Tunnel pipe roof construction dynamic deviation correction method based on laser ranging and optical fiber sensing

Abstract

The invention discloses a tunnel pipe roofing construction dynamic deviation correction method based on laser ranging and optical fiber sensing, and relates to the technical field of measurement and sensing, and the method comprises the steps: S1, collecting laser and optical fiber data, obtaining real strain through temperature coupling compensation, and calculating global pose parameters; s2, generating an eigenmode component based on the real strain and the global pose parameter, and screening and constructing an intelligent feature vector according to the contribution degree of the physical coupling feature; s3, constructing a physical information driving state estimation model in combination with pipe joint mechanical constraints, and outputting a pipe joint state estimation value in real time; s4, performing construction deviation prediction by using the pipe joint state estimation value, and generating an optimal control sequence and instruction; and S5, applying the control instruction to the construction device, and adaptively updating the key parameters of the system through a meta-optimization algorithm based on feedback. And through multi-modal data fusion and physical constraint intelligent prediction, high-precision deviation correction, closed-loop robust control and system self-adaptive optimization of the pipe joints are realized, and the intelligence and long-term precision of tunnel pipe curtain construction are improved.

Description

Technical Field

[0001] This invention relates to the field of measurement and sensing technology, specifically to a dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing. Background Technology

[0002] During tunnel jacking construction, the installation posture and positional accuracy of the tunnel segments are crucial factors in ensuring construction quality and structural safety. Current technologies primarily rely on manual measurement, single-sensor monitoring, or automatic control methods based on fixed models. However, existing methods have the following shortcomings: Measurement accuracy is limited: Traditional manual measurement and single-point sensor deployment methods are easily affected by tunnel environmental factors, such as insufficient lighting, construction vibration and equipment installation deviation, making it difficult to obtain high-precision pipe section displacement and attitude data.

[0003] Sensor data is significantly affected by interference: data collected by a single sensor is easily affected by environmental changes. For example, optical ranging may produce measurement errors due to dust or light fluctuations, and fiber optic strain data may be significantly affected by temperature changes and mechanical disturbances, thereby reducing the reliability of pipe section status judgment.

[0004] Lack of multimodal data fusion capability: Existing technologies usually rely on single-modal data for state estimation or control, without fully integrating multi-source sensing information such as fiber strain and laser ranging. This makes it difficult to effectively explore the potential correlation between data, limiting the improvement of construction deviation prediction and control accuracy.

[0005] Poor adaptability to dynamic working conditions: Existing control systems are mostly based on fixed parameters or offline calibration models, which cannot fully cope with dynamic factors such as changes in geological conditions, temperature fluctuations and construction vibrations during construction, resulting in a decrease in the accuracy of pipe section state estimation and control.

[0006] Insufficient closed-loop optimization capability: Existing methods lack real-time feedback and global optimization mechanisms for control execution effects, making it difficult to achieve adaptive updates of various parameters during construction. Deviations can easily accumulate during long-term construction, affecting overall construction quality and safety.

[0007] Limited level of intelligence: Traditional construction control systems rely heavily on preset control strategies and manual adjustments, lacking the ability to learn intelligently and predict conditions based on historical construction data. This makes it difficult to capture the movement patterns of pipe sections and the trends of construction deviations in real time, thus limiting the improvement of automation level and construction accuracy.

[0008] In summary, existing technologies have significant shortcomings in terms of measurement accuracy, multimodal information fusion, dynamic adaptability, closed-loop optimization, and intelligence level, making it difficult to meet the requirements of high-precision and high-reliability tunnel jacking construction. Summary of the Invention

[0009] Based on the shortcomings of the prior art described above, the purpose of this invention is to provide a dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing, so as to solve the above-mentioned technical problems.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing, comprising: S1: Acquire laser and fiber optic data, calculate the temperature coupling coefficient by least squares fitting through automatic identification of the temperature excitation window to compensate for the fiber optic data, obtain real strain data, and calculate global pose parameters through online optimization based on the cubic norm. S2: Generate eigenmode components based on real strain and global pose parameters, and filter them according to the calculated contribution of physical coupling features to construct intelligent feature vectors; S3: Based on the intelligent feature vector and the mechanical and physical constraints of the pipe section, a physical information-driven state estimation model is constructed to output the estimated state value of the pipe section in real time; S4: Using the estimated state values ​​of pipe sections, dynamically predict the construction deviations of pipe sections and generate the optimal control sequence and optimal control commands; S5: Apply the optimal control commands to the construction guidance, jack adjustment and grouting device, and adaptively update the key parameters of the system based on actual construction feedback data through a meta-optimization algorithm.

[0011] The present invention is further configured such that S1 includes: an initial calibration unit and an online optimization and calibration unit; The initial calibration unit includes: Collect laser ranging data arranged at known coordinate control points and construct a ranging dataset; The temperature excitation window is automatically identified, and the raw strain data and temperature measurement data of the fiber optic sensor in the window are collected to construct a strain dataset. A robust optimization method is used to solve for the pose parameters of the laser rangefinder in the global coordinate system in the ranging dataset, and these parameters are set as the initial global pose parameters. Using a pre-calibrated temperature coupling coefficient, temperature compensation is performed on the original strain data to output the true strain data of the structure.

[0012] The present invention is further configured such that the online optimization and calibration unit includes: During system operation, temperature changes are monitored during periods when the structure is in a stress-free state, and the temperature coupling coefficient is dynamically updated using linear regression analysis. Based on the real-time acquired structural strain data, the displacement changes of structural feature points are calculated to construct the structural displacement change. By combining the changes in the real-time range measurement dataset with the changes in structural displacement, an optimization method is used to complete the online correction of the initial global pose parameters, which are then set as the global pose parameters.

[0013] The present invention is further configured such that S2 includes: a multi-source data preprocessing unit, an adaptive signal decomposition unit, and an intelligent feature selection unit; The multi-source data preprocessing unit includes: Based on the real-time acquired ranging dataset, the difference between laser ranging values ​​at adjacent times is calculated to obtain the change in the ranging dataset, while the real strain data output by S1 is received. Digital filtering is used to eliminate noise interference on the range measurement dataset, and anomaly detection is used to remove abnormal measurement values ​​from the strain data. The processed range measurement dataset changes and strain data are combined into a multimodal data vector.

[0014] The present invention is further configured such that the adaptive signal decomposition unit includes: An adaptive variational mode decomposition algorithm is used to process multimodal data vectors. Through an iterative optimization process, the multimodal data vectors are decomposed into multiple time-frequency components at different frequencies and with time-varying characteristics, and multiple sets of intrinsic mode components are extracted. Simultaneously extract the center frequency parameters of each group of intrinsic mode components; Multiple sets of intrinsic mode components and their corresponding center frequency parameters are transmitted to the intelligent feature selection unit.

[0015] The present invention is further configured such that the intelligent feature selection unit includes: Based on the intrinsic mode components of each group and the preset characteristics of the current tunnel segment motion mode, the contribution of physical coupling characteristics is calculated. Based on a preset feature selection threshold, the intrinsic mode components with the highest contribution to the physical coupling features are selected. The selected intrinsic mode components are weighted and fused using their corresponding physical coupling feature contributions as weights to generate a fused intelligent feature vector.

[0016] The present invention is further configured such that S3 includes: Perform system initialization: Check if the pipe section status value at the previous moment is valid. If it is invalid, use the preset status value to initialize the status. If it is in an active state, it receives the intelligent feature vector obtained from S2 and the previous node state; A neural network model is established based on a preset neural network learning rate. The intelligent feature vector and the previous node state are used as network inputs. The preliminary estimate of the current state is obtained through forward propagation of the neural network. Based on the real-time sensor measurement data, coordinate transformation and calculation are performed using the global pose parameters of the S1 dual output to derive the reference state value.

[0017] The present invention is further configured to calculate the data difference between the preliminary estimate and the reference state value; Simultaneously, based on the Euler-Bernoulli beam theory, physical equations are established and constrained to obtain the predicted values ​​of the equations. The physical consistency deviation between the preliminary estimate and the predicted values ​​of the equations is then calculated. A multi-objective optimization function is constructed by combining data differences and physical consistency deviations, and the neural network model is optimized through the backpropagation algorithm. The system monitors changes in geological conditions and process parameters during construction, adjusts neural network parameters online based on real-time monitoring data, and outputs an estimated value of the pipe section status at the current moment.

[0018] The present invention is further configured such that S4 includes: Based on the estimated state of the pipe segment at the current moment, a probabilistic prediction method is used to perform multi-step forward state prediction, and the prediction uncertainty of the future state is quantified by Monte Carlo sampling or Bayesian inference method to obtain the future state prediction sequence. Based on the future state prediction sequence, and combined with the preset risk sensitivity coefficient, a control optimization problem using stochastic model prediction is constructed. The optimal control sequence is obtained by using an optimization algorithm that minimizes conditional risk value. Optimal control commands are generated based on the optimal control sequence. These commands include jack displacement, grouting pressure, and guide frame adjustment.

[0019] The present invention is further configured such that S5 includes: The optimal control command sequence is sent to the actuator to control the coordinated action of the jack system, grouting system and guide frame adjustment mechanism, and the status response data of the pipe section under actual control action is collected synchronously and recorded in the digital twin database; Based on historical data in the digital twin database, the cumulative control deviation between the actual state and the preset design target state is calculated; The meta-optimization algorithm is adopted to optimize the key parameters of each module offline with the goal of minimizing the long-term cumulative control deviation. The key parameters include: feature selection threshold in S2, neural network learning rate in S3, and risk sensitivity coefficient in S4. The optimized key parameters are sent back to the corresponding functional modules.

[0020] This invention provides a dynamic deviation correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing. The method comprises the following steps: S1: Acquiring laser and fiber optic data; compensating for the fiber optic data by automatically identifying the temperature excitation window and calculating the temperature coupling coefficient using least-squares fitting to obtain true strain data; and calculating global pose parameters through online optimization based on the cubic norm. S2: Generating eigenmode components based on the true strain and global pose parameters; filtering based on the calculated physical coupling feature contribution to construct an intelligent feature vector. S3: Constructing a physical information-driven state estimation model based on the intelligent feature vector and the mechanical and physical constraints of the pipe section, and outputting the pipe section state estimate in real time. S4: Using the pipe section state estimate, dynamically predicting the construction deviation of the pipe section and generating the optimal control sequence and optimal control command. S5: Applying the optimal control command to the construction guidance, jack adjustment, and grouting device; and adaptively updating the key system parameters based on actual construction feedback data using a meta-optimization algorithm. The resulting beneficial effects include: High-precision multimodal data fusion and real-time correction: By fusing laser ranging and fiber optic strain data in multiple modes and combining the contribution of physical coupling features to generate intelligent feature vectors, high-precision real-time estimation of the pipe segment status is achieved; effectively overcoming the limitations of single sensor accuracy and environmental interference, and improving the accuracy and reliability of pipe segment position and attitude control in tunnel pipe curtain construction.

[0021] Adaptive control based on physical constraints and intelligent prediction: It introduces a physical information-driven state estimation model and probabilistic prediction method, combined with pipe section mechanical constraints and risk-sensitive optimization algorithms, to achieve future state prediction and optimal control sequence generation; it can adaptively cope with construction vibration, geological changes and temperature disturbances, improve the dynamic correction capability of construction deviations, and achieve closed-loop robust control.

[0022] Closed-loop execution and system parameter meta-optimization: By applying the optimal control commands to the jacks, grouting and guide frame devices, and recording actual construction feedback data into the digital twin database, the meta-optimization algorithm is used to adaptively update key parameters, thereby achieving self-optimization and continuous accuracy improvement in the construction process, and improving the overall intelligence level and long-term construction accuracy of tunnel pipe curtain construction.

[0023] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 The flowchart illustrates a dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing, as an exemplary embodiment of the present invention. Detailed Implementation

[0025] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0026] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components. In actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0027] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention. Example

[0028] A dynamic correction method for tunnel pipe jacking construction based on laser ranging and fiber optic sensing, such as Figure 1 As shown, it includes: S1: Acquire laser and fiber optic data, calculate the temperature coupling coefficient by least squares fitting through automatic identification of the temperature excitation window to compensate for the fiber optic data, obtain real strain data, and calculate global pose parameters through online optimization based on the cubic norm. S2: Generate eigenmode components based on real strain and global pose parameters, and filter them according to the calculated contribution of physical coupling features to construct intelligent feature vectors; S3: Based on the intelligent feature vector and the mechanical and physical constraints of the pipe section, a physical information-driven state estimation model is constructed to output the estimated state value of the pipe section in real time; S4: Using the estimated state values ​​of pipe sections, dynamically predict the construction deviations of pipe sections and generate the optimal control sequence and optimal control commands; S5: Apply the optimal control commands to the construction guidance, jack adjustment and grouting device, and adaptively update the key parameters of the system based on actual construction feedback data through a meta-optimization algorithm.

[0029] The present invention is further configured such that S1 includes: an initial calibration unit and an online optimization and calibration unit; The initial calibration unit includes: Collect laser ranging data arranged at known coordinate control points and construct a ranging dataset; The temperature excitation window is automatically identified, and the raw strain data and temperature measurement data of the fiber optic sensor in the window are collected to construct a strain dataset. A robust optimization method is used to solve for the pose parameters of the laser rangefinder in the global coordinate system in the ranging dataset, and these parameters are set as the initial global pose parameters. Using a pre-calibrated temperature coupling coefficient, temperature compensation is performed on the raw strain data to output the true strain data of the structure. Specifically, laser ranging data is distance measurement data collected by laser rangefinders installed at control points with known coordinates in the tunnel construction area, used to determine the position and attitude of the laser rangefinders in the global coordinate system; laser ranging data acquisition involves arranging several control points with known three-dimensional coordinates on the stable bedrock or completed structure in the tunnel construction area, and installing a laser rangefinder at each control point. The distance measurement values ​​from each rangefinder to the target point on the surface of the pipe section are collected periodically and organized in chronological order to form a ranging dataset. The raw strain data of the fiber optic sensors are the strain measurement values ​​of distributed fiber optic sensors on the construction pipe section, reflecting the stress and deformation of the pipe section; the temperature measurement data are the real-time temperature values ​​at the corresponding locations of the fiber optic sensors, used for temperature compensation; fiber optic sensing data acquisition involves arranging distributed fiber optic sensors on or inside the construction pipe section, such as those based on Brillouin scattering or FBG principles, to monitor the strain distribution of the pipe section in real time and simultaneously collect the temperature measurement values ​​at the corresponding locations of each sensing unit. Strain and temperature data are organized according to sensor unit number and timestamp to form the original strain-temperature dataset. The initial global pose parameters are the three-dimensional position and orientation of the laser rangefinder in the global coordinate system, including displacement and orientation information. Solving for the initial global pose parameters involves inputting the obtained ranging dataset into a robust optimization algorithm, such as RANSAC registration or iterative weighted least squares, to minimize the ranging error and solve for the three-dimensional position and orientation of each laser rangefinder in the global coordinate system, i.e., the rotation matrix and translation vector. The temperature coupling coefficient is a correction coefficient reflecting the influence of temperature changes on fiber optic strain measurement, used to calculate the true strain of the structure from the original strain. Temperature coupling coefficient calibration is achieved by automatically analyzing the obtained fiber optic data to identify time periods where the pipe section structure is in a stress-free state, the strain data fluctuation variance is below a threshold, but the temperature change is significant, serving as the temperature excitation window. Within this window, linear regression or least squares is used to fit the relationship between the strain change and temperature change of each fiber optic sensing unit to obtain its temperature coupling coefficient. The true strain data of the structure is the actual strain of the structure caused by stress or deformation after temperature compensation, which is the strain data measured by the fiber optic sensor. By using the updated temperature coupling coefficient as the adjustment coefficient, temperature compensation is performed on the original strain data of the fiber optic sensor in all acquisition windows at the current moment to eliminate temperature-induced interference and obtain the true strain data of the structure.

[0030] The present invention is further configured such that the online optimization and calibration unit includes: During system operation, temperature changes are monitored during periods when the structure is in a stress-free state, and the temperature coupling coefficient is dynamically updated using linear regression analysis. Based on the real-time acquired structural strain data, the displacement changes of structural feature points are calculated to construct the structural displacement change. By combining the changes in the real-time range measurement dataset with the changes in structural displacement, an optimization method is used to complete the online correction of the initial global pose parameters, which are then set as the global pose parameters. Specifically, during construction and operation, the system identifies stable periods with no stress changes in the structure, such as nighttime when there is no construction or when ground disturbance is minimal. During these periods, structural strain mainly originates from temperature changes. Linear regression analysis is used to establish the relationship between temperature changes and fiber optic strain changes. The temperature coupling coefficient is recalculated using the latest monitoring data and automatically replaces the original coefficients in the system, achieving adaptive compensation for different environmental conditions. Distributed fiber optic sensing technology combined with time series analysis can utilize mature linear regression algorithms, such as least squares regression or recursive least squares, to achieve real-time coefficient updates. The displacement change of structural feature points is the displacement increment of a specific monitoring point on the structure calculated from real strain data, used to reflect the spatial movement of the pipe section under construction or environmental disturbance. By utilizing the structural geometric characteristics and the fiber optic deployment location, the local real strain data is integrated and converted into the displacement change of feature points. The method can use distributed fiber grating or Brillouin scattering strain measurement technology, combined with finite element model or geometric relationship, to map the strain curve into the displacement increment of feature points, forming a real-time structural displacement change, used to reflect the actual stress and deformation of the pipe section. The change in the ranging dataset is the difference between consecutive distance measurements acquired by the laser ranging system at adjacent time points. It represents the minute displacement change of the construction pipe segment within a short time interval, reflecting the instantaneous motion state of the pipe segment, and is used to identify the relative displacement change of the global reference point in the three-dimensional coordinate system. By calculating the distance difference of the laser ranging system at adjacent time points, the change in the reference point in three-dimensional space is obtained. Then, the structural displacement change and the laser ranging change are input into a robust optimization algorithm for data fusion. The robust optimization algorithm can adopt the following algorithms, but no limitation is made on the algorithm: iterative weighted least squares, RANSAC, or robust point cloud registration, etc., which automatically suppress the influence of outliers on the calculation. The optimization algorithm calculates the corrected global pose parameters, that is, the latest position and attitude of the laser rangefinder in the global coordinate system, realizing real-time spatial reference system update.

[0031] The present invention is further configured such that S2 includes: a multi-source data preprocessing unit, an adaptive signal decomposition unit, and an intelligent feature selection unit; The multi-source data preprocessing unit includes: Based on the real-time acquired ranging dataset, the difference between laser ranging values ​​at adjacent times is calculated to obtain the change in the ranging dataset, while the real strain data output by S1 is received. Digital filtering is used to eliminate noise interference on the range measurement dataset, and anomaly detection is used to remove abnormal measurement values ​​from the strain data. The processed ranging dataset changes and strain data are combined into a multimodal data vector. Specifically, the ranging dataset changes are the differences between adjacent distance measurements continuously acquired by the laser ranging system, representing the minute displacement changes of the construction pipe segment within a short time interval, reflecting the instantaneous motion state of the pipe segment, and used to identify the relative displacement changes of the global reference point in the three-dimensional coordinate system. Laser ranging data is read sequentially, and the differences between adjacent distance measurements are calculated to obtain a short-interval displacement increment sequence. The true strain data comes from the output of S1 and consists of temperature-compensated fiber optic strain measurements, used to reflect the deformation level of the pipe segment under actual stress. The digital filtering denoising and strain anomaly detection are implemented using the following methods: Digital filtering denoising applies a classic low-pass digital filter or wavelet threshold denoising technique to the displacement increment sequence to eliminate high-frequency random noise and retain the true structural motion information; strain anomaly detection utilizes anomaly detection methods based on statistical thresholds or isolated forests to remove isolated anomalous measurement points in the fiber optic strain data, ensuring the reliability of the true strain data. Finally, the filtered displacement increment sequence and the anomaly-removed true strain sequence are aligned by timestamps to form a two-dimensional time series data matrix, constituting a multimodal data vector. The multimodal data vector is a composite vector formed by aligning the processed displacement increments and true strain data on a time series, comprehensively representing the structure's motion and stress information. The specific implementation method is as follows: First, the ranging values ​​of each control point are read from the laser ranging system in chronological order. Comparing two adjacent measurements yields a displacement increment sequence representing minute structural displacement changes over a short period. Next, a low-pass digital filtering method is used to remove high-frequency random noise from the ranging sequence, preserving the true motion trend. Simultaneously, the temperature-compensated fiber optic strain data output from S1 is called, and anomaly detection is performed, for example, using an isolated forest algorithm to identify and remove sudden anomalies, ensuring data stability and reliability. Subsequently, the processed displacement increment sequence and the true strain sequence are synchronously aligned by timestamps to form a multimodal data vector containing motion and stress information.

[0032] The present invention is further configured such that the adaptive signal decomposition unit includes: An adaptive variational mode decomposition algorithm is used to process multimodal data vectors. Through an iterative optimization process, the multimodal data vectors are decomposed into multiple time-frequency components at different frequencies and with time-varying characteristics, and multiple sets of intrinsic mode components are extracted. Simultaneously extract the center frequency parameters of each group of intrinsic mode components; Multiple sets of intrinsic modal components and their corresponding center frequency parameters are passed to the intelligent feature selection unit. Specifically, the intrinsic modal components are time-series components with different center frequencies, independently characterizing local features, decomposed from the multimodal data vector using an adaptive variational mode decomposition algorithm, reflecting the dynamic characteristics of the structure at different time scales. Simultaneously, the center frequency parameter represents the energy concentration frequency of each set of intrinsic modal components, used to identify the dynamic characteristics represented by the component, such as high-frequency vibration or low-frequency displacement trends. This is achieved using a variational mode decomposition algorithm, which iteratively optimizes and extracts multiple narrowband modes simultaneously. This algorithm is more stable than traditional EMD and avoids endpoint effects. The specific decomposition process is as follows: the multimodal data vector is input into the VMD framework, and the center frequency of each mode is automatically adjusted in the frequency domain until an optimal balance is reached, giving each set of output intrinsic modal components a clear frequency band characteristic. After each iteration, the algorithm outputs the center frequency parameter of each set of modes to characterize the main energy distribution location of that mode. The specific implementation method involves inputting this multimodal data vector into the variational mode decomposition algorithm. The algorithm automatically divides multiple narrowband frequency bands through iterative optimization, updating the center frequency of each mode in each iteration until all modes converge and can characterize the dynamic characteristics at different scales. The output consists of multiple sets of intrinsic mode components, each corresponding to a specific center frequency parameter, identifying its main dynamic frequency band.

[0033] The present invention is further configured such that the intelligent feature selection unit includes: Based on the intrinsic mode components of each group and the preset characteristics of the current tunnel segment motion mode, the contribution of physical coupling characteristics is calculated. Based on a preset feature selection threshold, the intrinsic mode components with the highest contribution to the physical coupling features are selected. The selected intrinsic modal components are weighted and fused using their corresponding physical coupling feature contributions as weights to generate a fused intelligent feature vector. Specifically, the physical coupling feature contribution measures the correlation strength between each intrinsic modal component and the known motion mode of the tunnel segment, and is used to select modes that are highly coupled with actual physical phenomena. The intelligent feature vector is the comprehensive feature result of the selected high-contribution modes weighted and fused according to their contributions, and is the core input of subsequent prediction and control algorithms. Each group of intrinsic modal components is compared with a preset tunnel segment motion mode feature library. The physical coupling feature contribution of each group of modes is calculated using correlation coefficients, mutual information, or feature mapping algorithms based on graph convolutional networks. Based on a pre-determined feature selection threshold, such as a contribution greater than a certain set proportion, the group or multiple groups of modes with the highest contribution are retained. Finally, the selected modes are weighted and summed according to their respective contributions to generate a single intelligent feature vector that combines multi-frequency dynamic characteristics and actual physical meaning. Specific implementation method: Finally, each group of intrinsic modal components is coupled with the pre-established tunnel segment motion mode features for analysis. By calculating the matching degree between each modality group and the pattern in terms of temporal correlation, energy distribution, and feature mutual information, the physical coupling feature contribution of each modality is obtained. Based on a preset feature selection threshold, the set of modes with the highest contribution is selected, and these modal components are weighted and fused using their contribution values. The resulting intelligent feature vector retains multi-scale dynamic change information and is highly consistent with the actual motion pattern of the tube segment, making it a key input for subsequent attitude correction and predictive analysis.

[0034] The present invention is further configured such that S3 includes: Perform system initialization: Check if the pipe section status value at the previous moment is valid. If it is invalid, use the preset status value to initialize the status. If it is in an active state, it receives the intelligent feature vector obtained from S2 and the previous node state; A neural network model is established based on a preset neural network learning rate. The intelligent feature vector and the previous node state are used as network inputs. The preliminary estimate of the current state is obtained through forward propagation of the neural network. Based on real-time sensor measurement data, coordinate transformation and calculation are performed using the global pose parameters of the S1 dual output to derive the reference state value. Specifically, the previous segment state value is the estimated three-dimensional attitude and displacement information of the segment at the previous sampling time, used as the temporal input for the neural network; the preset state value is the initial value of the segment state used to start estimation during system initialization, such as the initial attitude obtained based on design drawings or construction measurements; the neural network learning rate is the step size for weight updates during neural network training, used to control the model convergence speed and stability; the reference state value is the reference value of segment attitude and displacement calculated by combining real-time sensor data with global pose parameters. By restarting the construction monitoring system after startup or data stream interruption, the system first checks whether the previous pipe segment status value exists and meets validity conditions. If it meets preset rules such as a normal data timestamp, a value within a reasonable engineering range, or a non-empty value, it is deemed valid. If the status value is invalid, the system reads a preset status value from the design drawings or initial survey data as the initial state input to the neural network to ensure the continuity of state estimation. If the status value is valid, it is directly used as the temporal part of the network input. Then, the system receives the intelligent feature vector from S2 and combines it with the previous pipe segment status value to form a temporal input vector. A temporal neural network is established according to a preset neural network learning rate. For establishing the temporal neural network, a Long Short-Term Memory (LSTM) network, a Gated Recurrent Unit (GRU), or other deep temporal models can be used. Forward propagation calculations are then performed to output a preliminary estimate of the current pipe segment status, including three-dimensional displacement, attitude angles, and structural deformation. The model output yields the preliminary state estimate. The system synchronously reads real-time data from laser ranging and fiber optic strain sensors. Based on the global pose parameters output by S1, it performs a coordinate transformation on these sensor data and then uses spatial geometry calculations to obtain a reference state value for the pipe segment consistent with the global coordinate system. This reference state value serves as the benchmark for data-driven estimation. In simpler terms, the specific implementation method is as follows: During tunnel jacking construction, the system first checks whether the pipe segment state at the previous moment is valid. If the data is missing or abnormal, the initial pose measured in the construction design drawings or previous measurements is used as the initial state input. Subsequently, the intelligent feature vector from multi-source data fusion is combined with the previous state value and input into a deep temporal neural network with a pre-set learning rate. A preliminary estimate of the current pipe segment state is obtained through forward propagation. Simultaneously, the system receives real-time data from laser ranging and fiber optic strain sensors and uses the global pose parameters obtained from S1 to perform coordinate transformation and calculate the reference state value for the current pipe segment.

[0035] The present invention is further configured to calculate the data difference between the preliminary estimate and the reference state value; Simultaneously, based on the Euler-Bernoulli beam theory, physical equations are established and constrained to obtain the predicted values ​​of the equations. The physical consistency deviation between the preliminary estimate and the predicted values ​​of the equations is then calculated. A multi-objective optimization function is constructed by combining data differences and physical consistency deviations, and the neural network model is optimized through the backpropagation algorithm. The system monitors changes in geological conditions and process parameters during construction, adjusts neural network parameters online based on real-time monitoring data, and outputs an estimated value of the pipe segment's state at the current moment. Specifically, data discrepancy is the deviation between the preliminary estimate and the reference state value, used to evaluate the error of the data-driven model; equation prediction values ​​are the predicted values ​​of pipe segment displacement and deformation calculated based on the mechanical model established by Euler-Bernoulli beam theory, used to provide physical constraints; physical consistency deviation is the deviation between the preliminary estimate and the equation prediction value, used to measure the consistency between the neural network output and the mechanical theory; the multi-objective optimization function is a comprehensive evaluation index that considers both data discrepancy and physical consistency deviation, used to guide the backpropagation update of the neural network. Based on the Euler-Bernoulli beam theory, a mechanical prediction model is established using known structural parameters of the pipe section, such as the elastic modulus, moment of inertia, and boundary conditions specified in the engineering documents. Under current loads and boundary conditions, the theoretical deformation and displacement of the pipe section are calculated, yielding the model output: the predicted values ​​from the equations. The data differences between the preliminary estimates and the reference state values, as well as the physical consistency deviation between the preliminary estimates and the predicted values ​​from the equations, are then calculated. These data differences and physical consistency deviations are used as a joint objective to construct a multi-objective optimization function. A backpropagation algorithm is employed to update the weights of the neural network parameters, ensuring the model simultaneously approximates sensor measurements while satisfying structural mechanical constraints. During construction, changes in geological conditions, such as surrounding rock pressure and groundwater level, as well as technological parameters, such as grouting pressure and jack thrust, are continuously monitored. The learning rate and weight update strategy of the neural network are adjusted in real time to achieve online adaptive optimization. The updated neural network outputs the estimated state of the pipe section at the current moment. In summary, the specific implementation method is as follows: Based on the Euler-Bernoulli beam theory and combined with the material and geometric properties of the pipe section, a mechanical prediction model is established. Under the current construction load and boundary conditions, the theoretical deformation and displacement are obtained, yielding the predicted values ​​of the equations. The system calculates the data differences between the preliminary estimates and the reference state values, as well as the physical consistency deviations between the preliminary estimates and the predicted values ​​of the equations. Based on this, a multi-objective optimization function is constructed, and the neural network weights are updated through a backpropagation algorithm. If geological conditions or process parameters change during construction, the system dynamically adjusts the learning rate and parameter update strategy of the neural network based on real-time monitoring data, achieving online adaptation of the model. The final output pipe section state estimate accurately reflects the current spatial attitude and structural deformation of the pipe section, providing a highly reliable input for subsequent deviation prediction and control.

[0036] The present invention is further configured such that S4 includes: Based on the estimated state of the pipe segment at the current moment, a probabilistic prediction method is used to perform multi-step forward state prediction, and the prediction uncertainty of the future state is quantified by Monte Carlo sampling or Bayesian inference method to obtain the future state prediction sequence. Based on the future state prediction sequence, and combined with the preset risk sensitivity coefficient, a control optimization problem using stochastic model prediction is constructed. The optimal control sequence is obtained by using an optimization algorithm that minimizes conditional risk value. Optimal control commands are generated based on the optimal control sequence. These commands include jack displacement, grouting pressure, and guide frame adjustment. Specifically, using the current segment state estimate output by S3 as the initial condition, a probabilistic prediction model is invoked for multi-step forward prediction. The prediction model can be one of the following: 1. Bayesian temporal network: Calculates the probability distribution of future states by establishing the prior distribution of state transitions and the posterior distribution of observed data; 2. Particle filter or unscented Kalman filter: Recursively estimates future states under nonlinear and non-Gaussian noise conditions. To quantify prediction uncertainty, Monte Carlo sampling can be used to generate a large number of possible future state trajectories, or the confidence interval of future states can be directly output using Bayesian inference methods to obtain a future state prediction sequence containing multiple future moments. Using the predicted future state sequence as input, a risk sensitivity coefficient is set based on construction safety and accuracy requirements. This risk sensitivity coefficient is a risk weight parameter preset by the user according to engineering safety requirements, used to balance expected performance and risk fluctuations in control optimization. It is used to weigh control performance against the potential risks of future uncertainty. Under the framework of stochastic model predictive control, an optimization problem is established with the goal of minimizing future construction deviations and control energy consumption. The risk sensitivity coefficient is embedded in the target weight, and risk measurement methods such as conditional value of risk are introduced to ensure that the control strategy not only optimizes average performance but also reduces the probability of high-risk events in extreme cases. Existing stochastic optimization algorithms are selected for solving the problem, such as: 1. Scenario optimization: Monte Carlo samples of future states are divided into scenarios, and the solution is obtained in each scenario with weighted results; 2. Partial robustness optimization: By establishing a confidence set for the uncertainty distribution, the most conservative but stable control strategy is obtained. During the iterative optimization process, the risk sensitivity coefficient is used as a regulating factor to ensure that the control strategy is robust under different future state scenarios. Finally, the optimal control sequence is obtained. The optimal control sequence is the set of optimal control variables over a period of time obtained through stochastic model predictive control, including the control actions at each future prediction time. The optimal control sequence is transformed into a set of instructions executable by the actual equipment: optimal control instructions. These instructions are engineering commands that translate the optimal control sequence into actual execution, including jack displacement adjustments, grouting pressure setpoints, and guide frame attitude adjustments. These are directly executed by the construction equipment and specifically include: 1. Jack displacement: The thrust displacement of each thrust cylinder, ensuring the horizontal and vertical position adjustment of the pipe section; 2. Grouting pressure: The target pressure value of the grouting pump, used to control the stability of the strata surrounding the tunnel lining; 3. Guide frame adjustment: The attitude correction amount of the guide frame, used to fine-tune the attitude of the pipe section to ensure the accuracy of the tunnel centerline. The control instructions are sent to the corresponding actuators in real time, driving the coordinated action of each device through the field control system. Specifically, during the dynamic correction process of tunnel pipe curtain construction, the system first reads the estimated state of the pipe section at the current moment output by the S3 module as the initial condition for prediction.Subsequently, a probabilistic prediction model is initiated to extrapolate the state over multiple future construction time steps. The prediction model, based on Bayesian time series methods or particle filtering techniques, recursively generates the probability distribution of future states while considering sensor noise and geological disturbances. To account for prediction uncertainty, the system generates a large number of future state trajectories through Monte Carlo sampling and statistically analyzes the distribution characteristics of these samples, forming a prediction sequence containing displacement, attitude, and structural deformation at multiple future time points, with confidence intervals marked for each time point in the sequence. After obtaining the future state prediction sequence, the system establishes an optimization problem for stochastic model predictive control based on a pre-set risk sensitivity coefficient. This optimization problem comprehensively considers future pipe segment attitude deviations, control energy consumption, and uncertainty risks, using the risk sensitivity coefficient as a weight to balance construction accuracy and safety margins under extreme conditions. The optimization objective adopts a conditional risk value minimization strategy, ensuring control stability is maintained even under high-risk conditions through distribution analysis of the future scenarios generated by Monte Carlo. Subsequently, scenario optimization or bilabial bar optimization algorithms are used for iterative solution, continuously refining the control strategy in multiple rounds of calculation, ultimately outputting the optimal control sequence over a certain time range. Finally, the system transforms the optimal control sequence into specific executable instructions. These include: displacement adjustment instructions for the jacks, used to precisely correct the horizontal and vertical positions of the tunnel sections in three-dimensional space; grouting pressure setpoints, used to stabilize the soil surrounding the tunnel lining and prevent excessive deformation; and attitude adjustment amounts for the guide frame, used to fine-tune the installation angle of the tunnel sections to ensure the accuracy of the tunnel centerline. The generated control instructions are sent to each actuator through the on-site automated control system, driving the jacks, grouting pumps, and guide frame equipment to operate in coordination, achieving dynamic correction and robust control of tunnel section construction deviations.

[0037] The present invention is further configured such that S5 includes: The optimal control command sequence is sent to the actuator to control the coordinated action of the jack system, grouting system and guide frame adjustment mechanism, and the status response data of the pipe section under actual control action is collected synchronously and recorded in the digital twin database; Based on historical data in the digital twin database, the cumulative control deviation between the actual state and the preset design target state is calculated; The meta-optimization algorithm is adopted to optimize the key parameters of each module offline with the goal of minimizing the long-term cumulative control deviation. The key parameters include: feature selection threshold in S2, neural network learning rate in S3, and risk sensitivity coefficient in S4. The optimized key parameters are then fed back to the corresponding functional modules. Specifically, the digital twin database is a database used to record the actual state data of the construction site in real time. It contains historical pipe section state estimates, sensor measurements, control commands and their execution results, supporting subsequent deviation calculations and parameter optimization. Cumulative control deviation refers to the cumulative quantitative index of the difference between the actual state of the construction pipe section and the preset design target state over a period of time, used to measure the long-term accuracy of the control system. The meta-optimization algorithm is a high-level optimization method used to automatically adjust the key parameters of the lower-level optimization modules. Typical algorithms include genetic algorithms, particle swarm optimization, or Bayesian optimization, which can search for the globally optimal parameter combination based on long-term construction data. Key parameters include: Feature selection threshold in S2: a threshold for screening physically coupled features in multimodal data, affecting the composition of intelligent feature vectors; Neural network learning rate in S3: the step size parameter of the neural network model in backpropagation optimization, directly affecting the convergence speed and accuracy of pipe section state estimation; Risk sensitivity coefficient in S4: an adjustment coefficient for risk preference in control stochastic model prediction optimization, used to balance control accuracy and risk tolerance during control optimization. Specific implementation method: In the real-time control stage of tunnel segment construction, the optimal control command sequence output from the previous stage S4 is first sent to the jacks, grouting system, and guide frame adjustment mechanism through the industrial automation communication system, enabling the three actuators to coordinate the attitude and position adjustment of the tunnel segment. The on-site sensor array continuously collects real-time status response data of the tunnel segment under actual control actions. This data includes displacement, attitude angle, strain, and the actual strain value after temperature compensation, and is stored in a digital twin database through a time synchronization mechanism. Subsequently, the system retrieves the actual status data within a construction cycle from the database, compares it hourly with the predefined design target state, calculates the control error at each moment within that cycle, and accumulates these errors to obtain a cumulative control deviation index reflecting long-term control accuracy. This cumulative control deviation serves as the target value for subsequent meta-optimization, used to measure the overall deviation level of the system's long-term operation. Based on this, the system initiates a meta-optimization algorithm. In the genetic algorithm mode, the initial ranges of feature selection threshold, neural network learning rate, and risk sensitivity coefficient are first defined, and multiple sets of parameter combinations are randomly generated as the initial population. Each set of parameter combinations is used to reproduce the control effect of the construction data, and the fitness is evaluated based on the calculated cumulative control deviation. Through continuous selection, crossover, and mutation operations, the genetic algorithm gradually eliminates suboptimal parameters and retains the optimal solution, eventually converging to the optimal parameter combination that minimizes the cumulative control deviation. If Bayesian optimization is used, the optimal parameter direction is predicted by a Gaussian process surrogate model, and the search region is updated through iterative sampling, similarly finding the optimal parameter set that minimizes the cumulative control deviation.Finally, the system feeds back the optimized key parameters to modules S2, S3, and S4 respectively: S2's intelligent feature selection unit automatically updates the feature selection threshold, thereby improving the accuracy of multimodal feature extraction; S3's neural network model updates its learning rate, improving the convergence performance and stability of state estimation; and S4's risk sensitivity coefficient is reset to better balance control accuracy and risk tolerance in future control optimizations. After the update, the next cycle of construction control will automatically adopt the optimized parameter combination, achieving system self-evolution and long-term performance improvement.

[0038] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic correction method for tunnel pipe jacking construction based on laser ranging and fiber optic sensing, characterized in that, include: S1: Acquire laser and fiber optic data, calculate the temperature coupling coefficient by least squares fitting through automatic identification of the temperature excitation window to compensate for the fiber optic data, obtain real strain data, and calculate global pose parameters through online optimization based on the cubic norm. S2: Generate eigenmode components based on real strain and global pose parameters, and filter them according to the calculated contribution of physical coupling features to construct intelligent feature vectors; S3: Based on the intelligent feature vector and the mechanical and physical constraints of the pipe section, a physical information-driven state estimation model is constructed to output the estimated state value of the pipe section in real time; S4: Using the estimated state values ​​of pipe sections, dynamically predict the construction deviations of pipe sections and generate the optimal control sequence and optimal control commands; S5: Apply the optimal control commands to the construction guidance, jack adjustment and grouting device, and adaptively update the key parameters of the system based on actual construction feedback data through a meta-optimization algorithm.

2. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 1, characterized in that, S1 includes: an initial calibration unit and an online optimization and calibration unit; The initial calibration unit includes: Collect laser ranging data arranged at known coordinate control points and construct a ranging dataset; The system automatically identifies the temperature excitation window, collects the raw strain data and temperature measurement data from the fiber optic sensor within the window, and constructs a strain dataset. A robust optimization method is used to solve for the pose parameters of the laser rangefinder in the global coordinate system in the ranging dataset, and these parameters are set as the initial global pose parameters. Using a pre-calibrated temperature coupling coefficient, temperature compensation is performed on the original strain data to output the true strain data of the structure.

3. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 2, characterized in that, The online optimization and calibration unit includes: During system operation, temperature changes are monitored during periods when the structure is in a stress-free state, and the temperature coupling coefficient is dynamically updated using linear regression analysis. Based on the real-time acquired structural strain data, the displacement changes of structural feature points are calculated to construct the structural displacement change. By combining the changes in the real-time range measurement dataset with the changes in structural displacement, an optimization method is used to complete the online correction of the initial global pose parameters, which are then set as the global pose parameters.

4. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 1, characterized in that, The S2 includes: a multi-source data preprocessing unit, an adaptive signal decomposition unit, and an intelligent feature selection unit; The multi-source data preprocessing unit includes: Based on the real-time acquired ranging dataset, the difference between laser ranging values ​​at adjacent times is calculated to obtain the change in the ranging dataset, while the real strain data output by S1 is received. Digital filtering is used to eliminate noise interference on the range measurement dataset, and anomaly detection is used to remove abnormal measurement values ​​from the strain data. The processed range measurement dataset changes and strain data are combined into a multimodal data vector.

5. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 4, characterized in that, The adaptive signal decomposition unit includes: An adaptive variational mode decomposition algorithm is used to process multimodal data vectors. Through an iterative optimization process, the multimodal data vectors are decomposed into multiple time-frequency components at different frequencies and with time-varying characteristics, and multiple sets of intrinsic mode components are extracted. Simultaneously extract the center frequency parameters of each group of intrinsic mode components; Multiple sets of intrinsic mode components and their corresponding center frequency parameters are transmitted to the intelligent feature selection unit.

6. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 5, characterized in that, The intelligent feature selection unit includes: Based on the intrinsic mode components of each group and the preset characteristics of the current tunnel segment motion mode, the contribution of physical coupling characteristics is calculated. Based on a preset feature selection threshold, the intrinsic mode components with the highest contribution to the physical coupling features are selected. The selected intrinsic mode components are weighted and fused using their corresponding physical coupling feature contribution as weights to generate a fused intelligent feature vector.

7. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 1, characterized in that, S3 includes: Perform system initialization: Check if the pipe section status value at the previous moment is valid. If it is invalid, use the preset status value to initialize the status. If it is in an active state, it receives the intelligent feature vector obtained from S2 and the previous node state; A neural network model is established based on a preset neural network learning rate. The intelligent feature vector and the previous node state are used as network inputs. The preliminary estimate of the current state is obtained through forward propagation of the neural network. Based on the real-time sensor measurement data, coordinate transformation and calculation are performed using the global pose parameters of the S1 dual output to derive the reference state value.

8. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 7, characterized in that, Calculate the data difference between the preliminary estimate and the reference state value; Simultaneously, based on the Euler-Bernoulli beam theory, physical equations are established and constrained to obtain the predicted values ​​of the equations. The physical consistency deviation between the preliminary estimate and the predicted values ​​of the equations is then calculated. A multi-objective optimization function is constructed by combining data differences and physical consistency deviations, and the neural network model is optimized through the backpropagation algorithm. The system monitors changes in geological conditions and process parameters during construction, adjusts neural network parameters online based on real-time monitoring data, and outputs an estimated value of the pipe section status at the current moment.

9. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 1, characterized in that, S4 includes: Based on the estimated state of the pipe segment at the current moment, a probabilistic prediction method is used to perform multi-step forward state prediction, and the prediction uncertainty of the future state is quantified by Monte Carlo sampling or Bayesian inference method to obtain the future state prediction sequence. Based on the future state prediction sequence, and combined with the preset risk sensitivity coefficient, a control optimization problem using stochastic model prediction is constructed. The optimal control sequence is obtained by using an optimization algorithm that minimizes conditional risk value. Optimal control commands are generated based on the optimal control sequence. These control commands include: jack displacement, grouting pressure, and guide frame adjustment.

10. The dynamic correction method for tunnel pipe curtain construction based on laser ranging and fiber optic sensing according to claim 1, characterized in that, S5 includes: The optimal control command sequence is sent to the actuator to control the coordinated action of the jack system, grouting system and guide frame adjustment mechanism, and the status response data of the pipe section under actual control action is collected synchronously and recorded in the digital twin database; Based on historical data in the digital twin database, the cumulative control deviation between the actual state and the preset design target state is calculated; The meta-optimization algorithm is adopted to optimize the key parameters of each module offline with the goal of minimizing the long-term cumulative control deviation. The key parameters include: feature selection threshold in S2, neural network learning rate in S3, and risk sensitivity coefficient in S4. The optimized key parameters are then sent back to the corresponding functional modules.

Images