelity data into the low-fidelity model, and output a low-fidelity prediction value; The high-fidelity data is fused with the low-fidelity prediction value to obtain extended input data, and a corresponding high-fidelity residual network is trained according to the extended input data to obtain a high-fidelity residual model, and a target multi-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model; The shield excavation data of the target tunnel is obtained, the shield excavation data is input into the target multi-fidelity model, and the mud cake blockage risk prediction result is output.

[0007] Optionally, the shield tunneling mud cake blockage prediction method, wherein the step of acquiring low-fidelity data of the target tunnel and training a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model, specifically includes: Obtaining a shield mud cake risk map of a target tunnel, performing data conversion on key parameters of the shield mud cake risk map, and obtaining characteristic parameters; Performing feature simulation on the feature parameters to obtain engineering feature parameters, and combining the feature parameters with the engineering feature parameters to obtain low-fidelity data; A low-fidelity neural network is constructed, and the low-fidelity neural network is iteratively trained according to the low-fidelity data. If the number of iterations of the low-fidelity neural network reaches a first preset number, it is determined that the low-fidelity neural network has completed training, thereby obtaining a low-fidelity model.

[0008] Optionally, the shield tunneling mud cake blockage prediction method, wherein the step of obtaining high-fidelity data of the target tunnel, inputting the high-fidelity data into the low-fidelity model, and outputting a low-fidelity prediction value, specifically includes: Acquire geological drilling data along the tunnel of the target tunnel and operating parameters of a shield machine, and perform normalization processing and outlier extraction processing on the geological drilling data along the tunnel and the operating parameters to obtain high-fidelity data; The high-fidelity data is input into the low-fidelity model, and a preliminary prediction of the probability of mud cake blockage is performed on the high-fidelity data to obtain a low-fidelity prediction value.

[0009] Optionally, the shield tunneling mud cake blockage prediction method, wherein the high-fidelity data is fused with the low-fidelity prediction value to obtain extended input data, and a corresponding high-fidelity residual network is trained according to the extended input data to obtain a high-fidelity residual model, and a target multi-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model, specifically comprising: Acquire data features of the low-fidelity prediction value, concatenate the data features with various input features of the high-fidelity data to obtain a target feature vector, and obtain extended input data according to the target feature vector; Dividing the extended input data according to a preset ratio to obtain an initial training set and an unseen high-fidelity data set, and performing model training on a high-fidelity residual network according to the initial training set to obtain an initial high-fidelity residual model; Inputting the unseen high-fidelity data set into the initial high-fidelity residual model, and predicting the unseen high-fidelity data set multiple times according to the Monte Carlo random dropout method to obtain multiple prediction residuals, calculating the mean of all the prediction residuals, and calculating the corresponding variance according to the mean; Sorting the data samples of the unseen high-fidelity data set according to the variance to obtain a sorting result, and extracting samples from the sorting result according to a variance threshold to obtain a target data sample; The target data sample is combined with the initial training set according to a sampling strategy with replacement to obtain a new target training set, and the high-fidelity residual network is model trained according to the new target training set to obtain a high-fidelity residual model, and a target multi-fidelity model is obtained based on the low-fidelity model and the high-fidelity residual model.

[0010] Optionally, the shield tunneling mud cake blockage prediction method, wherein the high-fidelity residual network is trained according to the new target training set to obtain a high-fidelity residual model, and a target high-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model, specifically includes: Iteratively training the high-fidelity residual network according to the new target training set, and if the number of iterations of the high-fidelity residual network reaches a second preset number, determining that the high-fidelity residual network has completed training, thereby obtaining a high-fidelity residual model; The low-fidelity prediction value is added to all the prediction residuals to obtain a multi-fidelity prediction value, and the multi-fidelity prediction value is fitted and corrected according to a correction formula to obtain a final prediction result; A multi-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model, and parameters of the multi-fidelity model are corrected according to the final prediction result to obtain a target multi-fidelity model.

[0011] Optionally, in the shield tunneling mud cake blockage prediction method, the correction formula is: ; in, For the final prediction result, is the predicted probability output by the low-fidelity model, is the probability residual output by the high-fidelity residual network, and All are linear regression correction coefficients.

[0012] Optionally, the shield tunneling mud cake blockage prediction method, wherein the shield tunneling data of the target tunnel is obtained, the shield tunneling data is input into the target multi-fidelity model, and the mud cake blockage risk prediction result is output, and then the method further includes: If the predicted probability of the mud cake blockage risk prediction result is less than the preset probability, it is determined that the target tunnel does not have the risk of mud cake blockage; If the predicted probability of the mud cake blockage risk prediction result is greater than or equal to the preset probability, it is determined that the target tunnel has a mud cake blockage risk, and a risk warning signal is generated to alert corresponding construction personnel.

[0013] Optionally, in the shield tunneling mud cake blockage prediction method, the shield tunneling mud cake blockage prediction system comprises: A first model training module, used for acquiring low-fidelity data of a target tunnel, and training a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model; A data prediction module, used for acquiring high-fidelity data of the target tunnel, inputting the high-fidelity data into the low-fidelity model, and outputting a low-fidelity prediction value; A second model training module is used to fuse the high-fidelity data with the low-fidelity prediction value to obtain extended input data, and train a corresponding high-fidelity residual network according to the extended input data to obtain a high-fidelity residual model, and obtain a target multi-fidelity model according to the low-fidelity model and the high-fidelity residual model; The risk prediction module is used to obtain the shield excavation data of the target tunnel, input the shield excavation data into the target multi-fidelity model, and output the mud cake blockage risk prediction result.

[0014] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a shield tunneling mud cake blockage prediction program stored in the memory and executable on the processor, wherein the shield tunneling mud cake blockage prediction program implements the steps of the shield tunneling mud cake blockage prediction method as described above when executed by the processor.

[0015] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a shield tunneling mud cake blockage prediction program, and when the shield tunneling mud cake blockage prediction program is executed by a processor, the steps of the shield tunneling mud cake blockage prediction method as described above are implemented.

[0016] In the present invention, low-fidelity data of a target tunnel is obtained, and a low-fidelity neural network is trained based on the low-fidelity data to obtain a low-fidelity model; high-fidelity data of the target tunnel is obtained, and the high-fidelity data is input into the low-fidelity model to output a low-fidelity prediction value; the high-fidelity data is fused with the low-fidelity prediction value to obtain extended input data, and a corresponding high-fidelity residual network is trained based on the extended input data to obtain a high-fidelity residual model, and a target multi-fidelity model is obtained based on the low-fidelity model and the high-fidelity residual model; shield excavation data of the target tunnel is obtained, and the shield excavation data is input into the target multi-fidelity model to output a mud cake blockage risk prediction result. The present invention uses low-fidelity data for preliminary modeling, reduces dependence on high-fidelity data, and reduces data collection and processing costs. It also accelerates model convergence through an active learning strategy, and uses a residual network to correct the deficiencies of the model to improve the model prediction accuracy, thereby improving the accuracy of mud cake blockage risk prediction. The MCD (Monte-Carlo Dropout) method is used to quantify prediction uncertainty, and the training set is dynamically updated in combination with an active learning strategy to effectively capture rare mud cake blockage samples and enhance the model's recognition ability for minority class samples. It also uses low-fidelity data for preliminary modeling, reduces dependence on high-fidelity data, and reduces data collection and processing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of a preferred embodiment of the shield tunneling mud cake blockage prediction method of the present invention; Figure 2 It is a schematic diagram of the risk of shield mud cake formation in the shield tunneling mud cake blockage prediction method of the present invention; Figure 3 It is a structural diagram of a preferred embodiment of the shield tunneling mud cake blockage prediction system of the present invention; Figure 4 It is a schematic diagram of the operating environment of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0019] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back...), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0020] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0021] The shield tunneling mud cake blockage prediction method described in the preferred embodiment of the present invention is as follows: Figure 1 As shown, the shield tunneling mud cake blockage prediction method comprises the following steps: Step S10: Obtain low-fidelity data of the target tunnel, and train a low-fidelity neural network based on the low-fidelity data to obtain a low-fidelity model.

[0022] The step S10 comprises: Step S11, obtaining a shield mud cake risk map of a target tunnel, performing data conversion on key parameters of the shield mud cake risk map, and obtaining characteristic parameters; Step S12: performing feature simulation on the feature parameters to obtain engineering feature parameters, and combining the feature parameters with the engineering feature parameters to obtain low-fidelity data; Step S13: construct a low-fidelity neural network, and iteratively train the low-fidelity neural network according to the low-fidelity data. If the number of iterations of the low-fidelity neural network reaches a first preset number, it is determined that the low-fidelity neural network has completed training, thereby obtaining a low-fidelity model.

[0023] Specifically, in an embodiment of the present invention, in order to solve the problem that the existing prediction method based on high-fidelity data has insufficient generalization ability in practical applications due to the high data collection cost and limited sample number, and the model built based on low-fidelity data is easy to obtain data but insufficient in accuracy, and the mud cake blockage event accounts for a very low proportion in the overall data, and the traditional model is difficult to capture the key minority class information. Therefore, the present invention aims to significantly improve the accuracy and robustness of mud cake blockage risk prediction by integrating the advantages of low-fidelity data and high-fidelity data, and introduce an active learning strategy assisted by MCD (Monte-Carlo Dropout, Monte Carlo random dropout method), reduce data collection costs, and meet the actual needs of shield tunneling construction safety and efficiency. Specifically, first, low-fidelity data is generated using existing empirical formulas, risk maps, and kernel density estimation methods, wherein low-fidelity data refers to data with low quality, poor accuracy, or incompleteness. Such data usually contains noise, errors, or missing information, and may not fully and accurately reflect the actual situation. The generation of low-fidelity data is based on the characteristic parameters summarized in the shield mud cake risk map, and the shield mud cake risk map is as follows: Figure 2 As shown, Figure 2 Ic represents the viscosity index of the soil, and Ip represents the plasticity index of the soil. The key parameters in the shield mud cake risk map are converted into numerical forms to generate characteristic parameters (i.e., plastic limit-water content and liquid limit-water content). Then, the KDE (Kernel Density Estimation) method is used to simulate and generate other characteristics (i.e., engineering characteristic parameters) that are consistent with the actual engineering characteristics, such as cutterhead speed, cutterhead torque, total thrust and penetration, etc., to finally form complete low-fidelity data.

[0024] After obtaining the low-fidelity data, the low-fidelity neural network is trained based on the low-fidelity data. The low-fidelity data is used as input and trained using an MLP (Multi-Layer Perceptron) regression neural network (6×64×128×32×1 network structure). During training, mean square error is used as the loss function, Adam is selected as the optimizer, the learning rate is set to 0.001, and the number of training iterations is 200. After the training is completed, the obtained low-fidelity neural network model can output the probability of mud cake blockage, that is, the low-fidelity prediction value.

[0025] Step S20: acquiring high-fidelity data of the target tunnel, inputting the high-fidelity data into the low-fidelity model, and outputting a low-fidelity prediction value.

[0026] The step S20 comprises: Step S21, obtaining geological drilling data along the tunnel of the target tunnel and operating parameters of the shield machine, and performing normalization processing and outlier extraction processing on the geological drilling data along the tunnel and the operating parameters to obtain high-fidelity data; Step S22: input the high-fidelity data into the low-fidelity model, perform a preliminary prediction of the mud cake blockage probability on the high-fidelity data, and obtain a low-fidelity prediction value.

[0027] Specifically, in an embodiment of the present invention, after obtaining a low-fidelity model, it is necessary to input high-fidelity data actually collected at the shield tunneling site into the low-fidelity model to obtain a low-fidelity prediction value, and the corresponding high-fidelity data acquisition process is specifically to obtain the geological drilling data along the tunnel of the target tunnel (for example, plastic limit, liquid limit, water content, etc.) and the operating parameters of the shield machine (for example, cutter head speed, cutter head torque, total thrust, penetration, etc.), and perform normalization and outlier extraction on the geological drilling data along the tunnel and the operating parameters to obtain high-fidelity data; then, the high-fidelity data is input into the low-fidelity model, and a preliminary prediction of the probability of mud cake blockage is performed on the high-fidelity data to obtain a low-fidelity prediction value, which is used as a preliminary estimate of the probability of mud cake blockage.

[0028] Step S30: fusing the high-fidelity data with the low-fidelity prediction value to obtain extended input data, training a corresponding high-fidelity residual network based on the extended input data to obtain a high-fidelity residual model, and obtaining a target multi-fidelity model based on the low-fidelity model and the high-fidelity residual model.

[0029] The step S30 comprises: Step S31, obtaining data features of the low-fidelity prediction value, concatenating the data features with various input features of the high-fidelity data to obtain a target feature vector, and obtaining extended input data according to the target feature vector; Step S32: dividing the extended input data according to a preset ratio to obtain an initial training set and an unseen high-fidelity data set, and performing model training on a high-fidelity residual network according to the initial training set to obtain an initial high-fidelity residual model; Step S33: input the unseen high-fidelity data set into the initial high-fidelity residual model, and perform multiple predictions on the unseen high-fidelity data set according to the Monte Carlo random dropout method to obtain multiple prediction residuals, calculate the mean of all the prediction residuals, and calculate the corresponding variance according to the mean; Step S34, sorting the data samples of the unseen high-fidelity data set according to the variance to obtain a sorting result, and extracting samples from the sorting result according to the variance threshold to obtain a target data sample; Step S35: combining the target data sample with the initial training set according to a sampling strategy with replacement to obtain a new target training set, and performing model training on the high-fidelity residual network according to the new target training set to obtain a high-fidelity residual model, and obtaining a target multi-fidelity model based on the low-fidelity model and the high-fidelity residual model.

[0030] Specifically, after obtaining the low-fidelity prediction value, it is necessary to fuse the low-fidelity prediction value with the high-fidelity data to form extended input data. In an embodiment of the present invention, a feature splicing method is adopted, that is, the data features of the low-fidelity prediction value are obtained, and the data features are spliced ​​with the various input features of the high-fidelity data to obtain a target feature vector, and the extended input data is obtained according to the target feature vector. The input features in the fused extended input data are a combination of the features of the high-fidelity data and the low-fidelity prediction value, and the output features are the target features of the high-fidelity data minus the low-fidelity prediction value. Afterwards, a high-fidelity residual network is trained based on the extended input data to learn the residual between the low-fidelity prediction value and the actual high-fidelity data to obtain a high-fidelity residual model.

[0031] In this training process, the MCD method is used to perform multiple inferences on samples in the high-fidelity data, calculate the mean and variance of the predicted output, actively select samples with higher uncertainty based on the uncertainty index to update the training set, and repeatedly train the high-fidelity residual network. Specifically, the extended input data used for high-fidelity residual model training is divided into an initial training set and an unseen high-fidelity data set according to a preset ratio, wherein the initial training set is used for the preliminary training of the high-fidelity residual network, and the unseen high-fidelity data set is used for active learning and updating; first, the initial training set is used for training The high-fidelity residual network is trained according to the initial training set to obtain an initial high-fidelity residual model. Subsequently, the initial high-fidelity residual model is used to predict samples in the unseen high-fidelity data set. Specifically, the same input sample in the unseen high-fidelity data set is predicted T times (for example, 10 times) by using the Monte Carlo random discard method, and the mean and variance of the T-times predicted output are calculated, and the variance is used as an indicator to measure the uncertainty of the prediction; then, the data samples of the unseen high-fidelity data set are predicted according to the variance. Uncertainty sorting is performed to obtain a sorting result, and samples are extracted from the sorting result according to a variance threshold in order to select a certain proportion of samples with larger variance to obtain a target data sample, i.e., a data sample with the highest uncertainty, and a sampling strategy with replacement is adopted to add the sample to the initial training set, i.e., the target data sample is combined with the initial training set to obtain a new target training set; and model training is performed on the high-fidelity residual network according to the new target training set until the initial high-fidelity residual model reaches a second preset number of training times, then it is determined that the high-fidelity residual network has completed training to obtain a high-fidelity residual model; finally, the low-fidelity prediction value is added to all the prediction residuals to obtain a multi-fidelity prediction value. In order to further improve the prediction accuracy, the multi-fidelity prediction value is fitted and corrected by linear regression. Specifically, the multi-fidelity prediction value is fitted and corrected according to a correction formula to obtain a final prediction result, a multi-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model, and the parameters of the multi-fidelity model are corrected according to the final prediction result to obtain a target multi-fidelity model.

[0032] Wherein, the correction formula is: ;in, For the final prediction result, is the predicted probability output by the low-fidelity model, is the probability residual output by the high-fidelity residual network, and All are linear regression correction coefficients.

[0033] Step S40, obtaining shield tunneling data of the target tunnel, inputting the shield tunneling data into the target multi-fidelity model, and outputting a mud cake blockage risk prediction result.

[0034] Specifically, in an embodiment of the present invention, the trained target multi-fidelity model is used to predict new data collected in real time at the shield tunneling site, and the probability of mud cake blockage is output. Specifically, the shield tunneling data of the target tunnel is obtained, the shield tunneling data is input into the target multi-fidelity model, and the mud cake blockage risk prediction result is output.

[0035] Furthermore, if the predicted probability of the mud cake blockage risk prediction result is less than a preset probability (for example, 50%), it is determined that the target tunnel does not have a mud cake blockage risk; if the predicted probability of the mud cake blockage risk prediction result is greater than or equal to the preset probability, it is determined that the target tunnel has a mud cake blockage risk, and a risk warning signal is generated to remind the corresponding construction personnel, thereby providing timely decision support for the construction site to ensure construction safety and efficiency. The present invention uses low-fidelity data for preliminary modeling, reduces dependence on high-fidelity data, and reduces data acquisition and processing costs; it also accelerates model convergence through active learning strategies, and uses residual networks to correct model deficiencies to improve the prediction accuracy of the model, thereby improving the accuracy of mud cake blockage risk prediction; the Monte Carlo Dropout method is used to quantify prediction uncertainty, and the training set is dynamically updated in combination with active learning strategies to effectively capture rare mud cake blockage samples and enhance the model's recognition ability for minority class samples; it also uses low-fidelity data for preliminary modeling to reduce dependence on high-fidelity data and reduce data acquisition and processing costs.

[0036] Furthermore, if Figure 3 As shown, based on the above-mentioned shield tunneling mud cake blockage prediction method, the present invention also provides a shield tunneling mud cake blockage prediction system, wherein the shield tunneling mud cake blockage prediction system includes: A first model training module 51 is used to obtain low-fidelity data of a target tunnel, and train a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model; A data prediction module 52, configured to obtain high-fidelity data of the target tunnel, input the high-fidelity data into the low-fidelity model, and output a low-fidelity prediction value; The second model training module 53 is used to fuse the high-fidelity data with the low-fidelity prediction value to obtain extended input data, and train the corresponding high-fidelity residual network according to the extended input data to obtain a high-fidelity residual model, and obtain a target multi-fidelity model according to the low-fidelity model and the high-fidelity residual model; The risk prediction module 54 is used to obtain the shield tunneling data of the target tunnel, input the shield tunneling data into the target multi-fidelity model, and output the mud cake blockage risk prediction result.

[0037] Furthermore, if Figure 4 As shown, based on the above-mentioned shield tunneling mud cake blockage prediction method, the present invention also provides a terminal accordingly, and the terminal includes a processor 10, a memory 20 and a display 30. Figure 4 Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0038] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal. Further, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a shield excavation mud cake blockage prediction program 40 is stored on the memory 20, and the shield excavation mud cake blockage prediction program 40 can be executed by the processor 10, thereby realizing the shield excavation mud cake blockage prediction method in the present application.

[0039] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, which is used to run the program code or process data stored in the memory 20, such as executing the shield tunneling mud cake blockage prediction method.

[0040] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 30 is used to display information on the terminal and to display a visual user interface. The components of the terminal communicate with each other via a system bus.

[0041] In one embodiment, when the processor 10 executes the shield tunneling mud cake blockage prediction program 40 in the memory 20, the following steps are implemented: Acquire low-fidelity data of the target tunnel, and train a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model; Acquire high-fidelity data of the target tunnel, input the high-fidelity data into the low-fidelity model, and output a low-fidelity prediction value; The high-fidelity data is fused with the low-fidelity prediction value to obtain extended input data, and a corresponding high-fidelity residual network is trained according to the extended input data to obtain a high-fidelity residual model, and a target multi-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model; The shield excavation data of the target tunnel is obtained, the shield excavation data is input into the target multi-fidelity model, and the mud cake blockage risk prediction result is output.

[0042] The step of acquiring low-fidelity data of the target tunnel and training a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model specifically includes: Obtaining a shield mud cake risk map of a target tunnel, performing data conversion on key parameters of the shield mud cake risk map, and obtaining characteristic parameters; Performing feature simulation on the feature parameters to obtain engineering feature parameters, and combining the feature parameters with the engineering feature parameters to obtain low-fidelity data; A low-fidelity neural network is constructed, and the low-fidelity neural network is iteratively trained according to the low-fidelity data. If the number of iterations of the low-fidelity neural network reaches a first preset number, it is determined that the low-fidelity neural network has completed training, thereby obtaining a low-fidelity model.

[0043] The step of obtaining high-fidelity data of the target tunnel, inputting the high-fidelity data into the low-fidelity model, and outputting a low-fidelity prediction value specifically includes: Acquire geological drilling data along the tunnel of the target tunnel and operating parameters of a shield machine, and perform normalization processing and outlier extraction processing on the geological drilling data along the tunnel and the operating parameters to obtain high-fidelity data; The high-fidelity data is input into the low-fidelity model, and a preliminary prediction of the probability of mud cake blockage is performed on the high-fidelity data to obtain a low-fidelity prediction value.

[0044] The step of fusing the high-fidelity data with the low-fidelity prediction value to obtain extended input data, training a corresponding high-fidelity residual network according to the extended input data to obtain a high-fidelity residual model, and obtaining a target multi-fidelity model according to the low-fidelity model and the high-fidelity residual model specifically includes: Acquire data features of the low-fidelity prediction value, concatenate the data features with various input features of the high-fidelity data to obtain a target feature vector, and obtain extended input data according to the target feature vector; Dividing the extended input data according to a preset ratio to obtain an initial training set and an unseen high-fidelity data set, and performing model training on a high-fidelity residual network according to the initial training set to obtain an initial high-fidelity residual model; Inputting the unseen high-fidelity data set into the initial high-fidelity residual model, and predicting the unseen high-fidelity data set multiple times according to the Monte Carlo random dropout method to obtain multiple prediction residuals, calculating the mean of all the prediction residuals, and calculating the corresponding variance according to the mean; Sorting the data samples of the unseen high-fidelity data set according to the variance to obtain a sorting result, and extracting samples from the sorting result according to a variance threshold to obtain a target data sample; The target data sample is combined with the initial training set according to a sampling strategy with replacement to obtain a new target training set, and the high-fidelity residual network is model trained according to the new target training set to obtain a high-fidelity residual model, and a target multi-fidelity model is obtained based on the low-fidelity model and the high-fidelity residual model.

[0045] The step of training the high-fidelity residual network according to the new target training set to obtain a high-fidelity residual model, and obtaining a target high-fidelity model according to the low-fidelity model and the high-fidelity residual model, specifically includes: Iteratively training the high-fidelity residual network according to the new target training set, and if the number of iterations of the high-fidelity residual network reaches a second preset number, determining that the high-fidelity residual network has completed training, thereby obtaining a high-fidelity residual model; The low-fidelity prediction value is added to all the prediction residuals to obtain a multi-fidelity prediction value, and the multi-fidelity prediction value is fitted and corrected according to a correction formula to obtain a final prediction result; A multi-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model, and parameters of the multi-fidelity model are corrected according to the final prediction result to obtain a target multi-fidelity model.

[0046] Wherein, the correction formula is: ; in, For the final prediction result, is the predicted probability output by the low-fidelity model, is the probability residual output by the high-fidelity residual network, and All are linear regression correction coefficients.

[0047] The step of obtaining shield tunneling data of the target tunnel, inputting the shield tunneling data into the target multi-fidelity model, and outputting a mud cake blockage risk prediction result further includes: If the predicted probability of the mud cake blockage risk prediction result is less than the preset probability, it is determined that the target tunnel does not have the risk of mud cake blockage; If the predicted probability of the mud cake blockage risk prediction result is greater than or equal to the preset probability, it is determined that the target tunnel has a mud cake blockage risk, and a risk warning signal is generated to alert corresponding construction personnel.

[0048] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a shield tunneling mud cake blockage prediction program, and when the shield tunneling mud cake blockage prediction program is executed by a processor, the steps of the shield tunneling mud cake blockage prediction method as described above are implemented.

[0049] In summary, the present invention provides a method, system, terminal and storage medium for predicting mud cake blockage in shield tunneling. The method comprises: obtaining low-fidelity data of a target tunnel, training a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model; obtaining high-fidelity data of the target tunnel, inputting the high-fidelity data into the low-fidelity model, and outputting a low-fidelity prediction value; fusing the high-fidelity data with the low-fidelity prediction value to obtain extended input data, and training a corresponding high-fidelity residual network according to the extended input data to obtain a high-fidelity residual model, and obtaining a target multi-fidelity model according to the low-fidelity model and the high-fidelity residual model; obtaining shield tunneling data of the target tunnel, inputting the shield tunneling data into the target multi-fidelity model, and outputting a mud cake blockage risk prediction result. The present invention combines low-fidelity data with high-fidelity data through a multi-fidelity model framework, which can effectively improve the efficiency of model training. By fully integrating low-fidelity data with high-fidelity data and using a residual network to correct the deficiencies of the low-fidelity model, the accuracy of mud cake blockage risk prediction can be significantly improved. The MCD (Monte-Carlo Dropout) method is used to quantify prediction uncertainty, and the training set is dynamically updated in combination with an active learning strategy to effectively capture rare mud cake blockage samples and enhance the model's recognition ability for minority class samples. Low-fidelity data is also used for preliminary modeling to reduce dependence on high-fidelity data and reduce data acquisition and processing costs.

[0050] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or terminal including the element.

[0051] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium that can be read by a computer, and the program can include the processes of the above-mentioned method embodiments when executed. The computer-readable storage medium can be a memory, a disk, an optical disk, etc.

[0052] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A method for predicting mud cake blockage in shield tunneling, characterized in that: The shield tunneling mud cake blockage prediction method comprises: Acquire low-fidelity data of the target tunnel, and train a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model; Acquire high-fidelity data of the target tunnel, input the high-fidelity data into the low-fidelity model, and output a low-fidelity prediction value; The high-fidelity data is fused with the low-fidelity prediction value to obtain extended input data, a corresponding high-fidelity residual network is trained according to the extended input data to obtain a high-fidelity residual model, and a target multi-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model; The shield excavation data of the target tunnel is obtained, the shield excavation data is input into the target multi-fidelity model, and the mud cake blockage risk prediction result is output.

2. The shield tunneling mud cake blockage prediction method according to claim 1 is characterized in that: The step of acquiring low-fidelity data of the target tunnel and training a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model specifically includes: Obtaining a shield mud cake risk map of a target tunnel, performing data conversion on key parameters of the shield mud cake risk map, and obtaining characteristic parameters; Performing feature simulation on the feature parameters to obtain engineering feature parameters, and combining the feature parameters with the engineering feature parameters to obtain low-fidelity data; A low-fidelity neural network is constructed, and the low-fidelity neural network is iteratively trained according to the low-fidelity data. If the number of iterations of the low-fidelity neural network reaches a first preset number, it is determined that the low-fidelity neural network has completed training, thereby obtaining a low-fidelity model.

3. The shield tunneling mud cake blockage prediction method according to claim 1, characterized in that: The acquiring high-fidelity data of the target tunnel, inputting the high-fidelity data into the low-fidelity model, and outputting a low-fidelity prediction value specifically includes: Acquire geological drilling data along the tunnel of the target tunnel and operating parameters of a shield machine, and perform normalization processing and outlier extraction processing on the geological drilling data along the tunnel and the operating parameters to obtain high-fidelity data; The high-fidelity data is input into the low-fidelity model, and a preliminary prediction of the probability of mud cake blockage is performed on the high-fidelity data to obtain a low-fidelity prediction value.

4. The shield tunneling mud cake blockage prediction method according to claim 1, characterized in that: The step of fusing the high-fidelity data with the low-fidelity prediction value to obtain extended input data, training a corresponding high-fidelity residual network according to the extended input data to obtain a high-fidelity residual model, and obtaining a target multi-fidelity model according to the low-fidelity model and the high-fidelity residual model specifically includes: Acquire data features of the low-fidelity prediction value, concatenate the data features with various input features of the high-fidelity data to obtain a target feature vector, and obtain extended input data according to the target feature vector; Dividing the extended input data according to a preset ratio to obtain an initial training set and an unseen high-fidelity data set, and performing model training on a high-fidelity residual network according to the initial training set to obtain an initial high-fidelity residual model; Inputting the unseen high-fidelity data set into the initial high-fidelity residual model, and predicting the unseen high-fidelity data set multiple times according to the Monte Carlo random dropout method to obtain multiple prediction residuals, calculating the mean of all the prediction residuals, and calculating the corresponding variance according to the mean; Sorting the data samples of the unseen high-fidelity data set according to the variance to obtain a sorting result, and extracting samples from the sorting result according to a variance threshold to obtain a target data sample; The target data sample is combined with the initial training set according to a sampling strategy with replacement to obtain a new target training set, and the high-fidelity residual network is model trained according to the new target training set to obtain a high-fidelity residual model, and a target multi-fidelity model is obtained based on the low-fidelity model and the high-fidelity residual model.

5. The shield tunneling mud cake blockage prediction method according to claim 4 is characterized in that: The performing model training on the high-fidelity residual network according to the new target training set to obtain a high-fidelity residual model, and obtaining a target high-fidelity model according to the low-fidelity model and the high-fidelity residual model, specifically includes: Iteratively training the high-fidelity residual network according to the new target training set, and if the number of iterations of the high-fidelity residual network reaches a second preset number, determining that the high-fidelity residual network has completed training, thereby obtaining a high-fidelity residual model; The low-fidelity prediction value is added to all the prediction residuals to obtain a multi-fidelity prediction value, and the multi-fidelity prediction value is fitted and corrected according to a correction formula to obtain a final prediction result; A multi-fidelity model is obtained according to the low-fidelity model and the high-fidelity residual model, and parameters of the multi-fidelity model are corrected according to the final prediction result to obtain a target multi-fidelity model.

6. The shield tunneling mud cake blockage prediction method according to claim 5, characterized in that: The correction formula is: ； in, For the final prediction result, is the predicted probability output by the low-fidelity model, is the probability residual output by the high-fidelity residual network, and All are linear regression correction coefficients.

7. The shield tunneling mud cake blockage prediction method according to claim 1, characterized in that: The step of obtaining shield tunneling data of the target tunnel, inputting the shield tunneling data into the target multi-fidelity model, and outputting a mud cake blockage risk prediction result further includes: If the predicted probability of the mud cake blockage risk prediction result is less than the preset probability, it is determined that the target tunnel does not have the risk of mud cake blockage; If the predicted probability of the mud cake blockage risk prediction result is greater than or equal to the preset probability, it is determined that the target tunnel has a mud cake blockage risk, and a risk warning signal is generated to alert corresponding construction personnel.

8. A shield tunneling mud cake blockage prediction system, characterized in that: The shield tunneling mud cake blockage prediction system comprises: A first model training module, used for acquiring low-fidelity data of a target tunnel, and training a low-fidelity neural network according to the low-fidelity data to obtain a low-fidelity model; A data prediction module, used for acquiring high-fidelity data of the target tunnel, inputting the high-fidelity data into the low-fidelity model, and outputting a low-fidelity prediction value; A second model training module is used to fuse the high-fidelity data with the low-fidelity prediction value to obtain extended input data, and train a corresponding high-fidelity residual network according to the extended input data to obtain a high-fidelity residual model, and obtain a target multi-fidelity model according to the low-fidelity model and the high-fidelity residual model; The risk prediction module is used to obtain the shield excavation data of the target tunnel, input the shield excavation data into the target multi-fidelity model, and output the mud cake blockage risk prediction result.

9. A terminal, characterized in that: The terminal includes: a memory, a processor, and a shield tunneling mud cake blockage prediction program stored in the memory and executable on the processor. When the shield tunneling mud cake blockage prediction program is executed by the processor, the steps of the shield tunneling mud cake blockage prediction method as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a shield tunneling mud cake blockage prediction program, and when the shield tunneling mud cake blockage prediction program is executed by a processor, the steps of the shield tunneling mud cake blockage prediction method according to any one of claims 1 to 7 are implemented.

Images