Slurry balance shield tunneling machine stagnant drainage prediction model and method based on multi-target different-weight regression

Abstract

The invention relates to the technical field of shield construction, in particular to a slurry balance shield tunneling machine stagnant drainage prediction model based on multi-target different-weight regression and a construction method thereof, and the construction method comprises the following steps: S1, collecting tunneling data, namely operation parameters, of a slurry balance shield tunneling machine; s2, grouping the tunneling data and performing Kalman filtering processing to obtain an accurate operation parameter set; s3, analyzing linear correlation degrees of different accurate operation parameters in the accurate operation parameter set by using Pearson's correlation to obtain correlation coefficients among the accurate operation parameters; s4, screening the accurate operation parameters corresponding to the correlation coefficients of which the absolute values are greater than a set threshold value to obtain a prediction factor set; s5, performing multi-target different-weight regression analysis on the prediction factor set, and constructing a stagnant elimination prediction model; s6, optimizing the stagnant elimination prediction model by using an ADMA algorithm; and S7, performing simulation verification on the optimized stagnant discharge prediction model by using Flomaster, and adjusting parameters of the stagnant discharge prediction model. The stagnant discharge prediction model can accurately predict the stagnant discharge phenomenon and can predict the stagnant discharge occurrence position.

Description

Technical Field

[0001] The invention relates to the technical field of shield construction, and in particular to a slurry balance shield machine stagnation prediction model based on multi-objective heterogeneous weight regression and a construction method thereof. Background Art

[0002] With the development of underground space construction, the quality requirements of shield construction are gradually increasing. The mud and water circulation system in the shield machine plays a vital role in the entire shield excavation process.

[0003] During the construction of a slurry shield, the slurry needs to be discharged through pipes. When the slurry shield enters the clay layer, the excavated clay will adhere to the cutterhead and the cabin or the slurry discharge pipeline, causing serious blockage problems. At the same time, due to various geological environments and construction factors, blockage and slurry stagnation are prone to occur in the slurry bin, air cushion bin and slurry discharge pipe. The occurrence of slurry stagnation will cause structural wear, difficulty in advancing the shield machine, and even cause landslides, seriously affecting construction efficiency and engineering safety. Therefore, the prediction and diagnosis of slurry stagnation problems of slurry shields will help to significantly improve the excavation efficiency and mechanical performance of shield machines.

[0004] The phenomenon of slurry shield machine can be divided into three categories according to the location of the slurry, namely, slurry in front of the cutterhead, slurry in the cutting chamber, and slurry in the mucking system. At present, the research on the prediction of slurry shield machine slurry is still relatively imperfect in China. It mainly analyzes the slurry phenomenon based on the slurry system pressure data by analyzing the construction balance principle of large-diameter slurry shield, and constructs a slurry warning model. This model divides the pressure difference into several levels according to the changes in the difference between the bubble chamber pressure and the shaft pressure of the excavation chamber and the suction pressure of the slurry pump. It does not fully consider different geological conditions, and the pressure difference threshold is relatively loose. It cannot accurately predict the slurry phenomenon, and cannot distinguish the specific slurry location of the shield machine. It still has certain defects. Summary of the invention

[0005] The object of the present invention is to provide a slurry shield machine drainage prediction model based on multi-objective heterogeneous weight regression and a construction method thereof, so as to solve the problems mentioned in the above background technology.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a method for constructing a slurry shield machine stagnation prediction model based on multi-objective heterogeneous weight regression, the construction method comprising the following steps:

[0007] S1. Collect the tunneling data of the slurry shield machine, i.e., the operating parameters;

[0008] S2, grouping the excavation data and performing Kalman filtering to obtain an accurate operating parameter set;

[0009] S3, using Pearson correlation analysis to analyze the linear correlation degree of different precise operating parameters in the precise operating parameter set, and obtaining the correlation coefficient between each precise operating parameter;

[0010] S4, screening the precise operating parameters corresponding to the correlation coefficients whose absolute values ​​are greater than the set threshold value to obtain a set of prediction factors;

[0011] S5. Perform multi-objective weighted regression analysis on the prediction factor set to construct a delay prediction model;

[0012] S6. Optimize the delay prediction model using ADMA algorithm;

[0013] S7. Use Flomaster to simulate and verify the optimized drainage prediction model and adjust the parameters of the drainage prediction model.

[0014] The present invention constructs a slurry shield machine drainage prediction model through the above-mentioned construction method, screens and establishes connections between the prediction factors in different drainage states and normal operation states of the slurry shield machine, utilizes the parameters collected during the operation of the shield machine, distinguishes the drainage state through multi-objective different-weight regression, further adjusts the control threshold of the drainage prediction model, improves the accuracy of drainage prediction, and verifies the accuracy of drainage prediction through pipe network simulation.

[0015] Preferably, in step S1, the excavation data includes propulsion speed, slurry inflow rate, slurry discharge rate, bin pressure, slurry discharge density and slurry discharge pump pressure.

[0016] The above-mentioned excavation data include all data generated by the slurry shield machine during its operation. The operation process includes the propulsion state, shutdown state and assembly state of the slurry shield machine. Among them, the propulsion state can be further divided into normal propulsion, cutterhead front discharge, cutting chamber discharge and mucking system discharge conditions. The above-mentioned excavation data are collected and subsequently sorted and screened.

[0017] Preferably, the step S2 specifically includes the following steps:

[0018] S21. Define the state matrix of the operating parameters of the slurry shield machine at time k, and its mathematical expression is as follows:

[0019] As shown in formula 1:

[0020]

[0021] In formula 1, Q out is the main slurry discharge flow, P st is the pressure of the mud tank, P b is the pressure difference between the air cushion chamber pressure and the slurry pump inlet pressure, T bis the propulsion force, ··· are the remaining operating parameters;

[0022] S22, after repeated prediction and update recursive process, the mathematical expression of the optimal value of Kalman filter is obtained, as shown in Formula 2:

[0023]

[0025] In formula 2, For accurate operating parameters, is the prediction of the operating parameters of the slurry shield machine at time k, K k is the Kalman gain, z k is the observation vector

[0026] Preferably, in step S3, the mathematical expression of the Pearson correlation analysis is as shown in Formula 3:

[0027]

[0028] In formula 3, and They represent the values ​​of two different precise operating parameters in the precise operating parameter set at time k, and They are and is the average value of , and r is the correlation coefficient between two different precise operating parameters.

[0029] The prediction factors will vary depending on the construction conditions, machine type, and include but are not limited to the following excavation data:

[0030] A. Main slurry flow rate Q in

[0031] B. Main slurry discharge flow Q out

[0032] C. The pressure difference P between the top pressure of the air cushion tank and the right middle pressure of the mud and water tank a

[0033] D. The pressure difference P between the air cushion tank pressure and the slurry pump inlet pressure b

[0034] E. Propulsion speed v t

[0035] F. Slurry density ρ out .

[0036] Preferably, in step S4, the threshold is set to 0.8-1, and more preferably, the threshold is set to 0.8.

[0037] By setting the threshold above, parameters with absolute values ​​of correlation coefficients greater than the set threshold and having practical engineering application significance are screened, highlighting the parameters that have a significant impact on the drainage lag phenomenon. The larger the set threshold, the more significant the impact of the screened parameters on the drainage lag phenomenon.

[0038] Preferably, the step S5 specifically includes the following steps:

[0039] S51, setting each working condition of the slurry shield machine to C, the mathematical expression of C is shown in Formula 4:

[0040] C={C1,C2,C3,C4}

[0041] Formula 4

[0042] In formula 4, C1, C2, C3, and C4 represent normal propulsion, front discharge of the cutter head, discharge of the cutting chamber, and discharge of the mucking system, respectively.

[0043] For multiple predictor sets X k Perform data standardization, randomly divide the standardized data into training data and test data according to a certain ratio, and divide the training data and test data into data sets X according to the working conditions at time k. 1jk , X 2jk , X 3jk , X 4jk , are the kth values ​​of the jth prediction factor under normal advancement, front cutterhead lag, cutting chamber lag, and mucking system lag conditions, respectively.

[0044] The X k Represents a set of multiple prediction factors obtained at different times k,

[0045] The ratio of the training data to the test data is (2-4):1;

[0046] S52, set the initial weight value of each working condition prediction factor, and establish a multi-objective different-weight regression model, whose mathematical expression is shown in Formula 5:

[0047]

[0048] In formula 5, w ij Represents the weight of the jth prediction factor under the i-th operating condition category;

[0049] For each working condition C i , define independent regression target values Said The mathematical expression of is shown in Equation 6:

[0050]

[0051] In Formula 6, X is the vector set of prediction factors, W i is the weight vector under the i-th working condition category,

[0052] The delay prediction model selects the fault category with the largest output target value as the delay diagnosis result, and its mathematical expression is shown in Formula 7:

[0053]

[0054] In formula 7, This is the predicted delay category.

[0055] The training data is used to determine the weight of each operating condition in step S52, and the test data is used to achieve the requirement that the minimum prediction accuracy of each operating condition category in the subsequent step S67 can reach 80%.

[0056] The ratio of the above training data to test data is preferably 7:3.

[0057] Preferably, step S6 specifically includes the following steps:

[0058] S61, define the loss function as cross entropy loss to measure the difference between the delay diagnosis result and the true label. The mathematical expression of the loss function is shown in Formula 8:

[0059]

[0060] In formula 8, k is the total number of sample data, i is the total number of working condition categories; if the i-th sample is of category c, y ic If yes, it is 1, otherwise it is 0; p ic is the probability that the i-th sample is diagnosed as operating condition category c;

[0061] S62, initializing ADMA algorithm parameters, wherein the ADMA algorithm parameters include: model weight coefficient θ, first-order moment estimate m, second-order moment estimate v, learning rate α, hyperparameters β1, β2, constant ∈, and number of iterations k;

[0062] S63. In each iteration, the gradient of the loss function is calculated by back propagation Update the first-order moment estimate m t and the second-order moment estimate v t , and calculate the bias-corrected first-order moment and the second moment t is the current iteration number;

[0063] S64, using the corrected moment estimate to update the weight coefficient of the delay prediction model, the weight coefficient matrix is ​​shown in Formula 9:

[0064]

[0065] In formula 9, θ t+1 is the updated weight coefficient of the delay prediction model,

[0066] S65, based on the optimized weight coefficient, calculate the loss function through the training data to obtain the fitness value of the current model;

[0067] S66, judging whether the number of iterations reaches the set number of iterations, if so, stopping the iteration and outputting the weight coefficient matrix of the delay prediction model optimized for each parameter; if not, going to step S63 to continue iterative optimization;

[0068] S67. Substitute the divided test data into the optimized delay prediction model for testing, and obtain the prediction accuracy of the delay prediction model. If the minimum prediction accuracy of each operating condition category can reach the set minimum accuracy value, the optimization is completed; if not, go to step S62 to adjust the learning rate α, hyperparameters β1, β2, and the number of iterations k, and continue iterative optimization.

[0069] Preferably, the hyperparameter β1 is set to 0.9, the hyperparameter β2 is set to 0.999, the constant ∈ is set to 1e-8, and the initial learning rate α is set to 0.01.

[0070] The above-mentioned set number of iterations is usually determined according to the convergence speed of the algorithm and the computing resources, and is preferably 100 times in the present invention; the above-mentioned set minimum accuracy value can be flexibly adjusted according to the actual shield tunneling data quality, and the model parameters can be adjusted multiple times to seek the best accuracy, but the minimum accuracy requirement must also reach 50% to ensure that the prediction model has practical engineering significance, and is preferably 80% in the present invention.

[0071] Preferably, step S7 specifically includes the following steps:

[0072] S71. Establish a one-dimensional model of the slurry shield machine through Flomaster, and import the above-mentioned drainage prediction model through the NewScript Based N-Arm Component script;

[0073] S72, setting different input boundary conditions and fluid properties, simulating the performance of the shield machine under different geological conditions, and observing the prediction accuracy of the delay prediction model for delay prediction through data post-processing;

[0074] S73. If the prediction accuracy is lower than the set prediction accuracy, the excavation data generated by the Flomaster simulation is returned to step S6 to further optimize the delay prediction model. If the prediction accuracy is greater than or equal to the set prediction accuracy, the delay prediction model is completed.

[0075] The above-mentioned set prediction accuracy can also be flexibly adjusted according to the actual shield tunneling data quality, and can be the same as or slightly smaller than the set minimum accuracy value. This is because the shield machine tunneling data used in Flomaster simulation is derived from the engineering, and many low-quality boundary conditions are ideally set in the software. The set prediction accuracy may not be particularly well adapted to the set minimum accuracy value. The set prediction accuracy can appropriately reduce the accuracy threshold of 5%-10% compared to the set minimum accuracy value. In the present invention, the set prediction accuracy is preferably 80%.

[0076] A drainage prediction model constructed by the construction method described above.

[0077] The beneficial effects of the present invention are:

[0078] 1. The present invention measures the pressure difference P of multiple key parameters (such as main slurry discharge flow, top pressure of air cushion bin and right middle pressure of mud and water bin) during the operation of shield machine. a ) was used for Pearson correlation analysis, with 80% as the screening threshold, and several tunneling data with greater influence on the accuracy of drainage prediction were screened in combination with engineering experience. As auxiliary parameters, it broke through the drainage prediction model based solely on the slurry shield pressure system and enhanced the reliability of the prediction.

[0079] 2. The present invention establishes a model based on a multi-objective different-weight regression method to ensure that each feature has a separate weight coefficient for different delay categories, which is closer to the actual situation. The weight coefficient of the model is continuously optimized through the ADMA algorithm to improve accuracy.

[0080] 3. The present invention uses Flomaster to simulate and verify the drainage model, effectively verifying the universality of the model under different geological conditions, and possessing dynamic adaptability, wide applicability and feedback optimization mechanism. The simulation can be adaptively adjusted according to different working conditions and fluid characteristics, and simulated under various geological conditions such as clay layers and sandy layers to ensure accuracy; if the prediction accuracy does not reach the set 80%, the simulation data can be fed back to the optimization stage to generate a prediction model that is more suitable for the new geological conditions, thereby improving its reliability and efficiency in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0082] Figure 1It is a flow chart of the method for constructing a slurry shield machine drainage prediction model based on multi-objective heterogeneous weight regression of the present invention. DETAILED DESCRIPTION

[0083] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0084] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0085] The present invention is now further described in conjunction with specific examples. The following examples are only for explaining the present invention, but do not constitute a limitation of the present invention. The test samples and test processes used in the following examples include the following (if the specific experimental conditions are not indicated in the examples, they are usually in accordance with conventional conditions or the conditions recommended by the reagent company; the reagents, consumables, etc. used in the following examples, if not otherwise specified, can all be obtained from commercial channels).

[0086] like Figure 1 As shown in FIG, a method for constructing a slurry shield machine delay prediction model based on multi-objective heterogeneous weight regression is provided. The specific steps of the construction method are as follows:

[0087] S1. Collect the excavation data of the slurry shield machine, that is, the operating parameters.

[0088] The excavation data refers to all data generated during the operation of the slurry shield machine, which includes but is not limited to the advancement, shutdown and assembly of the slurry shield machine. The collected excavation data are exemplified by the following data: advancement speed, slurry inflow rate, slurry discharge rate, bin pressure, slurry discharge density and slurry discharge pump pressure.

[0089] S2, grouping the excavation data and performing Kalman filtering to obtain an accurate operating parameter set, the specific steps are as follows:

[0090] S21. Define the state matrix of the operating parameters of the slurry shield machine at time k, and its mathematical expression is as follows:

[0091] As shown in formula 1:

[0092]

[0093] In formula 1, Q out is the main slurry discharge flow, P st is the pressure of the mud tank, P b is the pressure difference between the air cushion chamber pressure and the slurry pump inlet pressure, T b is the propulsion force, ··· are the remaining operating parameters;

[0094] S22, after repeated prediction and update recursive process, the mathematical expression of the optimal value of Kalman filter is obtained, as shown in Formula 2:

[0095]

[0096] In formula 2, For accurate operating parameters, is the prediction of the operating parameters of the slurry shield machine at time k, K k is the Kalman gain, z k is the observation vector.

[0097] The recursive process of repeated prediction and updating is as follows:

[0098] The Kalman filter assumes that the system can be described by a state-space model:

[0099] State update equation (prediction process):

[0100] x k =F k x k-1 +B k u k +w k

[0101] x k is the state vector of the system at time k (such as flow, pressure, etc.); F k B is the state transition matrix, which describes how the system transfers from the state at time k-1 to the state at time k; k is the control matrix, describing how the input affects the state change; u k is the input control vector; w k is the process noise, usually assumed to be zero-mean Gaussian white noise with a variance of Q k .

[0102] Observation equation (correction process):

[0103] z k =H k x k +v k

[0104] z k is the observation vector, i.e. the data measured by the sensor; H kis the observation matrix, describing how the state is mapped to the observation space; v k Assume that the noise is zero-mean Gaussian white noise with a variance of R k .

[0105] The Kalman filter has two main steps: prediction and update, which are performed alternately.

[0106] (1) Prediction Steps

[0107] In this step, the state estimate at the previous moment and the system model are used to predict the state at the current moment.

[0108] Status prediction:

[0109]

[0110] is the prediction of the state at time k.

[0111] Covariance prediction:

[0112]

[0113] is the predicted error covariance, which represents the uncertainty of the state estimate.

[0114] (2) Update steps

[0115] In this step, the predicted state is revised using the current measurement data to obtain a more accurate estimate.

[0116] Calculate the Kalman gain:

[0117]

[0118] Kalman gain K k Determines the degree of influence of the measurement on the state estimate. The larger the gain, the stronger the correction effect of the measurement on the estimate.

[0119] Status Update:

[0120]

[0121] Using the measured value z k Status of prediction Make corrections to get the updated state estimate

[0122] Covariance update:

[0123]

[0124] Update error covariance P k, represents the uncertainty of the new estimate.

[0125] Before Kalman filtering, the collected excavation data are grouped according to the working conditions of the shield machine. The above working conditions refer to the three states of the slurry shield machine: advancement, shutdown and assembly. Among them, the advancement state can be further divided into normal advancement, cutterhead front stagnation, cutting chamber stagnation and mucking system stagnation. Since short-term fluctuations or abnormal peaks may occur during the operation of the slurry shield machine, Kalman filtering is used to smooth the above grouped data to remove short-term fluctuations and retain long-term trends.

[0126] S3. Use Pearson correlation analysis to analyze the linear correlation degree of different precise operating parameters in the precise operating parameter set to obtain the correlation coefficient between each precise operating parameter. The mathematical expression of the Pearson correlation analysis is as follows:

[0127] As shown in formula 3:

[0128]

[0129] In formula 3, and They represent the values ​​of two different precise operating parameters in the precise operating parameter set at time k, and They are and is the average value of , and r is the correlation coefficient between two different precise operating parameters.

[0130] S4. Screen the precise operating parameters corresponding to the correlation coefficients whose absolute values ​​are greater than a set threshold and have practical engineering application significance to obtain a set of prediction factors. In this embodiment, the threshold is set to 0.8.

[0131] The prediction factors are the precise operating parameters after screening. The precise operating parameters are processed by Kalman filtering and are calibrated operating parameters. Therefore, the prediction factors will vary due to construction conditions and machine models, including but not limited to the following excavation data:

[0132] A. Main slurry flow rate Q in

[0133] B. Main slurry discharge flow Q out

[0134] C. The pressure difference P between the top pressure of the air cushion tank and the right middle pressure of the mud and water tank a

[0135] D. The pressure difference P between the air cushion tank pressure and the slurry pump inlet pressure b

[0136] E. Propulsion speed v t

[0137] F. Slurry density ρ out .

[0138] Based on the set of prediction factors, a multi-objective regression analysis with different weights is performed. In order to optimize the loss function and find the best model parameters, ADMA (Adaptive Moment Estimation, an adaptive moment estimation optimization algorithm) is used as the optimization method. Through the iterative update of the ADMA algorithm, the model can gradually adjust the weights of each working condition of the shield machine to minimize the overall cross entropy loss.

[0139] S5. Perform multi-objective weighted regression analysis on the prediction factor set to construct a delay prediction model. The specific steps are as follows:

[0140] S51, setting each working condition of the slurry shield machine to C, the mathematical expression of C is shown in Formula 4:

[0141] C={C1,C2,C3,C4}

[0142] Formula 4

[0143] In formula 4, C1, C2, C3, and C4 represent normal propulsion, front discharge of the cutter head, discharge of the cutting chamber, and discharge of the mucking system, respectively.

[0144] For multiple predictor sets X k Perform data standardization, randomly divide the standardized data into training data and test data according to a certain ratio, and divide the training data and test data into data sets X according to the working conditions at time k. 1jk , X 2jk , X 3jk , X 4jk , are the kth values ​​of the jth prediction factor under normal advancement, front cutterhead lag, cutting chamber lag, and mucking system lag conditions, respectively.

[0145] The X k Represents a set of multiple prediction factors obtained at different times k,

[0146] The ratio of the training data to the test data is (2-4):1;

[0147] S52, set the initial weight value of each working condition prediction factor, and establish a multi-objective different-weight regression model, whose mathematical expression is shown in Formula 5:

[0148]

[0149] In formula 5, w ij Represents the weight of the jth prediction factor under the i-th operating condition category;

[0150] For each working condition C i , define independent regression target values Said The mathematical expression of is shown in Equation 6:

[0151]

[0152] In Formula 6, X is the vector set of prediction factors, W i is the weight vector under the i-th working condition category,

[0153] The delay prediction model selects the fault category with the largest output target value as the delay diagnosis result, and its mathematical expression is shown in Formula 7:

[0154]

[0155] In formula 7, This is the predicted delay category.

[0156] S6. Use the ADMA algorithm to optimize the delay prediction model. The specific steps are as follows:

[0157] S61, define the loss function as cross entropy loss to measure the difference between the delay diagnosis result and the true label. The mathematical expression of the loss function is shown in Formula 8:

[0158]

[0159] In formula 8, k is the total number of sample data, i is the total number of working condition categories; if the i-th sample is of category c, y ic If yes, it is 1, otherwise it is 0; p ic is the probability that the i-th sample is diagnosed as operating condition category c;

[0160] S62, initializing ADMA algorithm parameters, wherein the ADMA algorithm parameters include: model weight coefficient θ, first-order moment estimate m, second-order moment estimate v, learning rate α, hyperparameters β1, β2, constant ∈, and number of iterations k;

[0161] S63. In each iteration, the gradient of the loss function is calculated by back propagation Update the first-order moment estimate m t and the second-order moment estimate v t , and calculate the bias-corrected first-order moment and the second moment t is the current iteration number;

[0162] S64, using the corrected moment estimate to update the weight coefficient of the delay prediction model, the weight coefficient matrix is ​​shown in Formula 9:

[0163]

[0164] In formula 9, θ t+1 is the updated weight coefficient of the delay prediction model,

[0165] S65, based on the optimized weight coefficient, calculate the loss function through the training data to obtain the fitness value of the current model;

[0166] S66, judging whether the number of iterations has reached 100 times, if so, stopping the iteration and outputting the weight coefficient matrix of the delay prediction model with optimized parameters; if not, going to step S63 to continue iterative optimization;

[0167] S67. Substitute the divided test data into the optimized delay prediction model for testing, and obtain the prediction accuracy of the delay prediction model. If the minimum prediction accuracy of each operating condition category can reach 80%, the optimization is completed; if not, go to step S62 to adjust the learning rate α, hyperparameters β1, β2, and the number of iterations k, and continue iterative optimization.

[0168] The above iteration number is selected as 100 times, and it is confirmed that the updated value of the weight parameter is a local optimal solution, but it is not limited to this. The iteration number can also be selected as 101, 102, 103, etc. As the iteration number increases, the optimization of the delay prediction model is theoretically more perfect.

[0169] The above ADMA is implemented based on the MATLAB computing platform.

[0170] The fitness value is an indicator obtained by calculating the loss function, which is used to measure the degree of fit of the current model to the training data. The loss function value is converted into an indicator reflecting the performance of the model. The fitness value can be the inverse of the loss function value (for example, the smaller the cross entropy loss, the higher the fitness value). Therefore, the larger the fitness value, the better the performance of the model on the training data, and the more accurately the target variable can be predicted. The prediction accuracy can be indirectly reflected by the fitness value, and the specific prediction accuracy can only be calculated in step S67.

[0171] In this embodiment, the hyperparameter β1 is set to 0.9, the hyperparameter β2 is set to 0.999, the constant ∈ is set to 1e-8, and the initial learning rate α is set to 0.01.

[0172] S7. Use Flomaster to simulate and verify the optimized drainage prediction model and adjust the parameters of the drainage prediction model. The specific steps are as follows:

[0173] S71. Establish a one-dimensional model of the slurry shield machine through Flomaster, and import the above-mentioned drainage prediction model through the NewScript Based N-Arm Component script;

[0174] S72, setting different input boundary conditions and fluid properties, simulating the performance of the shield machine under different geological conditions and different working conditions, and observing the prediction accuracy of the delay prediction model for delay prediction through data post-processing;

[0175] S73. If the prediction accuracy is lower than 80%, the excavation data generated by the Flomaster simulation is returned to step S6 to further optimize the drainage delay prediction model. If the prediction accuracy is greater than or equal to 80%, the drainage delay prediction model is completed and it is considered that the model has high universality.

[0176] The above 80% is used as a threshold for testing the prediction accuracy of the delay prediction model, which can be adjusted according to specific requirements, including but not limited to 81%, 82%, 83%, etc.

[0177] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0178] The above is a detailed introduction to the slurry shield machine delay prediction model based on multi-objective heterogeneous weight regression and its construction method provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a slurry shield machine drainage prediction model based on multi-objective heterogeneous weight regression, characterized in that: The construction method comprises the following steps: S1. Collect the tunneling data of the slurry shield machine, i.e., the operating parameters; S2, grouping the excavation data and performing Kalman filtering to obtain an accurate operating parameter set; S3, using Pearson correlation analysis to analyze the linear correlation degree of different precise operating parameters in the precise operating parameter set, and obtaining the correlation coefficient between each precise operating parameter; S4, screening the precise operating parameters corresponding to the correlation coefficients whose absolute values ​​are greater than the set threshold value to obtain a set of prediction factors; S5. Perform multi-objective weighted regression analysis on the prediction factor set to construct a delay prediction model; S6. Optimize the delay prediction model using ADMA algorithm; S7. Use Flomaster to simulate and verify the optimized drainage prediction model and adjust the parameters of the drainage prediction model.

2. The construction method according to claim 1, characterized in that: In step S1, the excavation data includes the advancing speed, the slurry inflow rate, the slurry discharge rate, the bin pressure, the slurry discharge density and the slurry discharge pump pressure.

3. The construction method according to claim 1, characterized in that: The step S2 specifically includes the following steps: S21. Define the state matrix of the operating parameters of the slurry shield machine at time k. Its mathematical expression is shown in Formula 1: In formula 1, Q out is the main slurry discharge flow, P st is the pressure of the mud tank, P b is the pressure difference between the air cushion chamber pressure and the slurry pump inlet pressure, T b is the propulsion force, ··· are the remaining operating parameters; S22, after repeated prediction and update recursive process, the mathematical expression of the optimal value of Kalman filter is obtained, as shown in Formula 2: In formula 2, For accurate operating parameters, is the prediction of the operating parameters of the slurry shield machine at time k, K k is the Kalman gain, z k is the observation vector.

4. The construction method according to claim 1, characterized in that: In step S3, the mathematical expression of the Pearson correlation analysis is shown in Formula 3: In formula 3, and They represent the values ​​of two different precise operating parameters in the precise operating parameter set at time k, and They are and is the average value of , and r is the correlation coefficient between two different precise operating parameters.

5. The construction method according to claim 1, characterized in that: In step S4, the threshold is set to 0.8-1.

6. The construction method according to claim 1, characterized in that: The step S5 specifically includes the following steps: S51, setting each working condition of the slurry shield machine to C, the mathematical expression of C is shown in Formula 4: C={C1,C2,C3,C4} Formula 4 In formula 4, C1, C2, C3, and C4 represent normal propulsion, front discharge of the cutter head, discharge of the cutting chamber, and discharge of the mucking system, respectively. For multiple predictor sets X k Perform data standardization, randomly divide the standardized data into training data and test data according to a certain ratio, and divide the training data and test data into data sets X according to the working conditions at time k. 1jk , X 2jk , X 3jk , X 4jk , are the kth values ​​of the jth prediction factor under normal advancement, front cutterhead lag, cutting chamber lag, and mucking system lag conditions, respectively. The X k Represents a set of multiple prediction factors obtained at different times k, The ratio of the training data to the test data is (2-4):1; S52, set the initial weight value of each working condition prediction factor, and establish a multi-objective different-weight regression model, whose mathematical expression is shown in Formula 5: In formula 5, w ij Represents the weight of the jth prediction factor under the i-th operating condition category; For each working condition C i , define independent regression target values Said The mathematical expression of is shown in Equation 6: In Formula 6, X is the vector set of prediction factors, W i is the weight vector under the i-th working condition category, The delay prediction model selects the fault category with the largest output target value as the delay diagnosis result, and its mathematical expression is shown in Formula 7: In formula 7, This is the predicted delay category.

7. The construction method according to claim 1, characterized in that: The step S6 specifically includes the following steps: S61, define the loss function as cross entropy loss to measure the difference between the delay diagnosis result and the true label. The mathematical expression of the loss function is shown in Formula 8: In formula 8, k is the total number of sample data, i is the total number of working condition categories; if the i-th sample is of category c, y ic If yes, it is 1, otherwise it is 0; p ic is the probability that the i-th sample is diagnosed as operating condition category c; S62, initializing ADMA algorithm parameters, wherein the ADMA algorithm parameters include: model weight coefficient θ, first-order moment estimate m, second-order moment estimate v, learning rate α, hyperparameters β1, β2, constant ∈, and number of iterations k; S63. In each iteration, the gradient of the loss function is calculated by back propagation Update the first-order moment estimate m t and the second-order moment estimate v t , and calculate the bias-corrected first-order moment and the second moment t is the current iteration number; S64, using the corrected moment estimate to update the weight coefficient of the delay prediction model, the weight coefficient matrix is ​​shown in Formula 9: In formula 9, θ t+1 is the updated weight coefficient of the delay prediction model, S65, based on the optimized weight coefficient, calculate the loss function through the training data to obtain the fitness value of the current model; S66, judging whether the number of iterations reaches the set number of iterations, if so, stopping the iteration and outputting the weight coefficient matrix of the delay prediction model optimized for each parameter; if not, going to step S63 to continue iterative optimization; S67. Substitute the divided test data into the optimized delay prediction model for testing, and obtain the prediction accuracy of the delay prediction model. If the minimum prediction accuracy of each operating condition category can reach the set minimum accuracy value, the optimization is completed; if not, go to step S62 to adjust the learning rate α, hyperparameters β1, β2, and the number of iterations k, and continue iterative optimization.

8. The construction method according to claim 7, characterized in that: The hyperparameter β1 is set to 0.9, the hyperparameter β2 is set to 0.999, the constant ∈ is set to 1e-8, and the initial learning rate α is set to 0.

01.

9. The construction method according to claim 1, characterized in that: The step S7 specifically includes the following steps: S71. Establish a one-dimensional model of the slurry shield machine through Flomaster, and import the above-mentioned drainage prediction model through the NewScript Based N-Arm Component script; S72, setting different input boundary conditions and fluid properties, simulating the performance of the shield machine under different geological conditions, and observing the prediction accuracy of the delay prediction model for delay prediction through data post-processing; S73. If the prediction accuracy is lower than the set prediction accuracy, the excavation data generated by the Flomaster simulation is returned to step S6 to further optimize the delay prediction model. If the prediction accuracy is greater than or equal to the set prediction accuracy, the delay prediction model is completed.

10. A drainage prediction model constructed by the construction method according to any one of claims 1 to 9.

Images