CN-122018342-A - Multi-source parameter coupling modeling shield dyke settlement prediction and closed-loop control method

CN122018342ACN 122018342 ACN122018342 ACN 122018342ACN-122018342-A

Abstract

The invention discloses a shield dyke settlement prediction and closed-loop control method of multisource parameter coupling modeling, in particular to the field of tunnel and underground engineering, which designs a closed-loop self-adaptive decision control method of acquisition, screening, prediction, interpretation, optimization and feedback, the system can continuously optimize the prediction model and the construction parameter adjustment strategy according to the actual effect in the construction process through a real-time feedback mechanism, so that the capability of dynamic adjustment and self optimization in the construction process is realized; the system carries out increment learning or on-line fine adjustment through settlement monitoring data and construction state response fed back in real time, so that the settlement prediction model and the multi-objective optimization model are continuously adapted to the change of stratum and construction working conditions, the prediction precision and the optimization effect are improved, and the accuracy and the robustness of long-term risk management and control in the construction process are ensured.

Inventors

SUN YANG
Zheng Xuru
WU WENQING
SHANG YUEFENG
SHAN XIAOBO
MENG FANTENG
LI MIAO
SUN HONGXUAN

Assignees

河海大学
中铁十四局集团大盾构工程有限公司
河海大学淮安研究院

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (5)

1. The shield dyke settlement prediction and closed-loop control method based on multi-source parameter coupling modeling is characterized by comprising the following steps of: S1, collecting multisource construction parameter data in the shield construction process, and synchronously recording corresponding earth surface subsidence monitoring data to form an initial parameter set; S2, based on a multisource parameter data set after data preprocessing, carrying out correlation analysis on multisource construction parameters and earth surface subsidence monitoring data, evaluating the correlation degree between each construction parameter and earth surface subsidence change, screening out key construction parameter combinations highly correlated with the earth surface subsidence change according to correlation analysis results, eliminating weak correlation parameters, and taking the key construction parameter combinations as input variables of a follow-up subsidence prediction model; S3, constructing a XGBoost surface subsidence prediction model for representing a nonlinear mapping relation between key construction parameters and surface subsidence, introducing a Bayesian optimization algorithm to perform self-adaptive optimization on core parameters of the XGBoost surface subsidence prediction model, and determining an optimal parameter combination by taking a prediction error minimization as a target; S4, performing interpretability analysis on the XGBoost ground surface subsidence prediction model subjected to Bayesian optimization by adopting a SHAP method, and quantifying the contribution degree and positive and negative influence directions of each key construction parameter on the ground surface subsidence prediction result by calculating the SHAP value of each key construction parameter; S5, constructing a multi-objective optimization model based on the contribution degree and positive and negative influence directions of the key construction parameters obtained in the S4 and combining construction safety constraint conditions, equipment operation constraint and construction efficiency indexes, and performing collaborative optimization on the key construction parameters to generate a pareto optimal solution set meeting the constraint conditions; Aiming at the pareto optimal solution set, carrying out multi-index decision screening treatment on the pareto optimal solution set, and selecting a construction parameter combination with optimal comprehensive goodness from the ordered pareto solution set as an optimal solution; the finally determined optimal construction parameter combination is used as an adjustment reference for specific construction, so that dynamic optimization regulation and control of the shield tunneling process is realized; S6, continuously collecting multisource construction parameter data and corresponding earth surface subsidence monitoring data in subsequent construction, comparing and analyzing actual monitoring results with model prediction results, calculating prediction errors and construction parameter response deviations, inputting newly-added data into an earth surface subsidence prediction model and a multi-target optimization model in real time, optimizing prediction model parameters in an incremental learning or online fine tuning mode, adjusting weight distribution, constraint boundaries and optimization target priorities of key construction parameters according to model updating results, and realizing collaborative iterative optimization of the prediction model and the multi-target optimization model.
2. The method for predicting settlement and controlling closed loop of shield dike body by multi-source parameter coupling modeling according to claim 1, wherein the correlation analysis adopts a gray correlation analysis method, and the specific operation of the analysis method is as follows: Firstly, carrying out min-max standardization processing on multisource construction parameter data and corresponding earth surface subsidence monitoring data, calculating correlation coefficients between each construction parameter and earth surface vertical displacement, and calculating corresponding gray correlation degree, sequencing the construction parameters according to the gray correlation degree, and selecting the construction parameters with the gray correlation degree higher than 70% as input variables of a subsidence prediction model, wherein the calculation formula is as follows: Initial parameter set: ; A reference characteristic is indicated and a reference characteristic is indicated, Representing the ith reference feature, which is each factor affecting sedimentation; The characteristics of the object are represented and, Representing the ith target feature, which is a feature related to the sedimentation index, Representing the number of samples; the min-max normalization is performed on the reference characteristic and the target characteristic: ; In the formula, And Respectively representing the normalized reference characteristic and target characteristic; representing an ith reference feature; Representing a jth target feature; Correlation coefficient: ; In the formula, Representing the association coefficient; Is a resolution coefficient; Gray correlation: ; In the formula, Is that And The closer the gray correlation value is to 1, the stronger the correlation is, and n represents the characteristic number; And sorting the key construction parameter sets according to a construction process time sequence to construct a data sample set taking the key construction parameters as characteristic variables and the earth surface subsidence monitoring data at corresponding moments as target variables.
3. The shield dyke subsidence prediction and closed-loop control method based on multi-source parameter coupling modeling according to claim 2 is characterized in that a training set and a test set are divided for a data sample set, the first 70% of the data sample set which is arranged according to a construction time sequence is used as the training set, the second 30% is used as the test set, the key construction parameter set is used as an input characteristic variable to be input into a XGBoost ground subsidence prediction model for training and predicting a model so as to establish a mapping relation between the key construction parameters and ground subsidence, the XGBoost ground subsidence prediction model is constructed by adopting a XGBoost model based on a gradient lifting frame, the model is constructed by taking classification and regression trees as a base learner, prediction residues of a preamble model are continuously fitted in an iterative training process, and an improved expected lifting acquisition function of the base learner is as follows: ; In the formula, Representing super parameters, threshold value For the observed value Is composed of a quantile of super parameter Determining; Indicating the construction parameters under the condition that the observed value is smaller than the threshold value Probability density of occurrence; Indicating parameters under the condition that the observed value is greater than or equal to a threshold value The probability density of the occurrence of the signal, A probability density function representing the occurrence of the observation y within the interval.
4. The method for shield dike settlement prediction and closed-loop control of multi-source parameter coupling modeling of claim 1, wherein the method is characterized in that a parameter self-adaptive optimization strategy based on historical sample prediction error feedback is adopted to adjust structural parameters and training parameters of a XGBoost ground surface settlement prediction model, and in the training process of the XGBoost ground surface settlement prediction model, the error is used for adjusting the parameters of the XGBoost ground surface settlement prediction model, and the specific adjustment is realized by the following steps: Prediction error feedback: ; In the formula, Representing the prediction error at the ith sample; Representing a predicted value of the model, and providing calculation data for a desired lifting acquisition function; representing an actual monitored value; The learning rate adjustment formula: ; In the formula, A control factor indicating the learning rate adjustment, controlling the adjustment amplitude; Representing a prediction error; regularization coefficient adjustment formula: ; In the formula, Representing the regularized coefficient after adjustment; representing a current regularization coefficient; Representing a hyper-parameter controlling the regularization coefficient increase; representing the prediction error.
5. The method for predicting settlement of shield dike body and controlling closed loop by coupling modeling of multi-source parameters according to claim 1, wherein the method is characterized in that data collected in the construction process is monitored in real time through incremental learning and adjusted according to new data, so as to ensure continuous optimization of settlement prediction along with construction, and comprises the following steps: Incremental learning formula: ; Assume that the model has a loss function of Wherein Representing parameters of the model, based on the errors of the historical samples, the parameters of the model being updated by the steps described above, wherein Representing current parameters of the model; representing the parameters after the update of the parameters, The learning rate is represented, and the step length of parameter updating is controlled; Representing the gradient of the loss function to the model parameters; Error-based adaptive optimization: ; In the formula, An exponentially weighted average representing the gradient, reflecting the change in gradient; A constant is represented, preventing divide by zero errors.

Description

Multi-source parameter coupling modeling shield dyke settlement prediction and closed-loop control method Technical Field The invention relates to the field of tunnel and underground engineering, in particular to a shield tunnel settlement prediction and closed-loop control method for multi-source parameter coupling modeling. Background The shield method is one of the most commonly used tunnel construction methods in the current urban rail transit and river-crossing tunnel engineering, and in the construction process of the shield tunnel penetrating through the dike body, the dike body is relatively poor in stability, complex in stratum structure and remarkable in water and soil pressure change, so that the earth surface subsidence overrun is extremely easy to be caused, and the dike body is possibly deformed, leaked and even safe accidents are possibly caused in serious cases. Through retrieval, the Chinese patent application with the publication number of CN113239439A provides a system and a method for predicting earth surface subsidence in shield construction, and the disclosed method realizes earth surface subsidence prediction in the construction process through a meta attribute extraction module, a subsidence data generator training module, a subsidence data generation module, a real-time subsidence prediction module and the like, so that the applicability and the accuracy of prediction are improved, and a large amount of data accumulation is not needed in the early stage. The Chinese patent application with publication number of CN107092990A provides a shield construction ground subsidence prediction system and method based on big data analysis, which utilizes a big data platform to collect, preprocess, extract features and build a prediction model for shield construction data, and performs ground subsidence prediction in the construction process, thereby improving the subsidence prediction efficiency and real-time performance. In addition, the Chinese patent application with the publication number of CN118013634A provides a stratum settlement prediction method and system caused by shield tunneling construction, discloses a space curve shield tunnel calculation model and an integral subsidence model of a shield cross section which are constructed by utilizing the tunneling posture of a shield machine, derives a settlement analysis solution, and theoretically realizes the prediction of stratum settlement. However, in practical engineering applications, the above disclosed sedimentation prediction method and prior art generally have the following disadvantages: On one hand, the existing method can not form a closed loop by prediction, attribution and regulation, so that the whole system is in an open loop state, cannot be dynamically adjusted and self-optimized according to actual construction feedback, and lacks the capability of continuous learning and adaptation to engineering change; On the other hand, the existing prediction model is mostly based on a black box algorithm, and the concrete contribution degree of each construction parameter to sedimentation cannot be clearly explained, so that constructors are difficult to obtain a regulation and control basis from a prediction result, often rely on experience to carry out single-target adjustment, and cannot realize quantitative collaborative optimization under multi-target constraint. Disclosure of Invention Aiming at the problems that a shield construction settlement prediction result is difficult to explain and the prediction result and construction control are mutually disjointed in the prior art, the invention provides a shield tunnel settlement prediction and closed-loop control method for multi-source parameter coupling modeling, and provides a closed-loop self-adaptive decision method applied to 'acquisition-screening-prediction-explanation-optimization-feedback' of shield construction. The deformation at least comprises vertical displacement deformation corresponding to surface subsidence. In order to achieve the purpose, the invention provides the technical scheme that the shield dike settlement prediction and closed-loop control method for multi-source parameter coupling modeling comprises the following steps: S1, collecting multi-source construction parameter data in the shield construction process, wherein the construction parameter data comprises tunnel geometric parameter data, geological parameter data and shield tunneling parameter data, and synchronously recording corresponding earth surface subsidence monitoring data to form an initial parameter set; S2, based on a multisource parameter data set after data preprocessing, carrying out correlation analysis on multisource construction parameters and earth surface subsidence monitoring data, evaluating the correlation degree between each construction parameter and earth surface subsidence change, screening out key construction parameter combinations highly correlated with the earth s