Cable-stayed bridge cross brace intelligent positioning method and device for construction state data driving

Abstract

The invention provides an intelligent positioning method and device for a cross brace of a cable-stayed bridge driven by construction state data. The method comprises the steps that phase-locked collection of camera exposure and double-angle oblique illumination is achieved based on a linear speed encoder and a cutterhead phase; executing speed compensation and geometric shearing on the continuous frames to obtain a space-time registration sequence; extracting a sub-pixel edge curve and generating a narrow-band region of interest along a normal direction; calculating a flicker ratio and a relative phase difference between alternating illumination frames, and generating a burr saliency map by combining a normal gradient and fine tooth period energy along the edge direction; inputting the saliency map and the normal brightness profile into a convolutional neural network, and outputting a pixel-level burr probability, an amplitude proxy amount and a period score; and self-adaptive threshold judgment is carried out by combining the linear speed, the tension and the foil thickness, and an alarm is given by associating with a cutter phase. The detection method and system provided by the invention realize high detection rate, low false alarm and millisecond-level delay under the conditions of high light reflection and linear speed greater than or equal to 200m / min, and are suitable for multi-tool-path online quality control and process linkage.

Description

Technical Field

[0001] This invention relates to the field of intelligent positioning technology, specifically to an intelligent positioning method and device for cross bracing of cable-stayed bridges driven by construction status data. Background Technology

[0002] As one of the main forms of long-span bridges, cable-stayed bridges typically require temporary or permanent cross bracing during the construction phase to improve the overall structural stiffness and spatial stability, and to limit the relative deformation of the bridge towers, main beams, and stay cables. The position, orientation, and coordination with surrounding components of the cross bracing directly affect the alignment control, internal force distribution, and the safety and durability of the final bridge structure during construction.

[0003] In current engineering practice, the arrangement and positioning of cross braces in cable-stayed bridges are mainly based on the theoretical alignment and structural layout given in the design. Traditional surveying methods such as total stations and levels are used for layout, and installation adjustments are made in conjunction with the experience of construction personnel. However, as construction conditions evolve, the bridge towers and main beams undergo significant time-varying deformation under the influence of multiple factors such as temperature, wind load, cable tension adjustments, and temporary support deformation. The actual spatial position often deviates from the theoretical design value. However, the positioning of the cross braces is still largely based on "static design values ​​plus a small number of measuring point corrections," making it difficult to reflect the true stress and geometric state of the structure in a timely manner.

[0004] To improve construction control precision, some projects have begun deploying monitoring equipment such as displacement gauges, strain gauges, cable tension gauges, and GNSS sensors to form a structural health monitoring system during the construction period. This system is used to collect data such as displacement at the top of the bridge tower, deflection of the main beam control section, cable force changes, support reactions, and ambient temperature. However, these monitoring systems typically rely on point sensors, with a limited number of measuring points and fixed locations. This makes it difficult to comprehensively cover the cross brace installation area and surrounding components. The monitoring information suffers from the problem of "local density and overall sparseness," failing to provide detailed information on the local alignment and attitude of the cross braces.

[0005] On the other hand, construction sites are generally equipped with a large number of video surveillance cameras for safety monitoring and recording of the construction process. In existing technologies, video images are mostly limited to manual visual inspection or simple video playback. Even if there are a few attempts to use image processing for displacement or deformation recognition, they are often single-point, offline post-processing analyses. They do not form a unified coordinate system and a unified time scale with multi-source sensor data and structural analysis models, and no systematic solution has been built for precise positioning and real-time verification of cross braces.

[0006] In recent years, digital twin technology has been gradually introduced into the bridge construction field. By establishing a high-precision finite element model corresponding to the actual project and combining monitoring data to calibrate the model parameters, it can reflect the actual stress and geometric state of the structure during the construction phase to a certain extent. However, existing digital twin applications mostly focus on the analysis and evaluation of overall alignment and internal forces, while applications for optimizing the installation position and real-time construction control of local components such as cross braces are still relatively few. In addition, model updates and calculations often rely on traditional finite element forward analysis, which involves large computational loads and limited response speed, making it difficult to support rapid evaluation of multiple schemes and real-time on-site decision-making.

[0007] With the development of data-driven methods such as deep learning, some studies have begun to explore the use of neural networks to predict bridge displacement, deflection, or cable forces to assist in structural condition assessment. However, existing methods are mostly based on sensor data from a single source, failing to fully utilize rich visual information such as video images, and also failing to combine the topological characteristics of the bridge structure with construction-stage working conditions for unified modeling. Furthermore, the lack of effective coupling between the neural network and the structural analysis model limits the physical meaning and interpretability of the network output, hindering its widespread application in engineering practice.

[0008] In summary, existing cross brace positioning technologies for cable-stayed bridges mainly suffer from the following problems: First, they rely excessively on static design alignment and limited measurement point information, making it difficult to accurately reflect the actual structural state under complex construction conditions; second, video monitoring resources are not effectively utilized, failing to provide high-precision displacement and attitude information for the cross brace installation area; third, digital twin models and data-driven models have not yet formed a collaborative mechanism, making it difficult to balance the reliability of physical constraints with the real-time performance of data-driven processes; and fourth, there is a lack of real-time attitude verification and deviation early warning methods based on multi-source data during cross brace installation, and the level of intelligent cross brace positioning and closed-loop control during the construction process needs to be improved.

[0009] Therefore, there is an urgent need for an intelligent positioning method and device for cable-stayed bridge cross braces that can integrate multi-source construction status data and video image information, and introduce data-driven algorithms such as digital twin models and convolutional neural networks to work together, so as to realize rapid optimization decision-making of cross brace positioning scheme and real-time verification of the installation process, thereby improving the alignment control accuracy and construction safety during the construction phase. Summary of the Invention

[0010] To address the aforementioned problems in existing technologies, this invention proposes an intelligent positioning method and device for cross braces of cable-stayed bridges driven by construction status data. This method collects multi-source construction status data, including tower and main girder displacements and cable tensions, and simultaneously acquires video images of the cross brace area. These images are preprocessed and time-aligned to form a multimodal construction status feature vector. A digital twin model of the cable-stayed bridge, calibrated using monitoring data, determines the geometric constraints of the cross brace. Within the design location and construction tolerance, multiple candidate positioning schemes for the cross brace are generated, and positioning feature parameters are constructed. The construction status features, working condition codes, and positioning features are input into a convolutional neural network to predict the structural response and comprehensively score the candidate schemes. The target cross brace positioning scheme is automatically selected, and installation control commands are generated. During the cross brace hoisting process, the end posture of the cross brace is inverted based on the video images. The measured posture is compared with the target posture, and an adjustment prompt is issued when the deviation exceeds the limit, thereby improving the positioning accuracy and construction safety of the cross brace during the construction phase.

[0011] This application provides a data-driven intelligent positioning method for the cross bracing of a cable-stayed bridge, including the following steps:

[0012] S1: Acquire multi-source construction status data during the construction phase of the cable-stayed bridge, and simultaneously acquire on-site video images of the cross bracing area collected by the camera, and generate a working condition coding vector based on the construction procedure information.

[0013] S2: Preprocess the multi-source construction status data to obtain the construction status feature vector; perform camera calibration, target area detection, feature point tracking and three-dimensional attitude inversion on the on-site video images to obtain the three-dimensional displacement and attitude parameters of the cross brace end and cross section, which constitute the video image feature vector, and align it with the construction status feature vector in time to obtain the fused multimodal construction status feature vector.

[0014] S3: Input the fused multimodal construction state feature vector into the digital twin model of the cable-stayed bridge construction stage, calibrate the model parameters, and obtain a digital twin model that matches the current construction state;

[0015] S4: Within the geometric constraints given by the digital twin model, generate multiple candidate schemes for cross bracing positioning based on the cross bracing design position and allowable construction deviations, and construct cross bracing positioning feature parameters for each candidate scheme.

[0016] S5: Input the construction state feature vector of the fused multimodal structure, the working condition encoding vector and the cross brace positioning feature parameters into the convolutional neural network prediction model to obtain the final target cross brace positioning scheme and the corresponding structural response index.

[0017] S6: Generate cross brace installation control commands based on the final target cross brace positioning scheme. During the cross brace installation process, continuously collect video images to estimate the attitude of the cross brace end. Compare the measured attitude with the structural response index. When the deviation exceeds the preset threshold, output an adjustment prompt.

[0018] Preferably, the multi-source construction status data includes bridge tower displacement, main beam displacement, cable tension, strain, temporary support reaction force, and environmental parameters.

[0019] Preferably, the step of generating a working condition coding vector based on construction procedure information includes: collecting current construction procedure information, which includes at least one or more of the following: the hoisting sequence number of the current main beam segment, the tensioning completion status of the corresponding stay cable, the tower-beam connection status, and the activation status of the temporary support; mapping the construction procedure information to a working condition category identifier representing the construction stage type according to a preset construction stage division rule; and representing the working condition category identifier as a working condition coding vector through multidimensional binary encoding.

[0020] Preferably, the preprocessing of multi-source construction status data to obtain the construction status feature vector includes: acquiring raw monitoring data collected by displacement sensors, strain gauges, cable force gauges, support reaction force gauges, and thermometers deployed on bridge towers, main beams, and stay cables; using low-pass filtering to suppress high-frequency noise in the raw monitoring data; performing coordinate transformation and dimension normalization on the displacement, strain, cable force, and support reaction force data of each monitoring point; and splicing the preprocessed displacement, strain, cable force, support reaction force, and temperature feature quantities in a fixed order according to a preset cross-section and component arrangement sequence to form the construction status feature vector.

[0021] Preferably, the process of constructing the video image feature vector includes: calibrating the internal and external parameters of the cameras deployed in the cross brace area, and establishing a spatial mapping relationship between the camera pixel coordinate system and the coordinate system of the entire cable-stayed bridge; then, performing target area detection on the on-site video image to extract image areas of the cross brace components, cross brace end markers, and preset cross brace cross-sectional structural features; selecting several feature points within the image area, and tracking the pixel coordinates of the feature points in adjacent video frames based on an optical flow tracking algorithm; using the camera calibration parameters and the initial three-dimensional coordinates of the feature points in the reference frame, combined with the perspective-n-point solving algorithm, inverting to obtain the three-dimensional coordinates of the cross brace end and preset cross-section feature points in the entire bridge coordinate system; calculating the difference between the three-dimensional coordinates and the three-dimensional coordinates of the corresponding feature points under the reference construction state to obtain the three-dimensional displacement of the cross brace end and cross section, and calculating the attitude angle parameters of the cross brace end and cross section based on the fitted cross brace axis and local cross-section normal vector; and concatenating the three-dimensional displacement parameters and attitude angle parameters in a preset order to obtain the video image feature vector.

[0022] Preferably, step S3 includes: extracting the three-dimensional displacement of the top and middle sections of the bridge tower, the deflection and planar displacement of the main beam control section, the cable force of the stay cables, and the measured response of the temporary support from the construction state feature vector fused with multimodal data, and associating the measured response with the corresponding calculation nodes and components in the digital twin model; selecting at least one of temperature load distribution, construction additional load, support stiffness, and material elastic modulus as the parameter to be calibrated, and performing forward structural response analysis calculation in the digital twin model with the parameter to be calibrated as the variable to obtain the model calculated response; constructing a parameter inverse analysis model with the difference between the model calculated response and the corresponding measured response as the objective function, and using iterative least squares and Kalman filtering algorithms to update the parameter to be calibrated multiple times; when the displacement deviation and cable force deviation at the key monitoring positions are both less than the preset threshold, a digital twin model matching the current construction state is obtained.

[0023] Preferably, step S4 includes: obtaining the design coordinates and design attitude parameters of both ends of the cross brace in the design full-bridge coordinate system, wherein the design attitude parameters include the cross brace axis direction and the design rotation angles around the longitudinal, transverse and vertical axes of the bridge; based on the calibrated digital twin model of the cable-stayed bridge construction stage, determining the geometric constraint range of the nodes at both ends of the cross brace in the vertical, longitudinal and transverse directions, the minimum clearance requirement between them and adjacent components, and the allowable additional displacement and internal force range of the tower, beam and cable after the cross brace is installed; within the geometric constraint range, performing combined perturbations on the vertical elevation, longitudinal position, transverse offset and attitude rotation angle around the cross brace design position at a preset step size to generate multiple cross brace positioning candidate schemes that satisfy the geometric constraints; for each cross brace positioning candidate scheme, calculating the three-dimensional coordinates of the nodes at both ends of the cross brace in the full-bridge coordinate system, the cross brace axis direction vector and the corresponding attitude angle, and adding the cable force adjustment amount that matches the candidate scheme, splicing them in a preset parameter order to form the cross brace positioning characteristic parameters of the cross brace positioning candidate scheme.

[0024] Preferably, the convolutional neural network prediction model includes:

[0025] The construction state feature convolutional subnetwork is used to perform two-dimensional convolutional encoding on the construction state feature vector and the working condition encoding vector that fuses multiple modes, so as to obtain the first feature map of the current construction state and construction stage.

[0026] A convolutional subnetwork for cross brace positioning features is used to convolutionally encode the cross brace positioning feature parameters of each cross brace positioning candidate scheme, thereby obtaining a second feature map that represents the spatial position and attitude features of different cross brace positioning schemes.

[0027] The digital twin response embedding module performs structural analysis and calculation on each cross brace positioning candidate scheme to obtain the corresponding basic structure response result, and splices the basic structure response result with the first feature map and the second feature map in the channel dimension to form a third feature map that integrates physical information.

[0028] The residual correction and scoring output module is used to perform multi-scale convolution and residual connection operations on the third feature map, learn the deviation relationship between the basic structure response result and the target structure response index, output the structural response index of each cross brace positioning candidate scheme and the comprehensive score value used to indicate the final target cross brace positioning scheme. The comprehensive score value is used to automatically select the final target cross brace positioning scheme from multiple cross brace positioning candidate schemes.

[0029] Preferably, comparing the measured posture with the structural response index during the installation of the cross brace includes:

[0030] During the installation of the cross brace, the measured attitude parameters of the cross brace end in the whole bridge coordinate system are obtained in real time based on the processing of on-site video images. The measured attitude parameters include at least the three-dimensional coordinates of the feature points of the cross brace end and the measured rotation angles around the longitudinal, transverse and vertical axes of the bridge.

[0031] Extract the target attitude parameters corresponding to the final target cross brace positioning scheme from the structural response index. The target attitude parameters include at least the target three-dimensional coordinates and the target rotation angle.

[0032] Under a unified full-bridge coordinate system, the difference vector between the measured three-dimensional coordinates and the target three-dimensional coordinates, as well as the difference between the measured rotation angle and the target rotation angle, are calculated to obtain the displacement deviation and the angle deviation.

[0033] Based on preset displacement deviation threshold and angle deviation threshold, the magnitude of the displacement deviation and angle deviation is compared. When either deviation exceeds the corresponding threshold, it is determined that the attitude deviation exceeds the limit, and an adjustment prompt or alarm signal is triggered. The adjustment prompt includes one or more of the following: reducing the hoisting speed, fine-tuning the position of the cross brace end, and / or pausing the hoisting operation.

[0034] This application also provides an intelligent positioning device for the cross bracing of a cable-stayed bridge, driven by construction status data, comprising:

[0035] Data acquisition module: acquires multi-source construction status data during the construction phase of the cable-stayed bridge, and simultaneously acquires on-site video images of the cross bracing area captured by cameras, and generates a working condition coding vector based on construction procedure information;

[0036] Feature vector generation module: preprocesses multi-source construction status data to obtain construction status feature vectors; performs camera calibration, target area detection, feature point tracking and 3D attitude inversion on on-site video images to obtain the 3D displacement and attitude parameters of the cross brace end and cross section, which constitute video image feature vectors, and aligns them with the construction status feature vectors in time to obtain fused multimodal construction status feature vectors.

[0037] Digital twin model update module: Input the fused multimodal construction state feature vector into the digital twin model of the cable-stayed bridge construction stage, calibrate the model parameters, and obtain a digital twin model that matches the current construction state;

[0038] Candidate solution generation module: Within the geometric constraints given by the digital twin model, multiple candidate solutions for cross bracing positioning are generated based on the cross bracing design position and allowable construction deviations, and cross bracing positioning feature parameters are constructed for each candidate solution.

[0039] Target cross brace positioning final scheme generation module: Input the fused multimodal construction state feature vector, the working condition encoding vector and the cross brace positioning feature parameters into the convolutional neural network prediction model to obtain the target cross brace positioning final scheme and the corresponding structural response index;

[0040] Installation control module: Generates cross brace installation control commands based on the final target cross brace positioning scheme. During the cross brace installation process, continuously collects video images to estimate the attitude of the cross brace end, compares the measured attitude with the structural response index, and outputs adjustment prompts when the deviation exceeds the preset threshold.

[0041] This invention provides a method and device for intelligent positioning of cross braces of cable-stayed bridges driven by construction status data, which can achieve the following beneficial technical effects:

[0042] 1. Compared to existing cross-bracing positioning methods that rely on a limited number of measuring points and static design alignment, this invention constructs a multi-modal sensing system combining "multi-source sensors + video images." Through unified preprocessing of monitoring data such as bridge tower displacement, main beam deflection, stay cable force, temporary support reaction force, and ambient temperature, and combined with three-dimensional attitude inversion of video images of the cross-bracing area, a construction state feature vector is formed under a unified coordinate system and time scale. This method overcomes the limitations of traditional point-based monitoring, which has limited measuring points and struggles to cover local areas of the cross-bracing. Furthermore, it fully utilizes on-site video resources originally intended only for safety monitoring, transforming them into quantifiable geometric measurement information. This allows for the high-precision real-time acquisition of spatial displacement and attitude at the ends of the cross-bracing and key sections, significantly improving the comprehensiveness and refinement of structural state identification during the construction phase, and providing a more reliable input foundation for subsequent intelligent positioning.

[0043] 2. This invention combines a digital twin model of the cable-stayed bridge construction stage calibrated with monitoring data with a convolutional neural network prediction model to achieve rapid and intelligent optimization of the cross brace positioning scheme while meeting physical constraints. The digital twin model provides the geometric constraint range and foundation structure response results, ensuring that all candidate schemes meet structural safety and geometric feasibility requirements. Based on this, the convolutional neural network performs deep feature extraction and residual correction on the fused multimodal construction state feature vector, working condition encoding vector, and cross brace positioning feature parameters. This enables efficient prediction of structural response indicators corresponding to different candidate schemes and outputs a comprehensive score, thereby automatically selecting the final target cross brace positioning scheme from a large number of candidate schemes. Compared with traditional schemes that rely on multiple finite element forward analyses, this invention significantly reduces redundant calculations and improves the real-time performance and intelligence level of cross brace positioning scheme evaluation and decision-making.

[0044] 3. This invention introduces a real-time attitude estimation and deviation comparison mechanism based on video images during the cross brace installation process, achieving closed-loop control between the intelligent positioning results of the cross brace and the actual construction process. By performing target area detection, feature point tracking, and 3D attitude inversion on continuously acquired on-site video images during the hoisting process, the measured 3D coordinates and rotation parameters of the cross brace end in the whole bridge coordinate system can be obtained in real time. These are then compared with the target attitude parameters corresponding to the final positioning scheme of the target cross brace, calculating the displacement deviation and angle deviation. When any deviation exceeds a preset threshold, an adjustment prompt of deceleration, fine-tuning, or pausing the hoisting is automatically issued. This invention not only corrects deviations in a timely manner during construction, avoiding structural stress concentration and construction safety risks caused by cross brace misalignment, but also reduces reliance on the experience judgment of construction personnel, reduces the probability of rework and the workload of measurement and verification, and significantly improves the accuracy and safety of cross brace installation during the construction phase of cable-stayed bridges. Attached Figure Description

[0045] 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 some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart of the steps of an intelligent positioning method for cross bracing of a cable-stayed bridge driven by construction status data according to the present invention.

[0047] Figure 2 This is a schematic diagram of an intelligent positioning device for the cross bracing of a cable-stayed bridge, driven by construction status data, according to the present invention. Detailed Implementation

[0048] 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.

[0049] Example 1:

[0050] In view of the aforementioned problems mentioned in the prior art, and in order to solve the above technical problems, as shown in the appendix. Figure 1 As shown: This application provides a data-driven intelligent positioning method for cross bracing of cable-stayed bridges based on construction status, including the following steps:

[0051] S1: Acquire multi-source construction status data during the construction phase of the cable-stayed bridge, and simultaneously acquire on-site video images of the cross bracing area collected by cameras, and generate a working condition coding vector based on the construction procedure information; In some embodiments, taking a double-tower double-cable-stayed bridge with a main span of 580m as an example, during the construction phase before the main beam closure, the construction status data of the cable-stayed bridge and the video images of the cross bracing area are acquired, and a working condition coding vector is generated based on the construction procedure information for subsequent intelligent positioning of the cross bracing.

[0052] First, during the construction phase of the cable-stayed bridge, three-dimensional displacement gauges were installed at key sections at the top, middle, and bottom of the two towers along their height direction. Deflection gauges and horizontal displacement gauges were installed at control sections in the middle and side spans of the main girder. Cable force gauges were installed on each stay cable, and support reaction force gauges were installed at the connection points between the main girder and temporary supports. Thermometers were also installed on the outer surface of the tower columns and on the web of the main girder. All types of sensors were connected to the on-site data acquisition terminal via wired or wireless means, with a sampling period set to 10 seconds. High-precision timestamps were uniformly added to generate multi-source construction status data, including tower displacement, main girder displacement, stay cable force, temporary support reaction force, and ambient temperature.

[0053] Secondly, one pan-tilt-zoom (PTZ) network camera is installed on each side of the bridge tower where the cross bracing is to be installed, and two fixed-focus industrial cameras are installed near the control section of the main beam. The camera resolution is set to 1920×1080, and the frame rate is set to 25 frames per second. Each camera transmits real-time video images of the cross bracing area to the video processing server via the construction site's local area network, and also adds a unified timestamp synchronized with the sensor system to ensure that subsequent multi-source data can be aligned in time.

[0054] Furthermore, the construction management system records the current construction process information, including: the hoisting sequence number of the main beam segment (e.g., segment n), the tensioning completion status of the stay cables on both sides of the corresponding segment (not tensioned, initially tensioned, or finally tensioned), the current tower-beam connection status (not connected, temporarily connected, or permanently connected), and the activation status of the temporary supports within the span where the cross braces are located (not activated, partially activated, or fully activated). This construction process information is automatically collected by the construction scheduling system and the tensioning control system, and can also be manually confirmed and entered by the supervising engineer on the construction management platform upon completion of the process.

[0055] In this embodiment, according to a pre-defined construction stage division rule, the "main beam segment hoisting sequence," "stay cable tensioning completion status," "tower-beam connection status," and "temporary support activation status" are mapped to multiple binary or multi-valued work condition identifiers. For example, the main beam segment number uses one-hot encoding to represent different segments, the stay cable tensioning status uses three-digit binary encoding to represent the three states of "not tensioned, initial tensioned, and final tensioned," and the tower-beam connection status and temporary support activation status use two-digit or three-digit binary encoding to represent different connection and activation conditions. Subsequently, the various work condition identifiers are concatenated in a fixed order along the feature dimension to form a work condition encoding vector of length M, which is used to characterize the work condition type of the current cross brace construction stage and serves as one of the inputs to the subsequent convolutional neural network prediction model.

[0056] Through the above steps, this embodiment, while ensuring time synchronization, completed the acquisition of multi-source construction status data and video images of the cross bracing area during the construction phase of the cable-stayed bridge. Based on the construction procedure information, a working condition coding vector with clear physical meaning was constructed, laying a data foundation for subsequent multimodal feature fusion and intelligent positioning of the cross bracing.

[0057] S2: Preprocess the multi-source construction status data to obtain the construction status feature vector; perform camera calibration, target area detection, feature point tracking and three-dimensional attitude inversion on the on-site video images to obtain the three-dimensional displacement and attitude parameters of the cross brace end and cross section, which constitute the video image feature vector, and align it with the construction status feature vector in time to obtain the fused multimodal construction status feature vector.

[0058] In some embodiments, based on the multi-source construction status data and on-site video images of the cross bracing area obtained in Embodiment 1, the sensor data and video data are preprocessed respectively to construct construction status feature vectors and video image feature vectors, and multimodal feature fusion is achieved at a unified time step. Specifically, the steps include the following.

[0059] The first step involves preprocessing the multi-source construction status data. The data acquisition terminal sequentially receives monitoring data such as bridge tower displacement, main girder displacement, cable tension, temporary support reaction force, and temperature at a sampling period of 10 seconds. For each type of monitoring data, a low-pass filter is used to filter the time series. The filter cutoff frequency is determined based on the bridge's natural frequency and the characteristics of construction load changes to suppress sensor noise and occasional impact interference. For individual missing or obviously abnormal data points, a combination of time interpolation and threshold rejection is used for repair or removal. Subsequently, using the designed full-bridge coordinate system as a reference, the displacement components of each monitoring point are transformed from the sensor's local coordinates to the X (longitudinal), Y (transverse), and Z (vertical) components in the full-bridge coordinate system. Furthermore, physical quantities with different dimensions, such as displacement, cable tension, reaction force, and temperature, are normalized to ensure that each feature quantity is within a similar numerical range, facilitating subsequent neural network training and inference. Finally, following a fixed order—bridge towers from top to bottom, main beams from one end to the other, stay cables in numerical order, and temporary supports in arrangement order—the normalized feature quantities of each monitoring point are sequentially spliced ​​together along the feature dimension to form a construction state feature vector of length N, which is used to characterize the construction state of the entire bridge at the current moment.

[0060] Second, camera calibration, target region detection, feature point tracking, and 3D pose inversion are performed on the on-site video images. Using a calibration board, multi-pose shots are taken within the field of view of each camera. The Zhang Zhengyou calibration method is used to solve for the camera's internal parameters (focal length, principal point coordinates, distortion coefficients) and external parameters (rotation matrix and translation vector), establishing a spatial mapping relationship between the camera pixel coordinate system and the full bridge coordinate system. Based on the calibration, for video frames of the cross brace area, a pre-trained target detection network (e.g., a detection model based on a convolutional neural network) is invoked to automatically identify the outline of the cross brace components, preset markers at the ends of the cross brace, and geometric feature regions on the tower columns or main beams near the cross brace cross section. Within each detected target region, the Shi-Tomasi corner detection algorithm is used to select several feature points, and the Lucas-Kanade optical flow algorithm is used to track these feature points in adjacent video frames, obtaining the pixel coordinate trajectory of the feature points over time.

[0061] After acquiring the pixel trajectories of feature points, using the 3D coordinates of the corresponding feature points under the construction baseline as a reference, the 3D coordinates of the cross brace end and preset cross-section feature points in the current video frame in the full-bridge coordinate system are obtained by combining the perspective-n-point (PnP) solution algorithm with camera calibration parameters. Subtracting the 3D coordinates of the corresponding feature points under the baseline construction state from the current 3D coordinates yields the 3D displacement components of the cross brace end and cross-section. Simultaneously, the cross brace axis direction vector is fitted using the feature points on both sides of the cross brace end, and the local cross-section normal vector is fitted using the feature points around the cross-section. Based on this, the attitude rotation angles of the cross brace end and cross-section relative to the full-bridge coordinate system around the longitudinal, lateral, and vertical axes are calculated. These 3D displacement components and attitude angle parameters are arranged and spliced ​​in a preset order to form a video image feature vector of length K, maintaining a consistent feature dimension definition with the construction state feature vector.

[0062] Thirdly, to achieve multimodal feature fusion, the construction status feature vector and the video image feature vector are time-aligned. Using a 10-second sampling period from the construction monitoring system as a unified time step and a reference time as a starting point, for each time step t, the construction status feature vector at the corresponding moment is directly read from the sensor data. For video data, one or more video frames closest to time t are selected based on their timestamps, and the corresponding video image feature vector is calculated using nearest-neighbor interpolation or averaging strategies. For a small amount of time deviation caused by network latency, the maximum allowable time difference is limited and frames exceeding the limit are removed to ensure that the time error of feature alignment does not exceed a preset tolerance.

[0063] Finally, at each time step t, the aligned construction state feature vector and the video image feature vector are concatenated along the feature dimension or fused using a linear transformation to form a fused multimodal construction state feature vector of length N+K or after dimensionality reduction. This fused feature vector serves as the common input for subsequent digital twin model calibration and convolutional neural network prediction model. This embodiment achieves unified preprocessing and feature fusion of multi-source construction state data and on-site video images through the above steps, providing high spatiotemporal resolution multimodal input data for cross-bracing intelligent positioning.

[0064] S3: Input the fused multimodal construction state feature vector into the digital twin model of the cable-stayed bridge construction stage, calibrate the model parameters, and obtain a digital twin model that matches the current construction state;

[0065] In some embodiments, the parameters of the digital twin model of the cable-stayed bridge construction stage are calibrated based on the fused multimodal construction state feature vector obtained in S2, so that the digital twin model can accurately reflect the actual stress and deformation under the current construction state. The digital twin model adopts the same structural topology and construction layout as the actual cable-stayed bridge, including component units such as bridge towers, main beams, stay cables, main cables (if any), supports, and temporary supports. The initial parameters are taken from the design drawings and construction plan, including section stiffness, material elastic modulus, support stiffness, initial cable force of stay cables, temperature field distribution, and additional loads during the construction stage.

[0066] For any time step t, the corresponding fused multimodal construction state feature vector includes the three-dimensional displacement of the key section of the bridge tower, the deflection and planar displacement of the main beam control section, the cable force of the stay cables, the reaction force of the temporary supports, the characteristic quantities of the ambient temperature field, and the three-dimensional displacement and attitude parameters of the cross brace ends and sections obtained from video image inversion. First, the three-dimensional displacement of the top and middle of the bridge tower, the vertical deflection and horizontal displacement of the control sections of the middle and side spans of the main beam, the measured cable force of each cableway, and the reaction force values ​​of the temporary support columns are extracted from this feature vector according to a preset index to form the measured response vector y_meas, and each component in it is mapped and associated one-to-one with the corresponding finite element nodes or elements in the digital twin model.

[0067] Secondly, model parameters that significantly influence the structural response during the construction phase are selected as the parameter vector p to be calibrated. These include at least one or more of the following: the reduction factor of the equivalent elastic modulus of the bridge tower and main girder components, the equivalent stiffness coefficient of the supports and temporary bracing, the correction amount of the initial cable force of the stay cables, the correction amount of the temperature load distribution, and the equivalent partial factor of the additional construction load. Using the construction sequence information of the current construction phase as constraints, the corresponding self-weight, construction load, and known construction conditions are applied to the digital twin model. The parameter p to be calibrated is used as a variable for forward structural analysis calculations, obtaining the calculated response quantities such as bridge tower displacement, main girder displacement, cable force, and support reaction force, forming the calculated response vector y_calc(p). Based on this, a parametric inverse analysis model is constructed with the difference between the measured response y_meas and the calculated response y_calc(p) as the objective function. Specifically, with... As the minimization objective, w_i represents the weighting coefficients of each measuring point or response component, used to reflect the importance of key locations (such as the tower top or the main span control section). The parameter vector p is iteratively updated multiple times using iterative least squares, the Levenberg-Marquardt algorithm, or the extended Kalman filter algorithm: in each iteration, the structural response is calculated based on the current parameters, the sensitivity of the objective function to the parameters is linearized and approximated, and the parameter vector p^(k+1) = p^(k) + Δp is updated until the objective function converges.

[0068] During the iteration process, this embodiment preferably sets multiple convergence criteria: First, the absolute value of the displacement deviation at key monitoring locations does not exceed a preset displacement threshold (e.g., L / 10000 or 5mm) and the relative error of the cable force deviation does not exceed a preset percentage (e.g., 5%); second, the parameter increment norm ||Δp|| of two consecutive iterations is lower than a preset convergence threshold; third, the decrease in the overall objective function J(p) is less than a preset proportion. When any one or more convergence conditions are met, the iteration stops, and the parameter vector p* and its corresponding structural analysis model are determined as a digital twin model matching the current construction state.

[0069] Through the above steps, this embodiment utilizes the multi-type measured response information contained in the fusion multimodal construction state feature vector to jointly calibrate key parameters such as temperature load distribution, support and temporary support stiffness, initial cable force of the cable stays, and additional construction loads of the digital twin model during the construction stage of the cable-stayed bridge. This enables the calculated displacement, cable force, and reaction force of the digital twin model under the current construction stage to be highly consistent with the monitoring data, thereby providing a reliable structural analysis basis for determining the geometric constraint range of the cross braces and assessing the physical feasibility of the cross brace positioning candidate schemes.

[0070] S4: Within the geometric constraints given by the digital twin model, generate multiple candidate schemes for cross bracing positioning based on the cross bracing design position and allowable construction deviations, and construct cross bracing positioning feature parameters for each candidate scheme.

[0071] In this embodiment, based on the digital twin model of the cable-stayed bridge construction stage obtained by S3 calibration, the geometric constraint range of the cross brace is analyzed, and multiple cross brace positioning candidate schemes are generated around the cross brace design position within the constraint range. At the same time, cross brace positioning feature parameters are constructed for each candidate scheme, which are used as inputs for the subsequent convolutional neural network prediction model.

[0072] First, based on the design drawings, obtain the design position and orientation parameters of the target cross brace in the full bridge coordinate system. Let the design three-dimensional coordinates of the two ends of the cross brace in the full bridge coordinate system (X for longitudinal direction, Y for transverse direction, Z for vertical direction) be as follows: , The design attitude parameters include the cross brace axis direction vector u0 and the design rotation angles around the X, Y, and Z axes. Meanwhile, based on the construction surveying specifications and design requirements, the allowable deviation ranges for the cross bracing in terms of vertical elevation, longitudinal position, transverse offset, and attitude angle are determined. For example, vertical deviation ±20mm, longitudinal deviation ±15mm, transverse deviation ±10mm, attitude angle deviation ±0.3°, etc.

[0073] Secondly, using the digital twin model obtained from S3 that matches the current construction status, geometric and stress analyses are performed on the tower columns or main beam connection areas where the cross braces are located. By post-processing the displacement field of components near the connection nodes, the feasible geometric ranges of the nodes at both ends of the cross braces in the vertical, longitudinal, and transverse directions are determined to avoid collisions or interference with adjacent components. At the same time, based on the digital twin model, the additional displacements and internal force changes caused by the application of the cross braces at different locations on the tower, beam, and cable are calculated, and constraints such as "the tower top displacement after the cross brace installation does not exceed a certain limit, the cable force change does not exceed a certain percentage, and the internal force of the main beam control section does not exceed the allowable increment" are given, thus forming a geometric-mechanical comprehensive constraint domain for the cross brace positioning.

[0074] Within the aforementioned constraint domain, this embodiment employs a combination of regular mesh perturbation and key area densification to generate multiple candidate cross brace positioning schemes. Centered on the design location, vertical positioning is performed in 5mm increments. , The end elevations are enumerated within the range, and discrete values ​​are obtained in the longitudinal and transverse directions within the allowable deviation range at 5mm increments. The attitude angles are based on the design rotation angle, and perturbation combinations are performed within a set range at 0.1° increments. For each perturbation combination, the corresponding end coordinates and attitude parameters are constructed as a candidate scheme for cross brace positioning, and a digital twin model is used for rapid verification. If the scheme causes additional displacements or internal forces at the tower top, main beam control section, or stay cables to exceed the allowable range, the scheme is marked as infeasible and eliminated, retaining the set of candidate schemes that satisfy both geometric and mechanical constraints.

[0075] For each retained candidate cross brace positioning scheme, this embodiment constructs a corresponding cross brace positioning feature parameter vector. Specifically, the three-dimensional coordinates of the two ends of the candidate scheme in the full bridge coordinate system are used as the first part of the features; the unit vector along the cross brace axis and the attitude rotation angles around the X, Y, and Z axes are used as the second part of the features; considering the coupling relationship between the cross brace and cable force adjustment, the cable force adjustment amount required to maintain the design alignment or control the displacement of key sections under the candidate scheme can be further calculated based on the digital twin model, and the cable force increments of each cableway are spliced ​​in numerical order as the third part of the features. Finally, the above three parts of features are concatenated in a preset order to form a cross brace positioning feature parameter vector of length P, and a unique index ID is assigned to each candidate scheme.

[0076] Through the above steps, this embodiment generates multiple physically feasible candidate schemes for cross brace positioning around the cross brace design location within the geometric and force constraints given by the calibrated digital twin model. For each candidate scheme, cross brace positioning feature parameters with clear geometric and mechanical significance are constructed, providing structured input data for the subsequent convolutional neural network prediction model to quickly evaluate the structural response of different candidate schemes and intelligently select the final target cross brace positioning scheme.

[0077] S5: Input the construction state feature vector of the fused multimodal structure, the working condition encoding vector and the cross brace positioning feature parameters into the convolutional neural network prediction model to obtain the final target cross brace positioning scheme and the corresponding structural response index.

[0078] In some embodiments, based on the fused multimodal construction state feature vector obtained in S2, the working condition encoding vector obtained in S1, and the cross brace positioning feature parameters constructed in S4, a convolutional neural network prediction model is used to predict and comprehensively evaluate the structural response of multiple cross brace positioning candidate schemes, thereby automatically generating the final target cross brace positioning scheme and the corresponding structural response index. The convolutional neural network prediction model is pre-trained offline and inferred with fixed parameters during the online construction phase.

[0079] The convolutional neural network prediction model adopts a multi-branch fusion structure, including a construction state feature convolutional branch, a cross brace positioning feature convolutional branch, and a fusion and output branch. The construction state feature convolutional branch takes the vector obtained by concatenating the multimodal construction state feature vector and the working condition encoding vector as input. This vector is rearranged into a two-dimensional feature matrix according to the time or component layout dimension. After processing through two one-dimensional or two-dimensional convolutional layers, a batch normalization layer, and a nonlinear activation function, a first feature map representing the current overall structural state and construction stage characteristics is obtained. The cross brace positioning feature convolutional branch takes the cross brace positioning feature parameters of each cross brace positioning candidate scheme as input. Through one or two convolutional layers, it extracts the local features of the candidate schemes in terms of spatial position, attitude, and cable force adjustment, forming a second feature map.

[0080] To enhance the physical interpretability of the prediction model and synergize with the digital twin model in Example 3, this example incorporates a digital twin response embedding module within the network. During the training phase, for multiple combinations of construction states and cross brace positioning, the calibrated digital twin model calculates the foundation structure response results, including tower displacement, deflection at key sections of the main beam, changes in cable tension, and support reactions. These results are used as prior physical information and concatenated with the first and second feature maps along the channel dimension to form a third feature map that integrates physical information. Subsequently, the third feature map is fed into a multi-scale residual convolution block. Convolutional kernels with different receptive fields extract cross-component and cross-section correlation features. Residual connections are used to learn the deviation relationship between the digital twin foundation response and the target structural response indicators, effectively compensating for uncertainties and nonlinear effects that are difficult to cover by relying solely on finite element analysis.

[0081] At the network output, this embodiment maps the high-dimensional features output by the residual convolutional block into two types of outputs through a fully connected layer: one type is the structural response index vector corresponding to each cross brace positioning candidate scheme, which includes at least the tower top displacement, the deflection of the main beam control section, the change in the cable force of key stay cables, and the stress level of components near the cross brace; the other type is a score value used to comprehensively evaluate the merits of each candidate scheme. The score value is automatically learned by the network during the training phase based on a preset comprehensive evaluation function (e.g., the smaller the displacement deviation, the smoother the cable force change, and the larger the safety margin, the higher the score). During the training process, the calculation results of the digital twin model or historical measured data are used as supervision labels for the structural response index. By minimizing the mean square error between the predicted value and the label value, and introducing a ranking loss term that reflects "high-scoring schemes are better", the network parameters are jointly optimized, so that the network can accurately predict the structural response and output a comprehensive score that conforms to engineering experience.

[0082] During the online construction phase, for a given time t, the corresponding fused multimodal construction state feature vector and working condition encoding vector remain unchanged. The cross-bracing positioning feature parameters of the multiple cross-bracing positioning candidate schemes generated and constrained in Example 4 are input one by one or in batches into the convolutional neural network prediction model to obtain the structural response index and comprehensive score value corresponding to each candidate scheme. Subsequently, among the candidate schemes that satisfy both safety and construction constraints, they are sorted from high to low according to their comprehensive score values. The candidate scheme with the highest comprehensive score value is selected as the final target cross-bracing positioning scheme, and the corresponding structural response index is recorded as the target structural response index.

[0083] Through the above steps, this embodiment utilizes a convolutional neural network prediction model to deeply fuse multimodal construction state feature vectors, working condition coding vectors, and cross brace positioning feature parameters, forming a nonlinear mapping from "current construction state + cross brace positioning scheme" to "structural response index + comprehensive score". While ensuring the physical constraints provided by the digital twin model, it achieves rapid evaluation and automatic selection of a large number of cross brace positioning candidate schemes, significantly improving the efficiency and intelligence of cross brace positioning scheme generation and decision-making, and providing a reliable basis for the generation of subsequent cross brace installation control commands.

[0084] In a preferred embodiment, the convolutional neural network prediction model adopts a multi-branch fusion structure, specifically including four parts: a construction state feature convolutional subnetwork, a cross brace positioning feature convolutional subnetwork, a digital twin response embedding module, and a residual correction and scoring output module. This is used to realize the structural response prediction and comprehensive evaluation of different cross brace positioning candidate schemes based on the physical constraints of the digital twin.

[0085] In this embodiment, for a given time t and any candidate cross brace positioning scheme, the network input includes three parts: a multimodal construction state feature vector fstate with length Ns; a work condition encoding vector fwork corresponding to that time with length Nw; and a cross brace positioning feature parameter vector floc with length Nl for the candidate cross brace positioning scheme. First, fstate and fwork are concatenated along their feature dimensions to obtain a joint construction state-work condition vector of length Ns+Nw. Then, based on the monitoring point arrangement order and work condition encoding layout rules, this vector is rearranged into a two-dimensional feature map of size Cs×Ls (e.g., Cs = 32 channels, Ls = (Ns+Nw) / 32 position units), which serves as the input to the construction state feature convolutional subnet. Similarly, the cross brace positioning feature parameter vector floc is rearranged into a two-dimensional feature map of size Cl×Ll (e.g., Cl = 16, Ll = Nl / 16), which serves as the input to the cross brace positioning feature convolutional subnet.

[0086] The construction state feature convolutional subnetwork is mainly used to extract the current overall structural state and construction stage features. Its structure includes a two-stage convolution-normalization-nonlinear module and a global pooling layer: The first-stage convolutional module uses several 3×3×3 two-dimensional convolutional kernels to perform convolution operations on the input construction state feature map, with an output channel count of, for example, 64. This is followed by a batch normalization layer and a ReLU activation layer to enhance feature representation and suppress overfitting. The second-stage convolutional module continues convolution based on the first-stage output using 1×3×3 or 3×3×3 convolutional kernels, maintaining or expanding the output channel count to 64~128, also incorporating batch normalization and ReLU activation. The global pooling layer performs global average pooling or max pooling on the second-stage convolutional output along its length dimension, compressing the two-dimensional feature map into a one-dimensional vector hstate, used to represent the current global construction state and working condition features.

[0087] The cross-bracing positioning feature convolutional subnetwork is used to mine the intrinsic relationship between the coordinates, attitude angles, and associated cable force adjustments at both ends of the cross-bracing. Its structure also employs a two-stage convolutional module: The first-stage convolutional module applies several 1×3×3 two-dimensional convolutional kernels (equivalent to one-dimensional convolution along the feature dimension) to the cross-bracing positioning feature map, with an output channel count of, for example, 32. Combined with batch normalization and ReLU activation, it is used to extract local geometric and cable force adjustment features at the ends of the cross-bracing. The second-stage convolutional module continues to use 1×3×3 or 1×5×5 convolutional kernels to strengthen the cross-positional correlation between different parameters, with an output channel count of, for example, 32~64. Subsequently, a global pooling layer is used to aggregate along the length dimension, resulting in a one-dimensional vector hloc, used to represent the overall spatial-mechanical features of a candidate solution.

[0088] A digital twin response embedding module is set up in this embodiment to introduce the basic structural response results of the calibrated digital twin model into the network. During the training phase, for each set of construction states and cross brace positioning candidate schemes, the calibrated digital twin model in Embodiment 3 is used first to calculate the basic structural response, such as tower top displacement, main beam key section deflection, cable force increment, and support reaction change, forming the basic structural response vector ydtydt. The digital twin response embedding module first projects ydtydt onto the same or similar feature dimensions as hstate and hloc through one or two fully connected layers to obtain the embedding vector hdthdt, and can be used with ReLU or LeakyReLU activation to enhance the nonlinear expression. Subsequently, hstate, hloc, and hdt are concatenated in the feature dimension to obtain the fusion vector hconcat. In the implementation that requires the construction of a two-dimensional feature map, hstate, hloc, and hdt can also be stacked in the channel dimension to form a third feature map for subsequent multi-scale convolution processing.

[0089] The residual correction and multi-scale convolution module is used to learn a "correction amount" based on the digital twin's base response and output a comprehensive score. Internally, it includes several residual convolutional blocks, each consisting of two convolutional layers and a shortened branch. Within each residual block, two sets of convolutional kernels of different scales (e.g., 1×3 1×3, 1×5 1×5 or 3×3 3×3, 3×5 3×5) are used to convolve the input feature map, combined with batch normalization and ReLU activation to capture structural response patterns at different spatial scales. The outputs of the two convolutional layers are concatenated or added along the channel dimension and then added element-wise to the input of the residual block to form residual connections. This allows the network to focus on learning the "biased portion above the digital twin's base response," improving the network's stability and physical plausibility.

[0090] Optionally, channel attention or self-attention modules can be inserted between residual blocks to assign higher weights to important monitoring locations and key cross-bracing parameters. After several residual blocks, the output feature map is globally pooled again to obtain a one-dimensional latent vector zresp for regressing the structural response index and a one-dimensional latent vector zscore for comprehensive scoring (which can be the same vector mapped from different output heads).

[0091] Dual Output Heads: Structural Response Regression and Comprehensive Scoring. This embodiment sets up two output heads at the network output end: Structural Response Regression Head: zresp is input to one or two fully connected layers. The output dimensions are consistent with the target structural response index dimensions, such as tower top X / Y / Z displacement, deflection of the control sections of the main beam's mid-span and side spans, increase in cable force of key stay cables, and maximum stress of components near the cross braces. During training, a digital twin model or historical measured data is used as a label, and mean squared error loss is employed for supervised learning of this output head to ensure the predicted structural response approximates the calculation results of the physical model as closely as possible. Comprehensive Scoring Output Head: zscore is input to another fully connected layer, outputting a single score value or a small number of scoring dimensions. The score value is used to measure the comprehensive merits of candidate schemes in terms of displacement deviation, cable force variation, safety margin, and subsequent construction convenience. During training, based on a pre-constructed comprehensive evaluation function or manually labeled merits, ranking loss or cross-entropy loss can be introduced to make schemes with high scores closer to the target alignment and more reasonable internal forces in terms of structural response.

[0092] In the online inference and target scheme selection phase, the construction state feature vector and working condition encoding vector remain unchanged at a certain time t. The cross-bracing positioning feature parameters of the multiple cross-bracing positioning candidate schemes retained in Example 4 are input into the aforementioned convolutional neural network prediction model to obtain the structural response index and comprehensive score value of each candidate scheme in batches. Subsequently, among the candidate schemes that meet the safety and construction constraints, the candidate scheme with the largest comprehensive score value is selected as the final target cross-bracing positioning scheme, and its corresponding structural response index is used as the target structural response benchmark for subsequent attitude comparison and early warning judgment.

[0093] Through the above-mentioned convolutional neural network structure design, this embodiment realizes rapid batch prediction and intelligent optimization of a large number of cross brace positioning candidate schemes based on physical constraints calibrated by the digital twin model. It organically integrates multi-source monitoring data, construction condition information, cross brace geometric-mechanical parameters and digital twin basic response, taking into account both the physical interpretability of the prediction results and computational efficiency, and significantly improving the real-time performance and reliability of intelligent positioning of cross braces in cable-stayed bridges.

[0094] S6: Generate cross brace installation control commands based on the final target cross brace positioning scheme. During the cross brace installation process, continuously collect video images to estimate the attitude of the cross brace end. Compare the measured attitude with the structural response index. When the deviation exceeds the preset threshold, output an adjustment prompt.

[0095] In this embodiment, based on the final target cross brace positioning scheme and corresponding structural response index obtained in S5, cross brace installation control commands are generated. During the cross brace hoisting process, video images are used to estimate the attitude of the cross brace end in real time. The measured attitude is compared with the target attitude. When the deviation exceeds a preset threshold, an adjustment prompt is automatically output to realize closed-loop control of the cross brace installation process.

[0096] First, when the convolutional neural network prediction model outputs the final positioning scheme for the target cross brace, the system reads the target's three-dimensional coordinates at both ends of the cross brace in the full bridge coordinate system, the cross brace axis direction vector, and the target attitude angle parameters around the longitudinal, transverse, and vertical axes of the bridge from the cross brace positioning feature parameters corresponding to the scheme. It also extracts information such as the tower top displacement, the deflection of the key section of the main beam, and the cable force of the key cableway under the target state from the structural response indicators. The cross brace installation control module generates specific installation control instructions based on the target's three-dimensional coordinates and attitude angles, including the target spatial position of the lifting points at both ends of the cross brace, the allowable lifting path deviation, the temporary support jacking amount, and the corresponding cable force fine-tuning suggestions. These control instructions are then issued to the on-site lifting command system and monitoring terminal in the form of a digital construction task sheet.

[0097] During the hoisting of the cross brace, cameras deployed in the cross brace area continuously capture video images, and the video stream is transmitted to the video processing server in real time. Following the method in Example 2, the video processing module performs target region detection and feature point tracking on each frame of video, inverting the current three-dimensional coordinates and attitude angles of the marker points at both ends of the cross brace in the full bridge coordinate system to form a measured attitude parameter vector. To ensure real-time performance, this embodiment preferably updates the attitude with a time step of 1 second, and performs a moving average of the attitude results for several consecutive frames to suppress measurement noise caused by instantaneous jitter.

[0098] The cross brace installation control module, under a unified full-bridge coordinate system, performs difference calculations between the measured three-dimensional coordinates at the current moment and the target three-dimensional coordinates corresponding to the final positioning scheme of the target cross brace, obtaining the displacement deviations of the two end nodes in the vertical, longitudinal, and transverse directions. Simultaneously, it calculates the difference between the current measured attitude angle and the target attitude angle, obtaining the angular deviation around the three axes. Based on the construction process, the system is divided into three stages: rough hoisting, positioning, and fine adjustment, with different displacement and angular deviation thresholds set for each. For example, in the rough hoisting stage, the allowable vertical deviation is no more than 20mm and the allowable angular deviation is no more than 0.5°; in the positioning stage, the allowable deviations are tightened to 10mm and 0.3°; and in the fine adjustment stage, they are further tightened to 5mm and 0.2°.

[0099] When the displacement or angular deviation in any direction exceeds the preset threshold corresponding to the current stage, the system automatically triggers an adjustment prompt. For minor deviations (such as slightly exceeding the threshold of the rough lifting stage), the system marks the direction of the deviation and the suggested adjustment amount in yellow on the lifting command terminal and monitoring screen, and prompts the crane operator to reduce the lifting speed or fine-tune the height of a certain end lifting point through voice broadcast. For deviations approaching or exceeding the threshold of the positioning / fine-tuning stage, the system issues a red alarm prompt, suggesting that the lifting be temporarily suspended, and that the corresponding end position be adjusted first or that the cable force be adjusted to restore the target posture. When the deviation seriously exceeds the threshold and shows a continuous increasing trend, the system can issue a control command to the lifting equipment to suspend the lifting according to the preset strategy, and push high-level alarm information to the construction supervisor.

[0100] Through the above implementation method, this embodiment uses the target attitude parameters corresponding to the final positioning scheme of the target cross brace in the structural response index as a benchmark. The measured attitude obtained from video image inversion is compared with the target attitude in real time, and intelligent adjustment prompts or control commands are output when limits are exceeded, realizing dynamic closed-loop control of the cross brace positioning result and the hoisting process. This method not only improves the alignment control accuracy of the cross brace installation process and reduces local stress concentration and construction safety risks caused by cross brace misalignment, but also reduces the repetitive process of traditional "measurement-stop-adjustment," which is beneficial to improving the overall efficiency and quality controllability of the cable-stayed bridge construction phase.

[0101] Preferably, the multi-source construction status data includes bridge tower displacement, main beam displacement, cable tension, strain, temporary support reaction force, and environmental parameters.

[0102] Preferably, the step of generating a working condition coding vector based on construction procedure information includes: collecting current construction procedure information, which includes at least one or more of the following: the hoisting sequence number of the current main beam segment, the tensioning completion status of the corresponding stay cable, the tower-beam connection status, and the activation status of the temporary support; mapping the construction procedure information to a working condition category identifier representing the construction stage type according to a preset construction stage division rule; and representing the working condition category identifier as a working condition coding vector through multidimensional binary encoding.

[0103] Preferably, the preprocessing of multi-source construction status data to obtain the construction status feature vector includes: acquiring raw monitoring data collected by displacement sensors, strain gauges, cable force gauges, support reaction force gauges, and thermometers deployed on bridge towers, main beams, and stay cables; using low-pass filtering to suppress high-frequency noise in the raw monitoring data; performing coordinate transformation and dimension normalization on the displacement, strain, cable force, and support reaction force data of each monitoring point; and splicing the preprocessed displacement, strain, cable force, support reaction force, and temperature feature quantities in a fixed order according to a preset cross-section and component arrangement sequence to form the construction status feature vector.

[0104] Preferably, the process of constructing the video image feature vector includes: calibrating the internal and external parameters of the cameras deployed in the cross brace area, and establishing a spatial mapping relationship between the camera pixel coordinate system and the coordinate system of the entire cable-stayed bridge; then, performing target area detection on the on-site video image to extract image areas of the cross brace components, cross brace end markers, and preset cross brace cross-sectional structural features; selecting several feature points within the image area, and tracking the pixel coordinates of the feature points in adjacent video frames based on an optical flow tracking algorithm; using the camera calibration parameters and the initial three-dimensional coordinates of the feature points in the reference frame, combined with the perspective-n-point solving algorithm, inverting to obtain the three-dimensional coordinates of the cross brace end and preset cross-section feature points in the entire bridge coordinate system; calculating the difference between the three-dimensional coordinates and the three-dimensional coordinates of the corresponding feature points under the reference construction state to obtain the three-dimensional displacement of the cross brace end and cross section, and calculating the attitude angle parameters of the cross brace end and cross section based on the fitted cross brace axis and local cross-section normal vector; and concatenating the three-dimensional displacement parameters and attitude angle parameters in a preset order to obtain the video image feature vector.

[0105] Preferably, step S3 includes: extracting the three-dimensional displacement of the top and middle sections of the bridge tower, the deflection and planar displacement of the main beam control section, the cable force of the stay cables, and the measured response of the temporary support from the construction state feature vector fused with multimodal data, and associating the measured response with the corresponding calculation nodes and components in the digital twin model; selecting at least one of temperature load distribution, construction additional load, support stiffness, and material elastic modulus as the parameter to be calibrated, and performing forward structural response analysis calculation in the digital twin model with the parameter to be calibrated as the variable to obtain the model calculated response; constructing a parameter inverse analysis model with the difference between the model calculated response and the corresponding measured response as the objective function, and using iterative least squares and Kalman filtering algorithms to update the parameter to be calibrated multiple times; when the displacement deviation and cable force deviation at the key monitoring positions are both less than the preset threshold, a digital twin model matching the current construction state is obtained.

[0106] Preferably, step S4 includes: obtaining the design coordinates and design attitude parameters of both ends of the cross brace in the design full-bridge coordinate system, wherein the design attitude parameters include the cross brace axis direction and the design rotation angles around the longitudinal, transverse and vertical axes of the bridge; based on the calibrated digital twin model of the cable-stayed bridge construction stage, determining the geometric constraint range of the nodes at both ends of the cross brace in the vertical, longitudinal and transverse directions, the minimum clearance requirement between them and adjacent components, and the allowable additional displacement and internal force range of the tower, beam and cable after the cross brace is installed; within the geometric constraint range, performing combined perturbations on the vertical elevation, longitudinal position, transverse offset and attitude rotation angle around the cross brace design position at a preset step size to generate multiple cross brace positioning candidate schemes that satisfy the geometric constraints; for each cross brace positioning candidate scheme, calculating the three-dimensional coordinates of the nodes at both ends of the cross brace in the full-bridge coordinate system, the cross brace axis direction vector and the corresponding attitude angle, and adding the cable force adjustment amount that matches the candidate scheme, splicing them in a preset parameter order to form the cross brace positioning characteristic parameters of the cross brace positioning candidate scheme.

[0107] Preferably, the convolutional neural network prediction model includes:

[0108] The construction state feature convolutional subnetwork is used to perform two-dimensional convolutional encoding on the construction state feature vector and the working condition encoding vector that fuses multiple modes, so as to obtain the first feature map of the current construction state and construction stage.

[0109] A convolutional subnetwork for cross brace positioning features is used to convolutionally encode the cross brace positioning feature parameters of each cross brace positioning candidate scheme, thereby obtaining a second feature map that represents the spatial position and attitude features of different cross brace positioning schemes.

[0110] The digital twin response embedding module performs structural analysis and calculation on each cross brace positioning candidate scheme to obtain the corresponding basic structure response result, and splices the basic structure response result with the first feature map and the second feature map in the channel dimension to form a third feature map that integrates physical information.

[0111] The residual correction and scoring output module is used to perform multi-scale convolution and residual connection operations on the third feature map, learn the deviation relationship between the basic structure response result and the target structure response index, output the structural response index of each cross brace positioning candidate scheme and the comprehensive score value used to indicate the final target cross brace positioning scheme. The comprehensive score value is used to automatically select the final target cross brace positioning scheme from multiple cross brace positioning candidate schemes.

[0112] Preferably, comparing the measured posture with the structural response index during the installation of the cross brace includes:

[0113] During the installation of the cross brace, the measured attitude parameters of the cross brace end in the whole bridge coordinate system are obtained in real time based on the processing of on-site video images. The measured attitude parameters include at least the three-dimensional coordinates of the feature points of the cross brace end and the measured rotation angles around the longitudinal, transverse and vertical axes of the bridge.

[0114] Extract the target attitude parameters corresponding to the final target cross brace positioning scheme from the structural response index. The target attitude parameters include at least the target three-dimensional coordinates and the target rotation angle.

[0115] Under a unified full-bridge coordinate system, the difference vector between the measured three-dimensional coordinates and the target three-dimensional coordinates, as well as the difference between the measured rotation angle and the target rotation angle, are calculated to obtain the displacement deviation and the angle deviation.

[0116] Based on preset displacement deviation threshold and angle deviation threshold, the magnitude of the displacement deviation and angle deviation is compared. When either deviation exceeds the corresponding threshold, it is determined that the attitude deviation exceeds the limit, and an adjustment prompt or alarm signal is triggered. The adjustment prompt includes one or more of the following: reducing the hoisting speed, fine-tuning the position of the cross brace end, and / or pausing the hoisting operation.

[0117] This application also provides an intelligent positioning device for the cross bracing of a cable-stayed bridge, driven by construction status data, such as... Figure 2 As shown, in a preferred embodiment, it includes the following hardware units:

[0118] The field sensor unit is used to collect multi-source construction status data during the construction phase of the cable-stayed bridge. It includes at least: three-dimensional displacement sensors deployed at key sections of the bridge towers; displacement / deflection sensors deployed at control sections of the main girder; cable force gauges installed on the stay cables; reaction force sensors deployed on supports and temporary supports; and temperature sensors deployed on the tower columns and main girder surfaces. Each sensor is electrically connected to the data acquisition and gateway unit via analog input interfaces, digital input interfaces, or fieldbuses (such as CAN, Modbus, Profibus, etc.).

[0119] The video acquisition unit is used to acquire on-site video images of the cross brace area. It includes several network cameras or industrial cameras deployed near the cross brace area, preferably high-definition network cameras supporting PoE power supply. Each camera is connected to a field switch via a network cable, and the field switch then connects to the video acquisition interface or network card of an industrial computer via an industrial Ethernet network, forming a video data channel.

[0120] Data acquisition and gateway unit. The data acquisition and gateway unit can be one or more data collectors, remote I / O modules or edge gateway devices, which are used to sample, quantify and preprocess analog and digital signals of on-site sensors, and then package them into construction status data frames after uniformly attaching time stamps. This unit communicates with the backend industrial computer through industrial Ethernet, fiber optic Ethernet or 4G / 5G wireless communication to achieve real-time upload of multi-source construction status data.

[0121] Industrial computer / server unit. The industrial computer / server unit is the core computing platform of this device, which at least includes a processor (CPU / GPU), a memory (RAM), a non-volatile storage device (SSD / HDD) and a network interface. The data acquisition module, feature vector generation module, digital twin model update module, candidate solution generation module, target cross brace positioning final solution generation module and installation control module can all be implemented by the processor in the industrial computer executing computer program instructions stored in the memory. This industrial computer communicates bidirectionally with the data acquisition and gateway unit, video acquisition unit and on-site control unit through an Ethernet interface, and stores multi-source construction status data, video image data, digital twin model parameters and convolutional neural network model parameters through a database or a file system.

[0122] Human-computer interaction and display terminal. The human-computer interaction and display terminal can be a display connected to the industrial computer, an industrial all-in-one machine or a tablet terminal, which is used to display the construction status monitoring interface, cross brace positioning candidate solutions and target solutions, structural response indicators and real-time attitude deviation information, and receive operations from construction personnel such as working condition parameter setting, threshold setting or solution confirmation. This terminal is connected to the industrial computer through Ethernet, serial bus or USB.

[0123] Lifting equipment and installation control unit. The installation control unit can include a PLC controller, a crane control system interface, a hydraulic jacking controller and related actuators, which are mainly used to receive the cross brace installation control instructions from the industrial computer and convert them into actual control actions such as lifting, horizontal movement, trolley movement, temporary support jacking and cable stay tensioning / relaxing of the crane hook. The industrial computer establishes a communication channel with the PLC or the crane equipment control system through a field bus (such as Profibus, Profinet, CANopen) or a dedicated communication protocol to achieve linkage control and status feedback of the hoisting process.

[0124] To ensure time consistency across multiple data sources, this device may also include a time synchronization unit, such as a GPS time synchronization device or an IEEE 1588 precision clock synchronization device, to provide a unified time reference for the data acquisition and gateway unit, industrial computer, and video acquisition unit. The field network preferably adopts an industrial Ethernet architecture, interconnecting the data acquisition and gateway unit, industrial computer, video acquisition unit, human-machine interface, and control unit through a core switch to form a closed-loop control network.

[0125] Under the aforementioned connection, the field sensor unit and video acquisition unit transmit multi-source construction status data and video image data to the industrial computer in real time through the data acquisition and gateway unit and the field switch, respectively. The functional modules running in the industrial computer perform feature extraction, multimodal fusion, digital twin model calibration, and intelligent selection of candidate and target schemes for cross bracing positioning on the received data. They generate cross bracing installation control commands and issue them to the lifting equipment through the installation control unit to achieve actual hoisting control. At the same time, the current construction status, cross bracing positioning scheme, and attitude deviation information are displayed to the construction personnel through human-computer interaction and display terminals. This constitutes an intelligent positioning device for cable-stayed bridge cross bracing that integrates "perception, calculation, decision-making, and execution" and is driven by construction status data.

[0126] Data acquisition module: acquires multi-source construction status data during the construction phase of the cable-stayed bridge, and simultaneously acquires on-site video images of the cross bracing area captured by cameras, and generates a working condition coding vector based on construction procedure information;

[0127] Feature vector generation module: preprocesses multi-source construction status data to obtain construction status feature vectors; performs camera calibration, target area detection, feature point tracking and 3D attitude inversion on on-site video images to obtain the 3D displacement and attitude parameters of the cross brace end and cross section, which constitute video image feature vectors, and aligns them with the construction status feature vectors in time to obtain fused multimodal construction status feature vectors.

[0128] Digital twin model update module: Input the fused multimodal construction state feature vector into the digital twin model of the cable-stayed bridge construction stage, calibrate the model parameters, and obtain a digital twin model that matches the current construction state;

[0129] Candidate solution generation module: Within the geometric constraints given by the digital twin model, multiple candidate solutions for cross bracing positioning are generated based on the cross bracing design position and allowable construction deviations, and cross bracing positioning feature parameters are constructed for each candidate solution.

[0130] Target cross brace positioning final scheme generation module: Input the fused multimodal construction state feature vector, the working condition encoding vector and the cross brace positioning feature parameters into the convolutional neural network prediction model to obtain the target cross brace positioning final scheme and the corresponding structural response index;

[0131] Installation control module: Generates cross brace installation control commands based on the final target cross brace positioning scheme. During the cross brace installation process, continuously collects video images to estimate the attitude of the cross brace end, compares the measured attitude with the structural response index, and outputs adjustment prompts when the deviation exceeds the preset threshold.

[0132] This invention provides a method and device for intelligent positioning of cross braces of cable-stayed bridges driven by construction status data, which can achieve the following beneficial technical effects:

[0133] 1. Compared to existing cross-bracing positioning methods that rely on a limited number of measuring points and static design alignment, this invention constructs a multi-modal sensing system combining "multi-source sensors + video images." Through unified preprocessing of monitoring data such as bridge tower displacement, main beam deflection, stay cable force, temporary support reaction force, and ambient temperature, and combined with three-dimensional attitude inversion of video images of the cross-bracing area, a construction state feature vector is formed under a unified coordinate system and time scale. This method overcomes the limitations of traditional point-based monitoring, which has limited measuring points and struggles to cover local areas of the cross-bracing. Furthermore, it fully utilizes on-site video resources originally intended only for safety monitoring, transforming them into quantifiable geometric measurement information. This allows for the high-precision real-time acquisition of spatial displacement and attitude at the ends of the cross-bracing and key sections, significantly improving the comprehensiveness and refinement of structural state identification during the construction phase, and providing a more reliable input foundation for subsequent intelligent positioning.

[0134] 2. This invention combines a digital twin model of the cable-stayed bridge construction stage calibrated with monitoring data with a convolutional neural network prediction model to achieve rapid and intelligent optimization of the cross brace positioning scheme while meeting physical constraints. The digital twin model provides the geometric constraint range and foundation structure response results, ensuring that all candidate schemes meet structural safety and geometric feasibility requirements. Based on this, the convolutional neural network performs deep feature extraction and residual correction on the fused multimodal construction state feature vector, working condition encoding vector, and cross brace positioning feature parameters. This enables efficient prediction of structural response indicators corresponding to different candidate schemes and outputs a comprehensive score, thereby automatically selecting the final target cross brace positioning scheme from a large number of candidate schemes. Compared with traditional schemes that rely on multiple finite element forward analyses, this invention significantly reduces redundant calculations and improves the real-time performance and intelligence level of cross brace positioning scheme evaluation and decision-making.

[0135] 3. This invention introduces a real-time attitude estimation and deviation comparison mechanism based on video images during the cross brace installation process, achieving closed-loop control between the intelligent positioning results of the cross brace and the actual construction process. By performing target area detection, feature point tracking, and 3D attitude inversion on continuously acquired on-site video images during the hoisting process, the measured 3D coordinates and rotation parameters of the cross brace end in the whole bridge coordinate system can be obtained in real time. These are then compared with the target attitude parameters corresponding to the final positioning scheme of the target cross brace, calculating the displacement deviation and angle deviation. When any deviation exceeds a preset threshold, an adjustment prompt of deceleration, fine-tuning, or pausing the hoisting is automatically issued. This invention not only corrects deviations in a timely manner during construction, avoiding structural stress concentration and construction safety risks caused by cross brace misalignment, but also reduces reliance on the experience judgment of construction personnel, reduces the probability of rework and the workload of measurement and verification, and significantly improves the accuracy and safety of cross brace installation during the construction phase of cable-stayed bridges.

[0136] The above provides a detailed description of an intelligent positioning method and device for cross bracing of cable-stayed bridges driven by construction status data. Specific examples have been used to illustrate the principles and implementation methods of the invention. The descriptions of the above embodiments are only for the purpose of helping to understand the core ideas of the invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas and methods of the invention. Therefore, the content of this specification should not be construed as a limitation of the invention.

Claims

1. A data-driven intelligent positioning method for cross bracing of cable-stayed bridges based on construction status, characterized in that: Including the following steps: S1: Acquire multi-source construction status data during the construction phase of the cable-stayed bridge, and simultaneously acquire on-site video images of the cross bracing area collected by the camera, and generate a working condition coding vector based on the construction procedure information. S2: Preprocess the multi-source construction status data to obtain the construction status feature vector; perform camera calibration, target area detection, feature point tracking and three-dimensional attitude inversion on the on-site video images to obtain the three-dimensional displacement and attitude parameters of the cross brace end and cross section, which constitute the video image feature vector, and align it with the construction status feature vector in time to obtain the fused multimodal construction status feature vector. S3: Input the fused multimodal construction state feature vector into the digital twin model of the cable-stayed bridge construction stage, calibrate the model parameters, and obtain a digital twin model that matches the current construction state; S4: Within the geometric constraints given by the digital twin model, generate multiple candidate schemes for cross bracing positioning based on the cross bracing design position and allowable construction deviations, and construct cross bracing positioning feature parameters for each candidate scheme. S5: Input the construction state feature vector of the fused multimodal structure, the working condition encoding vector and the cross brace positioning feature parameters into the convolutional neural network prediction model to obtain the final target cross brace positioning scheme and the corresponding structural response index. S6: Generate cross brace installation control commands based on the final target cross brace positioning scheme. During the cross brace installation process, continuously collect video images to estimate the attitude of the cross brace end. Compare the measured attitude with the structural response index. When the deviation exceeds the preset threshold, output an adjustment prompt.

2. The intelligent positioning method for cross bracing of cable-stayed bridges driven by construction status data as described in claim 1, characterized in that, The multi-source construction status data includes bridge tower displacement, main beam displacement, cable tension, strain, temporary support reaction force, and environmental parameters.

3. The intelligent positioning method for cross bracing of cable-stayed bridges driven by construction status data as described in claim 1, characterized in that, The process of generating a working condition coding vector based on construction procedure information includes: collecting current construction procedure information, which includes at least one or more of the following: the hoisting sequence number of the current main beam segment, the tensioning completion status of the corresponding stay cable, the tower-beam connection status, and the activation status of the temporary support; mapping the construction procedure information to a working condition category identifier that represents the construction stage type according to a preset construction stage division rule; and representing the working condition category identifier as a working condition coding vector through multidimensional binary encoding.

4. The intelligent positioning method for cross bracing of cable-stayed bridges driven by construction status data as described in claim 1, characterized in that, The preprocessing of multi-source construction status data to obtain the construction status feature vector includes: acquiring raw monitoring data collected by displacement sensors, strain gauges, cable force gauges, support reaction force gauges, and thermometers deployed on bridge towers, main beams, and stay cables; using low-pass filtering to suppress high-frequency noise in the raw monitoring data; performing coordinate transformation and dimension normalization on the displacement, strain, cable force, and support reaction force data of each monitoring point; and splicing the preprocessed displacement, strain, cable force, support reaction force, and temperature feature quantities in a fixed order according to a preset cross-section and component arrangement sequence to form the construction status feature vector.

5. The intelligent positioning method for cross bracing of cable-stayed bridges driven by construction status data as described in claim 1, characterized in that, The process of constructing the video image feature vector includes: calibrating the internal and external parameters of the cameras deployed in the cross brace area to establish a spatial mapping relationship between the camera pixel coordinate system and the overall bridge coordinate system of the cable-stayed bridge; then, performing target area detection on the on-site video images to extract image areas of the cross brace components, cross brace end markers, and preset cross brace cross-sectional structural features; selecting several feature points within the image areas and tracking the pixel coordinates of the feature points in adjacent video frames based on an optical flow tracking algorithm; using the camera calibration parameters and the initial three-dimensional coordinates of the feature points in the reference frame, combined with the perspective-n-point solving algorithm, inverting to obtain the three-dimensional coordinates of the cross brace end and preset cross-sectional feature points in the overall bridge coordinate system; calculating the difference between the three-dimensional coordinates and the three-dimensional coordinates of the corresponding feature points under the reference construction state to obtain the three-dimensional displacement of the cross brace end and cross-section, and calculating the attitude angle parameters of the cross brace end and cross-section based on the fitted cross brace axis and local cross-section normal vector; and concatenating the three-dimensional displacement parameters and attitude angle parameters in a preset order to obtain the video image feature vector.

6. The intelligent positioning method for cross bracing of cable-stayed bridges driven by construction status data as described in claim 1, characterized in that, Step S3 includes: extracting the three-dimensional displacement of the top and middle sections of the bridge tower, the deflection and planar displacement of the main beam control section, the cable force of the stay cables, and the measured response of the temporary support from the construction state feature vector fused with multimodal data, and associating the measured response with the corresponding calculation nodes and components in the digital twin model; selecting at least one of temperature load distribution, construction additional load, support stiffness, and material elastic modulus as the parameter to be calibrated, and performing forward structural response analysis calculation in the digital twin model with the parameter to be calibrated as the variable to obtain the model calculated response; constructing a parameter inverse analysis model with the difference between the model calculated response and the corresponding measured response as the objective function, and using iterative least squares and Kalman filtering algorithms to update the parameter to be calibrated multiple times; when the displacement deviation and cable force deviation at the key monitoring positions are both less than the preset threshold, a digital twin model matching the current construction state is obtained.

7. The intelligent positioning method for cross bracing of cable-stayed bridges driven by construction status data as described in claim 1, characterized in that, Step S4 includes: obtaining the design coordinates and design attitude parameters of both ends of the cross brace in the design full-bridge coordinate system, the design attitude parameters including the cross brace axis direction and the design rotation angles around the longitudinal, transverse and vertical axes of the bridge; based on the calibrated digital twin model of the cable-stayed bridge construction stage, determining the geometric constraint range of the nodes at both ends of the cross brace in the vertical, longitudinal and transverse directions, the minimum clearance requirement with adjacent components, and the allowable additional displacement and internal force range of the tower, beam and cable after the cross brace is installed; within the geometric constraint range, performing combined perturbations on the vertical elevation, longitudinal position, transverse offset and attitude rotation angle around the cross brace design position at a preset step size to generate multiple cross brace positioning candidate schemes that meet the geometric constraints; for each cross brace positioning candidate scheme, calculating the three-dimensional coordinates of the nodes at both ends of the cross brace in the full-bridge coordinate system, the cross brace axis direction vector and the corresponding attitude angle, and adding the cable force adjustment amount that matches the candidate scheme, splicing them in a preset parameter order to form the cross brace positioning characteristic parameters of the cross brace positioning candidate scheme.

8. The intelligent positioning method for cross bracing of cable-stayed bridges driven by construction status data as described in claim 1, characterized in that, The convolutional neural network prediction model includes: The construction state feature convolutional subnetwork is used to perform two-dimensional convolutional encoding on the construction state feature vector and the working condition encoding vector that fuses multiple modes, so as to obtain the first feature map of the current construction state and construction stage. A convolutional subnetwork for cross brace positioning features is used to convolutionally encode the cross brace positioning feature parameters of each cross brace positioning candidate scheme, thereby obtaining a second feature map that represents the spatial position and attitude features of different cross brace positioning schemes. The digital twin response embedding module performs structural analysis and calculation on each cross brace positioning candidate scheme to obtain the corresponding basic structure response result, and splices the basic structure response result with the first feature map and the second feature map in the channel dimension to form a third feature map that integrates physical information. The residual correction and scoring output module is used to perform multi-scale convolution and residual connection operations on the third feature map, learn the deviation relationship between the basic structure response result and the target structure response index, output the structural response index of each cross brace positioning candidate scheme and the comprehensive score value used to indicate the final target cross brace positioning scheme. The comprehensive score value is used to automatically select the final target cross brace positioning scheme from multiple cross brace positioning candidate schemes.

9. The intelligent positioning method for cross bracing of cable-stayed bridges driven by construction status data as described in claim 1, characterized in that, The comparison between the measured attitude and structural response indicators during the installation of the cross brace includes: During the installation of the cross brace, the measured attitude parameters of the cross brace end in the whole bridge coordinate system are obtained in real time based on the processing of on-site video images. The measured attitude parameters include at least the three-dimensional coordinates of the feature points of the cross brace end and the measured rotation angles around the longitudinal, transverse and vertical axes of the bridge. Extract the target attitude parameters corresponding to the final target cross brace positioning scheme from the structural response index. The target attitude parameters include at least the target three-dimensional coordinates and the target rotation angle. Under a unified full-bridge coordinate system, the difference vector between the measured three-dimensional coordinates and the target three-dimensional coordinates, as well as the difference between the measured rotation angle and the target rotation angle, are calculated to obtain the displacement deviation and the angle deviation. Based on preset displacement deviation threshold and angle deviation threshold, the magnitude of the displacement deviation and angle deviation is compared. When either deviation exceeds the corresponding threshold, it is determined that the attitude deviation exceeds the limit, and an adjustment prompt or alarm signal is triggered. The adjustment prompt includes one or more of the following: reducing the hoisting speed, fine-tuning the position of the cross brace end, and / or pausing the hoisting operation.

10. A smart positioning device for the cross bracing of a cable-stayed bridge driven by construction status data, characterized in that: include: Data acquisition module: acquires multi-source construction status data during the construction phase of the cable-stayed bridge, and simultaneously acquires on-site video images of the cross bracing area captured by cameras, and generates a working condition coding vector based on construction procedure information; Feature vector generation module: preprocesses multi-source construction status data to obtain construction status feature vectors; performs camera calibration, target area detection, feature point tracking and 3D attitude inversion on on-site video images to obtain the 3D displacement and attitude parameters of the cross brace end and cross section, which constitute video image feature vectors, and aligns them with the construction status feature vectors in time to obtain fused multimodal construction status feature vectors. Digital twin model update module: Input the fused multimodal construction state feature vector into the digital twin model of the cable-stayed bridge construction stage, calibrate the model parameters, and obtain a digital twin model that matches the current construction state; Candidate solution generation module: Within the geometric constraints given by the digital twin model, multiple candidate solutions for cross bracing positioning are generated based on the cross bracing design position and allowable construction deviations, and cross bracing positioning feature parameters are constructed for each candidate solution. Target cross brace positioning final scheme generation module: Input the fused multimodal construction state feature vector, the working condition encoding vector and the cross brace positioning feature parameters into the convolutional neural network prediction model to obtain the target cross brace positioning final scheme and the corresponding structural response index; Installation control module: Generates cross brace installation control commands based on the final target cross brace positioning scheme. During the cross brace installation process, continuously collects video images to estimate the attitude of the cross brace end, compares the measured attitude with the structural response index, and outputs adjustment prompts when the deviation exceeds the preset threshold.

Images