A shield under-penetration stratum stability real-time early warning method and system based on multi-sensor fusion
Patent Information
- Application Number
- CN202611149869.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-31
- Publication Date
- 2026-10-30
AI Technical Summary
[0002]盾构下穿铁路、公路、建筑物等敏感区域时，地层易出现沉降、变形、失稳等安全隐患，传统地层稳定性预警多采用单一传感器+固定阈值判断方式，存在数据源单一、无法适配复杂地质工况、误报漏报率高的问题；现有技术中部分预警方法未采用神经网络智能算法，无法挖掘地层监测数据的时序关联与非线性特征，难以实现地层稳定性状态的精准实时预警，无法满足盾构下穿施工的安全管控需求
[0015]经由上述的技术方案可知，与现有技术相比，本发明公开提供了一种基于多传感器融合的盾构下穿地层稳定性实时预警方法及系统，通过多源传感器数据融合、带地层物理约束的PCA降维与递归卡尔曼滤波动态融合，构建出贴合地层真实物理力学特性的地层孪生状态信息集，有效解决传统预警数据源单一、特征失真的问题；结合时变结构抗力模型与随机施工荷载效应模型构建地层稳定性结构功能函数，依托模糊逻辑复合神经网络深度挖掘监测数据的非线性关联，并通过地层结构变形与地表微裂缝的耦合及时序交叉相关分析动态修正风险值，大幅提升盾构下穿地层稳定性预警的实时性与精准度，显著降低误报、漏报率，可适配复杂地质工况，为盾构下穿施工安全管控提供高效可靠的技术保障。
Smart Images
Figure CN123016642A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of early warning technology for tunnel engineering, and more specifically to a real-time early warning method and system for the stability of shield tunneling under strata based on multi-sensor fusion. Background Technology
[0002] When tunnel boring machines (TBMs) pass under sensitive areas such as railways, highways, and buildings, the strata are prone to safety hazards such as settlement, deformation, and instability. Traditional strata stability early warning methods mostly use a single sensor and fixed threshold judgment method, which has the problems of single data source, inability to adapt to complex geological conditions, and high false alarm and missed alarm rates. Some early warning methods in the existing technology do not use neural network intelligent algorithms, which cannot explore the temporal correlation and nonlinear characteristics of strata monitoring data, making it difficult to achieve accurate real-time early warning of strata stability status and failing to meet the safety management requirements of TBM tunneling construction.
[0003] Therefore, how to achieve accurate and real-time early warning of the ground stability status during shield tunneling is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0004] In view of this, the present invention provides a method and system for real-time early warning of the stability of the ground when the shield tunnel passes under the strata based on multi-sensor fusion, which realizes accurate real-time early warning of the stability status of the strata when the shield tunnel passes under the strata.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: A real-time early warning method for the stability of shield tunneling under strata based on multi-sensor fusion, comprising: Mechanical data, surface micro-crack image data, and geological exploration data of the shield tunneling strata are collected by sensors to form a multi-source raw monitoring dataset; the multi-source raw monitoring dataset is preprocessed to remove invalid data and obtain a regular dataset. The regularized dataset is aligned by linear interpolation in the time dimension and matched by projection interpolation in the spatial dimension. Then, principal component analysis is performed, and a recursive Kalman filter algorithm is used to dynamically fuse multi-source information to construct a stratigraphic twin state information set. Temporal, frequency, spatial, and surface fracture features are extracted from the stratigraphic twin state information set to form a multidimensional coupled feature vector; a time-varying structural resistance model is constructed based on the time-varying law, and a construction load effect model is constructed by combining the random distribution law of construction load and surrounding rock pressure, and the stratigraphic stability structural function is obtained by coupling. By inputting multidimensional coupled feature vectors and structural function parameters into a pre-trained fuzzy logic composite neural network early warning model, the initial risk value of formation stability can be obtained. The deformation rate of the stratum structure and the change rate of surface micro-fracture density are normalized and coupled to obtain the coupling coefficient. Combined with time-series cross-correlation analysis, the dominant factors of damage transmission are determined, the initial risk value of the stratum stability is corrected, and three risk levels (low, medium, and high) are divided. Corresponding early warning strategies are triggered according to the risk level.
[0006] Preferably, the principal component analysis is PCA dimensionality reduction with formation physical constraints, and the formula is: ; in, For the standardized data of the i-th sample and the j-th indicator, For the original data of the j-th indicator of the i-th sample, Let j be the mean of the j-th indicator. Let j be the standard deviation of the j-th indicator. Let covariance matrix be the variance matrix. The total number of samples, These are the eigenvectors of the first k principal components. The dimensionality reduction is the fused feature matrix; the dimensionality reduction process applies constraints such as formation geometric continuity, stress-strain mechanical relationship, and material strength physical constraints, and outputs the dimensionality reduction feature matrix with constraint mechanism.
[0007] Preferably, the dynamic fusion of multi-source information using the recursive Kalman filter algorithm specifically includes: ; in, This represents the optimal estimate of the formation state at time t. This is a one-step prediction of the state at time t+1; This is the formation state transition matrix; Let be the state estimation error covariance matrix at time t; The prediction error covariance matrix at time t+1; The system process noise covariance matrix; Here is the filter gain matrix at time t+1; The sensor observation value at time t+1; The observation matrix is used to ultimately output the optimal estimate of the formation twin state. Update the optimal estimate of the state at time t+1.
[0008] Preferably, the time-domain features include the mean, standard deviation, skewness, kurtosis, and temporal rate of change of the monitoring data; the frequency-domain features include the power spectral density, dominant frequency, and frequency band energy obtained by FFT transformation; the spatial features include three-dimensional point cloud geometric deformation clusters, stratigraphic spatial curvature, and local subsidence gradient; and the surface crack features include crack surface density, average crack length, average crack aperture, and crack propagation direction.
[0009] Preferably, the construction of the time-varying structural resistance model based on time-varying laws specifically includes: ; ; ; Where R(t) is the time-varying structural resistance of the strata at time t; The time-varying attenuation coefficient of formation material strength. The coefficient for concrete deterioration rate is given by t, where t is the duration of construction disturbance. This represents the time-varying coefficient of the stiffness of the geological structure. This is the formation stiffness attenuation coefficient; This is the soil-structure collaborative working coefficient.
[0010] Preferably, the construction load effect model specifically includes: ; in, The dynamic load effect generated by the combined thrust and torque of the tunnel boring machine at time t; The soil displacement load effect generated by the synchronous grouting pressure at time t; The ground load effect caused by the soil chamber pressure at time t; The additional ground load effect caused by construction disturbance at time t.
[0011] Preferably, the formation stability structure function specifically includes: ; ; in, For formation stability structure function; This represents the average structural resistance. This represents the mean value of the load effect; The standard deviation of structural resistance; The standard deviation of the load effect; Let t be the formation stability reliability index; and These are the statistical mean and standard deviation of the function Z(t), respectively.
[0012] Preferably, the fuzzy logic composite neural network early warning model specifically includes: an input layer, a fuzzification layer, a fuzzy inference layer, a defuzzification layer, and an output layer, with each layer connected sequentially. The input layer receives multi-dimensional coupled feature vectors and structural function parameters; the fuzzification layer fuzzifies the multi-dimensional coupled feature vectors and structural function parameters into membership vectors; the fuzzy inference layer performs multi-rule inference through a fuzzy rule base to obtain a fuzzy output; the defuzzification layer converts the fuzzy result into a precise value, i.e., the initial risk value of formation stability; and the output layer outputs the initial risk value of formation stability.
[0013] Preferably, correcting the initial risk value of formation stability specifically includes: First, the deformation rate of the formation structure and the change rate of surface microfracture density are normalized to obtain the normalized deformation rate of the formation structure. The normalized rate of change of surface microfracture density Based on the normalized parameters, the coupling coefficient is calculated using the following formula: ; in, Let be the coupling coefficient at time t. These are the weighting coefficients for deformation rate and change rate, respectively. The weighting coefficients are used for collaborative coupling; the direction of damage propagation is determined by calculating the time-delay cross-correlation coefficient. ; in, For time delay The cross-correlation coefficient, For time delay step size, This represents the total length of the time-series data. , These are the time series mean values of the normalized deformation rate and the rate of change, respectively. Based on coupling coefficient Based on the damage transmission direction determination and the dominant factors of damage transmission, the initial risk value is dynamically corrected, as shown in the following formula: ; in, This is a risk amplification factor. As the initial risk value, The correction coefficient is the dominant factor.
[0014] A real-time early warning system for the stability of shield tunneling under strata based on multi-sensor fusion includes: The data processing unit collects mechanical data, surface micro-crack image data, and geological exploration data of the shield tunneling strata through sensors to form a multi-source raw monitoring dataset; the multi-source raw monitoring dataset is preprocessed to remove invalid data and obtain a regular dataset; The data integration unit performs time-dimensional linear interpolation alignment and spatial-dimensional projection interpolation matching on the regularized dataset, then performs principal component analysis, and uses a recursive Kalman filter algorithm to dynamically fuse multi-source information to construct a stratigraphic twin state information set. The function coupling unit extracts time-domain features, frequency-domain features, spatial features, and surface crack features from the stratigraphic twin state information set to form a multi-dimensional coupled feature vector; it constructs a time-varying structural resistance model based on time-varying laws, and constructs a construction load effect model by combining the random distribution laws of construction load and surrounding rock pressure, and couples to obtain the stratigraphic stability structural function. The initial risk determination unit inputs the multidimensional coupled feature vector and structure function parameters into a pre-trained fuzzy logic composite neural network early warning model to obtain the initial risk value of formation stability. The early warning unit is modified by normalizing and coupling the deformation rate of the stratum structure with the change rate of surface microcrack density to obtain the coupling coefficient. Combined with time-series cross-correlation analysis, the dominant factors of damage transmission are determined, the initial risk value of the stratum stability is modified, and three risk levels (low, medium, and high) are divided. The corresponding early warning strategy is triggered according to the risk level.
[0015] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a real-time early warning method and system for shield tunneling stability under strata based on multi-sensor fusion. By fusing multi-source sensor data, PCA dimensionality reduction with stratum physical constraints and dynamic fusion with recursive Kalman filtering, a stratum twin state information set that fits the real physical and mechanical properties of the stratum is constructed, effectively solving the problems of single data source and feature distortion in traditional early warning systems. The system combines time-varying structural resistance model and stochastic construction load effect model to construct stratum stability structural function, relies on fuzzy logic composite neural network to deeply mine the nonlinear correlation of monitoring data, and dynamically corrects the risk value through coupling of stratum structural deformation and surface microcracks and time series cross-correlation analysis. This significantly improves the real-time performance and accuracy of shield tunneling stability early warning, significantly reduces false alarm and missed alarm rates, and can be adapted to complex geological conditions, providing efficient and reliable technical support for the safety management of shield tunneling construction. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0017] Figure 1 A flowchart of the method steps provided by the present invention; Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] This invention discloses a real-time early warning method for the stability of shield tunneling under strata based on multi-sensor fusion, such as... Figure 1 As shown, it includes: Mechanical data, surface micro-crack image data, and geological exploration data of the shield tunneling strata are collected by sensors to form a multi-source raw monitoring dataset; the multi-source raw monitoring dataset is preprocessed to remove invalid data and obtain a regular dataset. The regular dataset is aligned by linear interpolation in the time dimension and matched by projection interpolation in the spatial dimension. Then, principal component analysis is performed, and the recursive Kalman filter algorithm is used to dynamically fuse multi-source information to construct a stratigraphic twin state information set. Temporal, frequency, spatial, and surface fracture features are extracted from the stratigraphic twin state information set to form a multidimensional coupled feature vector; a time-varying structural resistance model is constructed based on the time-varying law, and a construction load effect model is constructed by combining the random distribution law of construction load and surrounding rock pressure, and the stratigraphic stability structural function is obtained by coupling. By inputting multidimensional coupled feature vectors and structural function parameters into a pre-trained fuzzy logic composite neural network early warning model, the initial risk value of formation stability can be obtained. The coupling coefficient is obtained by normalizing the deformation rate of the stratum structure and the change rate of surface micro-fracture density. Combined with time-series cross-correlation analysis, the dominant factors of damage transmission are determined, the initial risk value of stratum stability is corrected, and three levels of risk (low, medium, and high) are divided. Corresponding early warning strategies are triggered according to the risk level.
[0020] In one specific embodiment, principal component analysis is performed as PCA dimensionality reduction with formation physical constraints, and the formula is as follows: ; in, For the standardized data of the i-th sample and the j-th indicator, For the original data of the j-th indicator of the i-th sample, Let j be the mean of the j-th indicator. Let j be the standard deviation of the j-th indicator. Let covariance matrix be the variance matrix. The total number of samples, These are the eigenvectors of the first k principal components. The dimensionality reduction is the fused feature matrix; the dimensionality reduction process applies constraints such as formation geometric continuity, stress-strain mechanical relationship, and material strength physical constraints, and outputs the dimensionality reduction feature matrix with constraint mechanism.
[0021] Conventional unconstrained principal component analysis (PCA) is a purely mathematical dimensionality reduction method that selects principal components solely by maximizing data variance. It completely ignores the geophysical laws, structural mechanical properties, and spatial distribution characteristics of the strata that the tunnel boring machine (TBM) passes through, making it prone to problems such as physical distortion, spurious correlations of features, and breaks in mechanical logic. Therefore, it is necessary to forcibly apply three types of physical constraints during the dimensionality reduction process: stratum geometric continuity, stress-strain mechanical relationship, and material strength. The goal is to ensure that the dimensionality-reduced data achieves both dimensionality reduction and redundancy elimination, while strictly adhering to the actual physical state of the strata. This guarantees that the subsequent twin information sets, stability modeling, and early warning results have engineering rationality and reliability.
[0022] The strata (soil, surrounding rock, and surface structure) that the tunnel boring machine (TBM) passes through are spatially continuous media. Surface settlement, stratum displacement, and 3D point cloud deformation all follow a gradual and continuous spatial distribution pattern, without any unexplained abrupt changes or jumps. Unconstrained PCA, based solely on mathematical variance, may misclassify measurement noise, local anomalies, and sensor drift as high-variance principal components, disrupting the spatial continuity of the strata and resulting in a severe discrepancy between the constructed stratum twin state and the actual geological spatial morphology. Applying geometric continuity constraints forces the dimensionality reduction process to preserve the topological relationships of the strata, the continuity of the settlement gradient, and the geometric consistency of the point cloud, avoiding spatial feature distortion and ensuring the accurate and effective extraction of subsequent spatial and fracture features.
[0023] The stress, strain, displacement, and surrounding rock pressure between the strata and the supporting structure strictly follow the constitutive relations of geotechnical mechanics and the laws of elastic / elastoplastic mechanics. They are strongly coupled physical quantities, rather than independent random variables. Pure mathematical PCA would blindly decompose the inherent mechanical relationship between stress and strain, splitting strongly correlated mechanical quantities into independent principal components, causing the data after dimensionality reduction to violate the basic logic of mechanics. Applying this constraint can ensure that the mechanical coupling characteristics are preserved in the dimensionality reduction process, so that the fused data still conforms to the laws of stress and deformation. This provides mechanically logical input for subsequent time-varying structural resistance modeling, load effect calculation, and structural function construction, avoiding physical paradoxes in stability calculations.
[0024] The strength, stiffness, and resistance of the soil, concrete support, and reinforced steel structure all have natural physical upper limits and time-varying decay laws, and cannot exceed the strength limit of the material itself.
[0025] Unconstrained PCA generates spurious principal components that exceed the material strength threshold and do not conform to the degradation law, directly leading to the distortion of the time-varying structural resistance model calculation and unreasonable values in reliability indicators and risk values. Applying material strength constraints can limit the dimensionality reduction features to the range of the material's true mechanical properties, prevent non-physical features from entering the early warning model, and avoid misjudgment and missed judgment in early warning from the source.
[0026] Stratigraphic geometric continuity constraints include: Settlement gradient constraints: ,in , The settlement (m) is the amount of sediment at adjacent monitoring points. The distance between monitoring points is (m). The maximum allowable settlement gradient; Point cloud deformation smoothness constraints: ,in The angle between the normal vectors of adjacent point clouds.
[0027] Stress-strain mechanical constraints include: Elastic phase: and Where E is the elastic modulus, Poisson's ratio (0.35-0.45 for silty clay, 0.2 for concrete) It is the elastic strain limit; Elastic-plastic stage: ,in ω is the internal friction angle of the formation, and c is the cohesion.
[0028] Physical constraints on material strength include: Strength upper limit constraint: ,in The ultimate strength of the material; Time-varying decay constraint: , where k is the attenuation coefficient.
[0029] Constraint Implementation Mechanism
[0030] Geometric constraints: For data involving hypersinking gradients and the angle between hypernormal vectors, a penalty term is added to the covariance matrix. (Abnormal data) =1, normal data =0); Mechanical constraints: Calculating mechanical deviation values Introducing weighting factors Revise the principal component screening criteria ; Strength constraint: Threshold truncation correction is applied to eigenvalues that exceed the strength threshold. .
[0031] Finally, based on the corrected covariance matrix and principal component selection criteria, a dimensionality-reduced feature matrix that satisfies the physical constraints is output. .
[0032] In one specific embodiment, the dynamic fusion of multi-source information using the recursive Kalman filter algorithm specifically includes: ; in, This represents the optimal estimate of the formation state at time t. This is a one-step prediction of the state at time t+1; This is the formation state transition matrix; Let be the state estimation error covariance matrix at time t; The prediction error covariance matrix at time t+1; The system process noise covariance matrix; Here is the filter gain matrix at time t+1; The sensor observation value at time t+1; The observation matrix is used to ultimately output the optimal estimate of the formation twin state. Update the optimal estimate of the state at time t+1.
[0033] In one specific embodiment, the time-domain features include the mean, standard deviation, skewness, kurtosis, and temporal rate of change of the monitoring data; the frequency-domain features include the power spectral density, dominant frequency, and frequency band energy obtained by FFT transformation; the spatial features include the geometric deformation clusters of three-dimensional point clouds, the spatial curvature of the strata, and the local subsidence gradient; and the surface crack features include crack surface density, average crack length, average crack aperture, and crack propagation direction.
[0034] In one specific embodiment, constructing a time-varying structural resistance model based on time-varying laws specifically includes: ; ; ; Where R(t) is the time-varying structural resistance of the strata at time t; The time-varying attenuation coefficient of formation material strength. The coefficient for concrete deterioration rate is given by t, where t is the duration of construction disturbance. This represents the time-varying coefficient of the geological structure stiffness. This is the formation stiffness attenuation coefficient; It is the soil-structure collaborative working coefficient, reflecting the collaborative working capacity of the soil and lining structure under the disturbance of shield tunneling construction.
[0035] In one specific embodiment, constructing the construction load effect model specifically includes: ; in, The dynamic load effect generated by the combined thrust and torque of the tunnel boring machine at time t; The soil displacement load effect generated by the synchronous grouting pressure at time t; The ground load effect caused by the soil chamber pressure at time t; The additional ground load effect caused by construction disturbance at time t.
[0036] Synergistic dynamic load effect of tunneling thrust and torque for: ; in, Let t be the total thrust of the shield tunneling in real time. The angle between the thrust and the tangent direction of the cutter head. The thrust-torque-stress conversion factor is determined based on the formation type. Let t be the real-time angular velocity of the cutterhead.
[0037] Synchronous grouting pressure soil displacement load effect for: ; in, The real-time grouting pressure at time t is the synchronous grouting pressure. The total cross-sectional area of the grouting holes is calculated from the number of grouting holes and the area of each hole. This refers to the equivalent cross-sectional area of the tunnel boring machine acting on the stratum. This is the grouting pressure-load effect conversion factor; Soil chamber pressure stratum load effect for: ; in, Let t be the real-time pressure of the soil chamber. Let t be the vertical distance between the monitoring point and the center of the soil chamber. is the soil chamber pressure attenuation coefficient, determined based on the formation permeability; This is the equivalent coefficient of the soil chamber pressure.
[0038] Additional load effects of construction disturbance for: ; in, Let t be the tunneling speed of the shield machine. The density of the strata is taken from the measured geological data. Let t be the depth of the shield tunnel. This is the disturbance load factor.
[0039] In one specific embodiment, the formation stability structure function specifically includes: ; ; in, For formation stability structure function; This represents the average structural resistance. This represents the mean value of the load effect; The standard deviation of structural resistance; The standard deviation of the load effect; Let t be the formation stability reliability index; and These are the statistical mean and standard deviation of the function Z(t), respectively.
[0040] In one specific embodiment, the fuzzy logic composite neural network early warning model specifically includes: an input layer, a fuzzification layer, a fuzzy inference layer, a defuzzification layer, and an output layer, with each layer connected sequentially. The input layer receives multidimensional coupled feature vectors and structural function parameters; the fuzzification layer fuzzifies the multidimensional coupled feature vectors and structural function parameters into membership vectors; the fuzzy inference layer performs multi-rule inference through a fuzzy rule base to obtain fuzzy output; the defuzzification layer converts the fuzzy result into a precise value, i.e., the initial risk value of formation stability; and the output layer outputs the initial risk value of formation stability.
[0041] The input vector of the fuzzy logic composite neural network early warning model consists of two parts: Multidimensional coupled feature vectors: ; which includes: Temporal characteristics: settlement rate, differential settlement, mean square value of vibration acceleration, etc. Frequency domain characteristics: dominant frequency, energy percentage, spectral peak value; Spatial characteristics: stratigraphic curvature, subsidence cluster density, deformation gradient; Characteristics of surface cracks: crack surface density, average length, and rate of change of crack aperture; Structure function parameters: ; Corresponding to time-varying structural resistance, total load effect, structural function value, and reliability index.
[0042] Final input dimensions: ; The blurring layer applies to each input dimension. The data is then fuzzified and divided into three fuzzy sets: low (L), medium (M), and high (H).
[0043] Using Gaussian membership function: ; in, For input Membership degree of fuzzy set A; The central value of the Gaussian function; This is the width coefficient of the Gaussian function.
[0044] The output of the fuzzification layer is a membership vector: ; The fuzzy rule base of the fuzzy inference layer is in the form of:
[0045] The total number of rules is .
[0046] The inference layer uses product inference, with the trigger strength of a single rule as follows: ; in, The trigger strength for the j-th rule; Let be the fuzzy linguistic variable for the i-th input to the j-th rule.
[0047] The sum of all rules is used to obtain the total fuzzy output: ; in, The overall membership distribution of the risk value p; Output the membership function of the fuzzy set for the j-th rule.
[0048] The defuzzing layer uses the center of gravity (CoG) method to obtain accurate initial risk values. : .
[0049] In one specific embodiment, correcting the initial risk value of formation stability specifically includes: First, the deformation rate of the formation structure and the change rate of surface microfracture density are normalized to obtain the normalized deformation rate of the formation structure. The normalized rate of change of surface microfracture density Based on the normalized parameters, the coupling coefficient is calculated using the following formula: ; in, Let be the coupling coefficient at time t. These are the weighting coefficients for deformation rate and change rate, respectively. The weighting coefficients are used for collaborative coupling; the direction of damage propagation is determined by calculating the time-delay cross-correlation coefficient. ; in, For time delay The cross-correlation coefficient, For time delay step size, This represents the total length of the time-series data. , These are the time series mean values of the normalized deformation rate and the rate of change, respectively. Rules for determining dominant factors: If it exists and Surface crack damage is dominant (crack propagation precedes significant stratum deformation, and cracks are the source of damage transmission). If it exists and : Formation structure deformation is dominant (formation deformation precedes fracture propagation, and deformation is the source of damage transmission); like and Deformation and cracks are synergistically dominant (both develop synchronously with no clear sequence of transmission). like (all ): Weak correlation (damage is dominated by other factors, treated as synergistic).
[0050] Based on coupling coefficient Based on the damage transmission direction determination and the dominant factors of damage transmission, the initial risk value is dynamically corrected, as shown in the following formula: ; in, This is a risk amplification factor. As the initial risk value, The correction coefficient is the dominant factor.
[0051] A real-time early warning system for the stability of shield tunneling under strata based on multi-sensor fusion includes: The data processing unit collects mechanical data, surface micro-crack image data, and geological exploration data of the shield tunneling strata through sensors to form a multi-source raw monitoring dataset; it preprocesses the multi-source raw monitoring dataset to remove invalid data and obtain a regular dataset. The data integration unit performs time-dimensional linear interpolation alignment and spatial-dimensional projection interpolation matching on the regularized dataset, then performs principal component analysis, and uses the recursive Kalman filter algorithm to dynamically fuse multi-source information to construct a stratigraphic twin state information set. The function coupling unit extracts time-domain features, frequency-domain features, spatial features, and surface fracture features from the stratigraphic twin state information set to form a multi-dimensional coupled feature vector; a time-varying structural resistance model is constructed based on time-varying laws, and a construction load effect model is constructed by combining the random distribution laws of construction load and surrounding rock pressure, and the stratigraphic stability structural function is obtained through coupling. The initial risk determination unit inputs the multidimensional coupled feature vector and structure function parameters into a pre-trained fuzzy logic composite neural network early warning model to obtain the initial risk value of formation stability. The early warning unit is modified by normalizing and coupling the deformation rate of the stratum structure with the change rate of surface micro-fracture density to obtain the coupling coefficient. Combined with time-series cross-correlation analysis, the dominant factors of damage transmission are determined, the initial risk value of stratum stability is modified, and three levels of risk (low, medium, and high) are divided. The corresponding early warning strategy is triggered according to the risk level.
[0052] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the methods disclosed in the embodiments; relevant parts can be found in the method section.
[0053] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A real-time early warning method for the stability of shield tunneling under strata based on multi-sensor fusion, characterized in that, include: The mechanical data, surface micro-crack image data, and geological exploration data of the shield tunneling under the strata are collected by sensors to form a multi-source raw monitoring dataset. The multi-source raw monitoring dataset is preprocessed to remove invalid data and obtain a regularized dataset; The regularized dataset is aligned by linear interpolation in the time dimension and matched by projection interpolation in the spatial dimension. Then, principal component analysis is performed, and a recursive Kalman filter algorithm is used to dynamically fuse multi-source information to construct a stratigraphic twin state information set. Temporal, frequency, spatial, and surface fracture features are extracted from the stratigraphic twin state information set to form a multidimensional coupled feature vector; a time-varying structural resistance model is constructed based on the time-varying law, and a construction load effect model is constructed by combining the random distribution law of construction load and surrounding rock pressure, and the stratigraphic stability structural function is obtained by coupling. By inputting multidimensional coupled feature vectors and structural function parameters into a pre-trained fuzzy logic composite neural network early warning model, the initial risk value of formation stability can be obtained. The deformation rate of the stratum structure and the change rate of surface micro-fracture density are normalized and coupled to obtain the coupling coefficient. Combined with time-series cross-correlation analysis, the dominant factors of damage transmission are determined, the initial risk value of the stratum stability is corrected, and three risk levels (low, medium, and high) are divided. Corresponding early warning strategies are triggered according to the risk level.
2. The method for real-time early warning of shield tunneling stability under strata based on multi-sensor fusion according to claim 1, characterized in that, The principal component analysis described is a PCA dimensionality reduction method with formation physical constraints, and the formula is as follows: ； in, For the standardized data of the i-th sample and the j-th indicator, For the original data of the j-th indicator of the i-th sample, Let j be the mean of the j-th indicator. Let j be the standard deviation of the j-th indicator. Let covariance matrix be the variance matrix. The total number of samples, These are the eigenvectors of the first k principal components. The dimensionality reduction is the fused feature matrix; the dimensionality reduction process applies constraints such as formation geometric continuity, stress-strain mechanical relationship, and material strength physical constraints, and outputs the dimensionality reduction feature matrix with constraint mechanism.
3. The method for real-time early warning of shield tunneling stability under strata based on multi-sensor fusion according to claim 1, characterized in that, The specific steps of using the recursive Kalman filter algorithm for dynamic fusion of multi-source information include: ； in, This represents the optimal estimate of the formation state at time t. This is a one-step prediction of the state at time t+1; This is the formation state transition matrix; Let be the state estimation error covariance matrix at time t; The prediction error covariance matrix at time t+1; The system process noise covariance matrix; Here is the filter gain matrix at time t+1; The sensor observation value at time t+1; The observation matrix is used to ultimately output the optimal estimate of the formation twin state. Update the optimal estimate of the state at time t+1.
4. The method for real-time early warning of shield tunneling stability under strata based on multi-sensor fusion according to claim 1, characterized in that, The time-domain features include the mean, standard deviation, skewness, kurtosis, and temporal rate of change of the monitoring data; the frequency-domain features include the power spectral density, dominant frequency, and frequency band energy obtained by FFT transformation; the spatial features include three-dimensional point cloud geometric deformation clusters, stratigraphic spatial curvature, and local subsidence gradient; and the surface crack features include crack surface density, average crack length, average crack aperture, and crack propagation direction.
5. The method for real-time early warning of shield tunneling stability under strata based on multi-sensor fusion according to claim 1, characterized in that, The construction of the time-varying structural resistance model based on time-varying laws specifically includes: ； ； ； Where R(t) is the time-varying structural resistance of the strata at time t; The time-varying attenuation coefficient of formation material strength. The coefficient for concrete deterioration rate is given by t, where t is the duration of construction disturbance. This represents the time-varying coefficient of the stiffness of the geological structure. This is the formation stiffness attenuation coefficient; This is the soil-structure collaborative working coefficient.
6. The method for real-time early warning of shield tunneling stability under strata based on multi-sensor fusion according to claim 5, characterized in that, The construction load effect model specifically includes: ； in, The dynamic load effect generated by the combined thrust and torque of the tunnel boring machine at time t; The soil displacement load effect generated by the synchronous grouting pressure at time t; The ground load effect caused by the soil chamber pressure at time t; The additional ground load effect caused by construction disturbance at time t.
7. A real-time early warning method for the stability of shield tunneling under strata based on multi-sensor fusion as described in claim 6, characterized in that, The specific structural function of the formation stability includes: ； ； in, For formation stability structure function; This represents the average structural resistance. This represents the mean value of the load effect; The standard deviation of structural resistance; The standard deviation of the load effect; Let t be the formation stability reliability index; and These are the statistical mean and standard deviation of the function Z(t), respectively.
8. The method for real-time early warning of shield tunneling stability under strata based on multi-sensor fusion according to claim 1, characterized in that, The fuzzy logic composite neural network early warning model specifically includes: an input layer, a fuzzification layer, a fuzzy inference layer, a defuzzification layer, and an output layer, with each layer connected sequentially. The input layer receives multi-dimensional coupled feature vectors and structural function parameters; the fuzzification layer fuzzifies the multi-dimensional coupled feature vectors and structural function parameters into membership vectors; the fuzzy inference layer performs multi-rule inference through a fuzzy rule base to obtain a fuzzy output; the defuzzification layer converts the fuzzy result into a precise value, i.e., the initial risk value of formation stability; and the output layer outputs the initial risk value of formation stability.
9. A real-time early warning method for the stability of shield tunneling under strata based on multi-sensor fusion according to claim 1, characterized in that, The specific steps to correct the initial risk value of formation stability include: First, the deformation rate of the formation structure and the change rate of surface microfracture density are normalized to obtain the normalized deformation rate of the formation structure. The normalized rate of change of surface microfracture density Based on the normalized parameters, the coupling coefficient is calculated using the following formula: ； in, Let be the coupling coefficient at time t. These are the weighting coefficients for deformation rate and change rate, respectively. The weighting coefficients are used for collaborative coupling; the direction of damage propagation is determined by calculating the time-delay cross-correlation coefficient. ； in, For time delay The cross-correlation coefficient, For time delay step size, This represents the total length of the time-series data. , These are the time series mean values of the normalized deformation rate and the rate of change, respectively. Based on coupling coefficient Based on the damage transmission direction determination and the dominant factors of damage transmission, the initial risk value is dynamically corrected, as shown in the following formula: ； in, This is a risk amplification factor. As the initial risk value, The correction coefficient is the dominant factor.
10. A real-time early warning system for the stability of shield tunneling under strata based on multi-sensor fusion, employing the real-time early warning method for the stability of shield tunneling under strata based on multi-sensor fusion as described in any one of claims 1-9, characterized in that, include: The data processing unit collects mechanical data, surface micro-crack image data, and geological exploration data of the shield tunneling through the strata using sensors, forming a multi-source raw monitoring dataset. The multi-source raw monitoring dataset is preprocessed to remove invalid data and obtain a regularized dataset; The data integration unit performs time-dimensional linear interpolation alignment and spatial-dimensional projection interpolation matching on the regularized dataset, then performs principal component analysis, and uses a recursive Kalman filter algorithm to dynamically fuse multi-source information to construct a stratigraphic twin state information set. The function coupling unit extracts time-domain features, frequency-domain features, spatial features, and surface crack features from the stratigraphic twin state information set to form a multi-dimensional coupled feature vector; it constructs a time-varying structural resistance model based on time-varying laws, and constructs a construction load effect model by combining the random distribution laws of construction load and surrounding rock pressure, and couples to obtain the stratigraphic stability structural function. The initial risk determination unit inputs the multidimensional coupled feature vector and structure function parameters into a pre-trained fuzzy logic composite neural network early warning model to obtain the initial risk value of formation stability. The early warning unit is modified by normalizing and coupling the deformation rate of the stratum structure with the change rate of surface microcrack density to obtain the coupling coefficient. Combined with time-series cross-correlation analysis, the dominant factors of damage transmission are determined, the initial risk value of the stratum stability is modified, and three risk levels (low, medium, and high) are divided. The corresponding early warning strategy is triggered according to the risk level.