Machine learning-based installation precision dynamic regulation method and system for skew-span steel box arch bridge

Abstract

The application relates to the technical field of construction control, and particularly discloses a dynamic regulation and control method and system for installation precision of a skew-span steel box arch bridge based on machine learning, which comprises the following steps: collecting real-time data in the construction process of the skew-span steel box arch bridge, wherein the real-time data comprises arch rib real-time stress, arch rib real-time deformation, real-time environmental wind speed and real-time hoisting parameters; constructing a prediction model capable of analyzing the coupling effect of wind load and hoisting eccentric load based on a large amount of historical construction data; inputting the real-time data into the prediction model to obtain real-time coupling effect analysis results; outputting hoisting sequence optimization instructions based on a decision engine, the coupling effect analysis results and preset mechanical constraint conditions of the skew-span steel box arch bridge; and generating a cable force grading adjustment scheme based on the deviation of the arch rib real-time deformation and a preset installation precision threshold value; the method effectively improves the construction efficiency of the skew-span steel box arch bridge, guarantees the construction safety and installation precision, and ensures the stability and reliability of the bridge structure.

Description

Technical Field

[0001] This invention relates to the field of construction control technology, and in particular to a method and system for dynamic control of installation accuracy of skew-span steel box arch bridges based on machine learning. Background Technology

[0002] Cable-span steel box arch bridges are widely used in modern transportation infrastructure construction due to their unique shape and excellent mechanical properties. However, the installation process of cable-span steel box arch bridges is extremely challenging, especially in terms of the extremely high requirements for installation accuracy. This not only affects the appearance of the bridge but also directly relates to its structural stability and load-bearing capacity, and even more importantly, its safety and durability during subsequent use. During the construction of cable-span steel box arch bridges, multiple complex factors intertwine and affect installation accuracy. Wind load, as an indispensable natural factor, alters the stress state of the structure, inducing vibration and displacement; while uneven loading during hoisting, due to the complexity of construction operations, may lead to uneven stress on the structure, further affecting installation accuracy. In addition, as construction progresses, the various construction stages are interconnected, and early installation deviations may accumulate in subsequent construction, increasing the difficulty of controlling installation accuracy. Currently, applying machine learning technology to the control of installation accuracy in cable-span steel box arch bridges is a promising direction. This will not only help improve the quality of bridge construction and reduce later maintenance costs caused by installation accuracy issues but also promote the intelligent and refined advancement of bridge construction technology, showing broad application prospects in the future bridge construction field.

[0003] However, existing installation accuracy control technologies for skeletal steel box arch bridges are insufficient to effectively assess the combined effects of complex factors and obtain real-time analysis results of coupling effects. They also cannot adjust the construction sequence in a timely manner to reduce the impact of adverse factors. Consequently, it is impossible to generate a scientific and reasonable cable force grading adjustment scheme, making it difficult to dynamically and accurately control the installation accuracy, ultimately affecting the overall installation quality of the skeletal steel box arch bridge.

[0004] Therefore, this invention proposes a method and system for dynamic control of installation accuracy of slanted steel box arch bridges based on machine learning. Summary of the Invention

[0005] This invention provides a method and system for dynamic control of installation accuracy of skew-span steel box arch bridges based on machine learning. Leveraging the powerful data processing and analysis capabilities of machine learning, it delves into the potential relationships between various data points during construction, enabling precise control of installation accuracy. A predictive model is constructed using extensive historical construction data to analyze the coupling effect between wind load and hoisting eccentric load. Real-time construction data, including arch rib stress, deformation, ambient wind speed, and hoisting parameters, are comprehensively collected and input into the model to obtain real-time analysis results of the coupling effect. This is then combined with a decision engine and preset mechanical constraints to output optimized hoisting sequence instructions. Simultaneously, a graded adjustment scheme for cable force is generated based on the deviation between the real-time arch rib deformation and preset accuracy thresholds. This achieves scientific management of the entire process from data acquisition and effect prediction to construction instruction optimization and accuracy control, effectively improving the construction efficiency of skew-span steel box arch bridges, ensuring construction safety and installation accuracy, and guaranteeing the stability and reliability of the bridge structure.

[0006] This invention provides a machine learning-based method for dynamically adjusting the installation accuracy of a slant-span steel box arch bridge, comprising:

[0007] Real-time data was collected during the construction of the skew-span steel box arch bridge, including real-time stress of the arch ribs, real-time deformation of the arch ribs, real-time ambient wind speed, and real-time hoisting parameters.

[0008] A predictive model capable of analyzing the coupling effect of wind load and hoisting off-center load was constructed based on a large amount of historical construction data.

[0009] Real-time data is input into the prediction model to obtain real-time analysis results of coupling effects;

[0010] Based on the analysis results of the decision engine and coupling effect, as well as the preset mechanical constraints of the skew-span steel box arch bridge, the hoisting sequence optimization command is output.

[0011] Based on the deviation between the real-time deformation of the arch rib and the preset installation accuracy threshold, a graded adjustment scheme for cable force is generated.

[0012] Preferably, real-time data is collected during the construction of the skew-span steel box arch bridge, including:

[0013] The real-time axial and radial stresses of the arch ribs during the construction of the skew-span steel box arch bridge are collected by fiber optic grating sensors uniformly arranged along the longitudinal direction of the arch ribs.

[0014] The three-dimensional coordinates of the arch rib during the construction of the sloping steel box arch bridge are collected in real time by laser trackers set on the piers on both sides of the arch rib, and the lateral bending deformation and axial deviation are calculated based on the real-time collected three-dimensional coordinates as the real-time deformation of the arch rib.

[0015] The instantaneous wind speed and direction during the construction of the sloping steel box arch bridge are collected by wind speed sensors installed at the top of the arch ribs as real-time environmental wind speed.

[0016] Real-time segmental physical parameters, real-time hoisting posture parameters, and real-time motion state parameters are collected during the construction of the skew-span steel box arch bridge as real-time hoisting parameters.

[0017] Preferably, a predictive model is constructed based on a large amount of historical construction data to analyze the coupling effect of wind load and hoisting off-center load, including:

[0018] Simulate the combined working conditions of wind load and hoisting eccentric load from a large amount of historical construction data to obtain a preset set of sample data;

[0019] All sample data are divided into training and validation sets according to a preset ratio. A neural network model is trained based on the training and validation sets. At the same time, the mean square error between the predicted deformation and the actual deformation is minimized by the Adam optimizer until the mean square error does not exceed the preset error threshold. This yields a predictive model that can analyze the coupling effect between wind load and hoisting off-center load.

[0020] Preferably, the decision engine construction process also includes:

[0021] The reward function for the reinforcement learning algorithm is generated based on the actual deformation, allowable deformation, and hoisting delay rate.

[0022] A decision engine is built based on the reward function of the reinforcement learning algorithm.

[0023] Preferably, based on the decision engine and coupling effect analysis results, as well as the preset mechanical constraints of the skew-span steel box arch bridge, the hoisting sequence optimization instructions are output, including:

[0024] A lifting sequence thread is generated based on each optional lifting sequence;

[0025] The reward value for each hoisting sequence thread is determined based on the reward function of the reinforcement learning algorithm in the decision engine, and hoisting sequence optimization instructions are generated based on the hoisting sequence thread corresponding to the maximum reward value.

[0026] Preferably, based on the deviation between the real-time deformation of the arch rib and the preset installation accuracy threshold, a graded adjustment scheme for cable force is generated, including:

[0027] The real-time deflection value of the arch rib is determined based on the real-time deformation of the arch rib.

[0028] Use the preset maximum preset deflection value as the preset installation accuracy threshold;

[0029] When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold does not exceed the first deviation threshold, the first proportional initial tensioning will be executed as a cable force grade adjustment scheme.

[0030] When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold exceeds the first deviation threshold but does not exceed the second deviation threshold, the first to second proportional graded tensioning will be performed as a graded adjustment scheme for cable force.

[0031] When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold exceeds the second deviation threshold, the maximum proportional tensioning and the cable force correction of all hangers calculated by the matrix influence method will be re-adjusted as the cable force grade adjustment scheme.

[0032] Preferably, the cable force correction for all suspenders is calculated based on the matrix influence method, including:

[0033] An influence matrix is ​​constructed using the hanger numbers of the slanted steel box arch bridge as row vectors and the key control sections of the arch ribs as column vectors.

[0034] A basic deviation vector is generated based on the deviation between the real-time deflection values ​​of all key control sections of the arch ribs and the preset installation accuracy threshold.

[0035] The base deviation vector is corrected based on the preset wind-induced correction coefficient and temperature correction coefficient to obtain the corrected deviation vector;

[0036] The multidimensional cable force correction vector is solved by regularized least squares method based on the influence matrix and the correction deviation vector;

[0037] The cable force correction vector is filtered by safety constraints, equilibrium constraints, and construction feasibility constraints to obtain the cable force correction amount for all suspenders.

[0038] Preferably, the multidimensional cable force correction vector is subjected to equilibrium constraint filtering, including:

[0039] A finite element model of the construction process of the skew-span steel box arch bridge is generated. In the finite element model, a force flow monitoring section is selected at a preset distance along the longitudinal direction of the arch rib. Multiple feature points are selected in each force flow monitoring section. The axial normal stress direction, radial normal stress direction, and shear stress direction of each feature point are extracted. The force flow direction angle of each feature point is calculated based on the axial normal stress direction, radial normal stress direction, and shear stress direction of each feature point.

[0040] Continuous sections with the same structural or stress characteristics in the finite element model are treated as sections to be analyzed.

[0041] Based on the force flow direction angle of all feature points in the finite element model, the force flow abrupt change section is selected from all sections to be analyzed.

[0042] The strength requirements of the equilibrium constraints are determined based on the distribution location and degree of abrupt changes in all force flow segments.

[0043] The multidimensional cable force correction vector is filtered based on the strength requirements of the equilibrium constraint.

[0044] Preferably, based on the force flow direction angles of all feature points in the finite element model, the force flow abrupt change zones are selected from all the zones to be analyzed, including:

[0045] The ratio of the force flow direction angle difference of the feature points at the same position in each group of adjacent force flow monitoring sections to the distance between adjacent sections is taken as the corresponding direction angle change rate.

[0046] The standard deviation of the force flow direction angle of all feature points in the section to be analyzed is defined as the force flow continuity index.

[0047] The ratio of all abrupt changes in orientation angles to all feature points in the section to be analyzed is taken as the proportion of extreme points.

[0048] Based on the rate of change of all directional angles, the force-flow continuity index, and the proportion of extreme points in each analysis segment, the force-flow abrupt change segments are selected from all analysis segments.

[0049] This invention provides a machine learning-based dynamic control system for the installation accuracy of a slant-span steel box arch bridge, comprising:

[0050] The data acquisition module is used to collect real-time data during the construction of the slanted steel box arch bridge. The real-time data includes real-time stress of the arch ribs, real-time deformation of the arch ribs, real-time ambient wind speed, and real-time hoisting parameters.

[0051] The model building module is used to build a predictive model based on a large amount of historical construction data that can analyze the coupling effect of wind load and hoisting off-center load.

[0052] The coupling analysis module is used to input real-time data into the prediction model to obtain real-time coupling effect analysis results;

[0053] The sequence optimization module is used to output hoisting sequence optimization instructions based on the decision engine and coupling effect analysis results, as well as the preset mechanical constraints of the skew-span steel box arch bridge.

[0054] The scheme generation module is used to generate cable force graded adjustment schemes based on the deviation between the real-time deformation of the arch rib and the preset installation accuracy threshold.

[0055] The beneficial effects of this invention compared to existing technologies are as follows: It leverages the powerful data processing and analysis capabilities of machine learning to deeply explore the potential relationships between various data points during construction, enabling precise control of installation accuracy. A predictive model capable of analyzing the coupling effect between wind load and hoisting eccentric load is constructed using a large amount of historical construction data. Real-time coupling effect analysis results are obtained by comprehensively collecting real-time construction data such as arch rib stress, deformation, ambient wind speed, and hoisting parameters. This data is then combined with a decision engine and preset mechanical constraints to output optimized hoisting sequence instructions. Simultaneously, a cable force grading adjustment scheme is generated based on the deviation between the real-time arch rib deformation and preset accuracy thresholds. This achieves scientific management of the entire process from data acquisition and effect prediction to construction instruction optimization and accuracy control, effectively improving the construction efficiency of the skew-span steel box arch bridge, ensuring construction safety and installation accuracy, and guaranteeing the stability and reliability of the bridge structure.

[0056] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0057] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0059] Figure 1 This is a schematic diagram of a machine learning-based dynamic control method for the installation accuracy of a slanted steel box arch bridge in an embodiment of the present invention.

[0060] Figure 2 This is a flowchart illustrating the real-time data acquisition process during the construction of the skew-span steel box arch bridge in this embodiment of the invention.

[0061] Figure 3 This is a flowchart of the prediction model construction method in an embodiment of the present invention. Detailed Implementation

[0062] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0063] like Figure 1 As shown, this invention provides a method for dynamically adjusting the installation accuracy of a slanted steel box arch bridge based on machine learning, comprising:

[0064] Real-time data was collected during the construction of the skew-span steel box arch bridge, including real-time stress of the arch ribs, real-time deformation of the arch ribs, real-time ambient wind speed, and real-time hoisting parameters.

[0065] A predictive model capable of analyzing the coupling effect of wind load and hoisting off-center load was constructed based on a large amount of historical construction data.

[0066] Real-time data is input into the prediction model to obtain real-time analysis results of coupling effects;

[0067] Based on the analysis results of the decision engine and coupling effect, as well as the preset mechanical constraints of the skew-span steel box arch bridge, the hoisting sequence optimization command is output.

[0068] Based on the deviation between the real-time deformation of the arch rib and the preset installation accuracy threshold, a graded adjustment scheme for cable force is generated.

[0069] In this embodiment, the skew-span steel box arch bridge and its construction process are described: The skew-span steel box arch bridge is a type of structure used in bridge construction, characterized by its unique shape and excellent mechanical properties. The construction process involves multiple stages, including component hoisting and cable tension adjustment, and is affected by various factors, such as environmental factors (wind load) and construction operation factors (hoisting off-center loading), requiring extremely high installation accuracy.

[0070] In this embodiment, a large amount of historical construction data refers to the data accumulated in many past slanted steel box arch bridge construction projects, covering information on all aspects of construction, such as the stress and deformation data of the arch ribs at different construction stages, the environmental wind speed and hoisting parameters at each stage, and the actual deformation results under the corresponding wind load and hoisting off-center load combination conditions.

[0071] In this embodiment, a predictive model for the coupling effect of wind load and hoisting eccentric load is analyzed: it is used to analyze the effect of the combined action of wind load and hoisting eccentric load, that is, the influence of the combined action of wind load and hoisting eccentric load on the installation accuracy of arch ribs under the high wind environment of plateau.

[0072] In this embodiment, the real-time coupling effect analysis results are as follows: the real-time stress of the arch rib, the real-time deformation of the arch rib, the real-time environmental wind speed and the real-time hoisting parameters collected in real time during the construction of the skew-span steel box arch bridge are input into the prediction model for analyzing the coupling effect of wind load and hoisting off-center load. The model outputs the analysis results on the effect (or the impact on the installation accuracy of the arch rib) under the combined action of the current wind load and hoisting off-center load.

[0073] In this embodiment, the decision engine is a mechanism used to combine the analysis results of real-time coupling effects with preset mechanical constraints to assist in making decisions about the hoisting sequence.

[0074] In this embodiment, the preset mechanical constraints of the skew-span steel box arch bridge are: restrictions set in advance during the design and construction planning stage of the skew-span steel box arch bridge, based on considerations of bridge structural mechanics and construction safety, such as the maximum stress that the structure can withstand, the range of deformation, and the load-bearing capacity of each component. These conditions restrict various operations during the construction process.

[0075] In this embodiment, the hoisting sequence optimization instruction is an instruction that guides the hoisting sequence during the construction of a sloping steel box arch bridge. Example of an optimization instruction:

[0076] For the skewed steel box arch bridge, a symmetrical alternating hoisting method is adopted, namely, left arch foot segment → right arch foot segment → left 1 / 4 span segment → right 1 / 4 span segment;

[0077] Dynamic adjustments are made based on real-time monitoring data (e.g., when the stress on a certain side arch rib exceeds 250MPa, priority is given to hoisting the opposite side segment to balance the stress).

[0078] The goal is to control the cumulative deformation of the arch rib to ≤28mm and ensure the accuracy of the 20mm expansion joint reserved at the closure joint.

[0079] like Figure 2 As shown, in order to obtain comprehensive and accurate real-time data during the construction of a skew-span steel box arch bridge, a method for collecting real-time data during the construction process is proposed, including:

[0080] The real-time axial and radial stresses of the arch ribs during the construction of the skew-span steel box arch bridge are collected by fiber optic grating sensors uniformly arranged along the longitudinal direction of the arch ribs.

[0081] The three-dimensional coordinates of the arch rib during the construction of the sloping steel box arch bridge are collected in real time by laser trackers set on the piers on both sides of the arch rib, and the lateral bending deformation and axial deviation are calculated based on the real-time collected three-dimensional coordinates as the real-time deformation of the arch rib.

[0082] The instantaneous wind speed and direction during the construction of the sloping steel box arch bridge are collected by wind speed sensors installed at the top of the arch ribs as real-time environmental wind speed.

[0083] Real-time segmental physical parameters, real-time hoisting posture parameters, and real-time motion state parameters are collected during the construction of the skew-span steel box arch bridge as real-time hoisting parameters.

[0084] In this embodiment, fiber optic grating sensors are uniformly arranged along the longitudinal direction of the arch rib of the inclined steel box arch bridge. During construction, the axial and radial stresses of the arch rib are collected in real time; for example, the sensors record stress changes when components are hoisted. Laser trackers are installed on the piers on both sides of the arch rib to collect the three-dimensional coordinates of the arch rib in real time during construction; for example, the trackers record coordinate changes when segments are installed. Then, based on the three-dimensional coordinates of the arch rib collected in real time by the laser trackers, the lateral bending deformation and axial deviation are calculated. For example, deformation and deviation are calculated after segment installation to assess construction progress. Additionally, the top of the arch rib, referring to the highest point of the arch rib of the inclined steel box arch bridge, can be used to install equipment such as wind speed sensors.

[0085] This embodiment collects segmental physical parameters (such as single segment weight, cross-sectional dimensions, and center of gravity coordinates), hoisting posture parameters (such as stage hoisting angle: vertical tilt angle, lateral sway angle; force distribution at the hoisting point, i.e., the tension value of each sling) and motion state parameters (such as hoisting speed, horizontal moving speed, and relative displacement during high-altitude docking) during the construction of the skewed steel box arch bridge.

[0086] like Figure 3 As shown, in order to construct a predictive model that can analyze the coupling effect of wind load and hoisting eccentric load, a predictive model based on a large amount of historical construction data is proposed, including:

[0087] Simulate the combined working conditions of wind load and hoisting eccentric load from a large amount of historical construction data to obtain a preset set of sample data;

[0088] All sample data are divided into training and validation sets according to a preset ratio. A neural network model is trained based on the training and validation sets. At the same time, the mean square error between the predicted deformation and the actual deformation is minimized by the Adam optimizer until the mean square error does not exceed the preset error threshold. This yields a predictive model that can analyze the coupling effect between wind load and hoisting off-center load.

[0089] This embodiment simulates the combination of wind load and hoisting off-center load in a large amount of historical construction data to obtain a set of pre-set sample data, such as simulating different combinations of wind force and hoisting off-center load, and generating corresponding samples.

[0090] The preset ratio mentioned in this embodiment is a pre-set ratio value, which may be used for sample division, such as dividing the samples into training set and test set according to a preset ratio of 7:3.

[0091] In this embodiment, a neural network model is trained using a training set and a validation set. The model parameters are continuously adjusted with the help of the Adam optimizer to minimize the mean square error between the predicted deformation and the actual deformation. When the mean square error is less than or equal to a preset error threshold, a prediction model that can analyze the coupling effect of wind load and hoisting off-center load is obtained. For example, the model is trained to predict the deformation of a bridge under specific working conditions and optimized until the error meets the requirements.

[0092] In this embodiment, the predicted deformation refers to the deformation calculated by the model, while the actual deformation is the deformation measured in a real construction scenario. For example, the model predicts that the bridge will deform by 5 centimeters, while the actual measured deformation is 4.8 centimeters.

[0093] The preset error threshold in this embodiment is a pre-set error limit value used to determine whether the model training meets the standard. For example, if it is set to 0.5, the requirement is met when the mean square error of the model does not exceed 0.5.

[0094] To generate a reward function based on actual deformation, allowable deformation, and hoisting delay rate, and then construct a decision engine to achieve more scientific construction decisions, the construction process of the decision engine is proposed, which also includes:

[0095] The reward function for the reinforcement learning algorithm is generated based on the actual deformation, allowable deformation, and hoisting delay rate.

[0096] A decision engine is built based on the reward function of the reinforcement learning algorithm.

[0097] The actual deformation in this embodiment refers to the actual deformation values ​​of structural components such as the arch ribs obtained through measurement during the actual construction of the skew-span steel box arch bridge. It reflects the actual morphological changes of the structure at the time of construction. For example, after the installation of a certain key component, the actual deformation of the arch rib was measured to be 3 millimeters.

[0098] The allowable deformation is the maximum range of deformation that a structural component can withstand during construction, predetermined according to bridge design standards and safety regulations. It is an important indicator for ensuring the safety and function of a bridge structure. For example, the design may specify that the allowable deformation of a certain arch rib is 5 millimeters during a specific construction stage.

[0099] The hoisting delay rate measures the percentage delay between the planned and actual construction period during the hoisting of a sloping steel box arch bridge. For example, if the planned hoisting period is 30 days but it actually takes 35 days, then the hoisting delay rate = (35-30)÷30×100%≈16.7%.

[0100] In this embodiment, the reward function in the reinforcement learning algorithm is constructed using the actual deformation, the allowable deformation, and the hoisting delay rate as variables:

[0101] Reward = 0.6 × (1 - actual deformation / allowable deformation) + 0.4 × (1 - project delay rate).

[0102] This embodiment builds a decision engine based on the reinforcement learning algorithm reward function constructed above. The decision engine optimizes and adjusts various decisions during the construction process according to different reward values ​​fed back by the reward function, such as adjusting the hoisting sequence and construction technology, to achieve better construction results. For example, based on the reward function feedback, the decision engine decides to add temporary support for a certain hoisting stage to reduce deformation and avoid delays in the construction period.

[0103] To optimize the hoisting sequence, a hoisting sequence optimization instruction is proposed based on the results of the decision engine and coupling effect analysis, as well as the preset mechanical constraints of the skew-span steel box arch bridge. This instruction includes:

[0104] A lifting sequence thread is generated based on each optional lifting sequence;

[0105] The reward value for each hoisting sequence thread is determined based on the reward function of the reinforcement learning algorithm in the decision engine, and hoisting sequence optimization instructions are generated based on the hoisting sequence thread corresponding to the maximum reward value.

[0106] The optional hoisting sequence in this embodiment refers to the various feasible arrangements for the hoisting of different components during the construction of the skew-span steel box arch bridge. For example, for three different arch rib segments A, B, and C, the optional hoisting sequences may include ABC, ACB, BAC, etc.

[0107] In this embodiment, each optional hoisting sequence is transformed into an independent hoisting sequence thread. A thread can be understood as a process unit that executes tasks according to a specific hoisting sequence; each thread simulates or executes hoisting operations according to its corresponding hoisting sequence. For example, for the optional hoisting sequence ABC, a dedicated thread is generated to be responsible for performing the relevant simulations or operations according to this sequence.

[0108] In this embodiment, a reward function based on a reinforcement learning algorithm in the decision engine is used to evaluate each hoisting sequence thread, thereby determining the reward value for each thread. The reward value reflects the rationality and quality of the hoisting sequence in actual construction. For example, for the ABC hoisting sequence thread, if the actual deformation is small, does not exceed the allowable deformation, and has a low hoisting delay rate when hoisting in this order, the reward function will give a higher reward value; conversely, if there is excessive deformation or serious delays, the reward value will be lower.

[0109] The decision engine, based on reinforcement learning algorithms, treats the hoisting sequence as an "action sequence" and evaluates the merits of different sequences through a reward function. Specific rules are as follows:

[0110] Basic sequence library: Preset 3 types of initial sequences (symmetrical alternation, from arch foot to arch crown, from arch crown to arch foot), such as "left arch foot segment → right arch foot segment → left 1 / 4 span segment → right 1 / 4 span segment" (symmetrical alternation);

[0111] Dynamic adjustment logic:

[0112] When the coupling effect analysis result is "low risk" (wind speed < 5m / s, off-center load < 3%): prioritize "efficiency priority sequence" (continuous hoisting of adjacent segments on the same side to shorten the construction period);

[0113] When the analysis result is "medium risk" (wind speed 5-10m / s, eccentric load 3%-5%): switch to "symmetric balance sequence" (alternate left and right hoisting, hoist one segment on each side and then switch sides) to reduce asymmetrical load;

[0114] When the analysis result is "high risk" (wind speed ≥10m / s or off-center load ≥5%): trigger "pause-dispersion sequence" (first hoist the already lifted segment to the temporary fixed position, and then hoist the opposite segment after an interval of 30 minutes to avoid load superposition).

[0115] To develop tensioning schemes with different proportions as cable force grading adjustment schemes based on varying deviations between the real-time arch rib deflection value and the preset installation accuracy threshold, and to achieve precise cable force adjustment, a cable force grading adjustment scheme is proposed based on the deviation between the real-time arch rib deformation and the preset installation accuracy threshold. This scheme includes:

[0116] The real-time deflection value of the arch rib is determined based on the real-time deformation of the arch rib.

[0117] Use the preset maximum preset deflection value as the preset installation accuracy threshold;

[0118] When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold does not exceed the first deviation threshold, the first proportional initial tensioning will be executed as a cable force grade adjustment scheme.

[0119] When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold exceeds the first deviation threshold but does not exceed the second deviation threshold, the first to second proportional graded tensioning will be performed as a graded adjustment scheme for cable force.

[0120] When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold exceeds the second deviation threshold, the maximum proportional tensioning and the cable force correction of all hangers calculated by the matrix influence method will be re-adjusted as the cable force grade adjustment scheme.

[0121] In this embodiment, the real-time deformation of the arch rib refers to the change in the spatial position of the arch rib acquired in real time during the construction of the slanted steel box arch bridge. For example, the lateral bending deformation and axial deviation of the arch rib calculated by data collected by a laser tracker are all considered real-time deformation of the arch rib.

[0122] In this embodiment, the real-time deflection value of the arch rib is determined based on the real-time deformation of the arch rib: the real-time three-dimensional coordinates of the key control sections of the arch rib (such as the arch top, 1 / 4 span, and arch foot) are collected by a laser tracker, and the difference between its vertical coordinate and the design coordinate is calculated, which is the real-time deflection value of the corresponding section. The difference of the arch top section is taken as the representative deflection value of the arch rib in the current state.

[0123] In this embodiment, the real-time deflection value of the arch rib is a value that is calculated based on the real-time deformation of the arch rib and reflects the current degree of vertical deformation of the arch rib.

[0124] In this embodiment, the maximum preset deflection value is the maximum deflection value that the arch rib can withstand, which is preset during the bridge design stage based on the bridge's function, structural characteristics, and safety standards.

[0125] In this embodiment, the first deviation threshold is, for example, 5 mm.

[0126] In this embodiment, a first proportional initial tensioning is performed: when the deviation between the real-time deflection value of the arch rib and the preset maximum deflection value reaches a first deviation threshold, the hanger is initially tensioned, and the tensioning force is executed according to a preset first proportion. This step aims to initially control the deformation of the arch rib by adjusting the hanger cable force; for example, the first proportion is 30%, that is, the hanger is tensioned to 30% of its design cable force.

[0127] In this embodiment, the second deviation threshold is, for example, 15 mm.

[0128] In this embodiment, graded tensioning from the first ratio to the second ratio is performed: when the deviation between the real-time deflection value of the arch rib and the preset maximum deflection value reaches the second deviation threshold, the suspension rod is tensioned further based on the initial tensioning already performed in the first ratio. The tensioning force is gradually increased from the first ratio to the second ratio. This graded approach is used to more precisely control the deformation of the arch rib and the distribution of internal forces in the structure. For example, the tensioning is gradually increased from 30% to 70%.

[0129] In this embodiment, maximum proportional tensioning is performed: if the real-time deflection value of the arch rib is still not effectively controlled after the initial tensioning and staged tensioning, and approaches or reaches the preset maximum deflection value, the suspender is tensioned to the maximum extent possible to adjust the arch rib deformation and bring it back to a safe range. The maximum proportion is usually 100% of the design cable force.

[0130] In this embodiment, the cable force corrections for all hangers calculated using the matrix influence method are readjusted. The matrix influence method is used to calculate the cable force corrections for all hangers, taking into account the mutual influence between hangers and their impact on the overall arch rib structure. Then, based on the calculated cable force corrections, the hanger cable forces are adjusted again to further optimize the stress state and deformation of the arch rib, ensuring the bridge structure meets design requirements. For example, if the matrix influence method determines that a certain hanger's cable force needs to be increased by 5kN, a readjustment operation is performed on it.

[0131] To calculate the cable force correction for all suspenders, a matrix influence method is proposed for calculating the cable force correction for all suspenders, including:

[0132] An influence matrix is ​​constructed using the hanger numbers of the slanted steel box arch bridge as row vectors and the key control sections of the arch ribs as column vectors.

[0133] A basic deviation vector is generated based on the deviation between the real-time deflection values ​​of all key control sections of the arch ribs and the preset installation accuracy threshold.

[0134] The base deviation vector is corrected based on the preset wind-induced correction coefficient and temperature correction coefficient to obtain the corrected deviation vector;

[0135] The multidimensional cable force correction vector is solved by regularized least squares method based on the influence matrix and the correction deviation vector;

[0136] The cable force correction vector is filtered by safety constraints, equilibrium constraints, and construction feasibility constraints to obtain the cable force correction amount for all suspenders.

[0137] In this embodiment, an influence matrix is ​​constructed using the hanger numbers of the slanted steel box arch bridge (e.g., 12 pairs of hangers for the entire bridge, numbered G1-G12) as row vectors and the key control sections of the arch rib (2 at the arch foot, 2 at the 1 / 4 span, 1 at the arch crown, and 1 at the 3 / 4 span, for a total of 6 sections) as column vectors. In the analysis of the slanted steel box arch bridge, each hanger number is arranged sequentially to form rows, and the key sections on the arch rib used to control the structural state are used as columns, forming a matrix. The elements of this matrix reflect the influence relationship of each hanger on each key control section of the arch rib. For example, if the influence of hanger 1 on key control section 1 of the arch rib is 0.5, then 0.5 is recorded at the corresponding position in the matrix for subsequent analysis of the effect of hanger cable force changes on different sections of the arch rib.

[0138] The matrix element M(i,j) represents the vertical deflection change (mm) at the j-th control section when the i-th suspender is tensioned by a unit force (100kN). Its value is determined as follows:

[0139] A refined finite element model was established using ANSYS to simulate the structural response under different load conditions (including the coupling effect of wind load and hoisting off-center load), and the baseline value of M(i,j) was initially calculated.

[0140] For the hangers (such as G1-G4) on the skew side (asymmetric stress side), considering the additional effect of the lateral offset of the arch rib, their M(i,j) value is multiplied by a correction factor of 1.2 (because skew-span structures are prone to asymmetric deflection).

[0141] In this embodiment, a basic deviation vector is generated based on the deviation between the real-time deflection values ​​of all key control sections of the arch rib and a preset installation accuracy threshold: the real-time deflection value of each control section of the arch rib is measured and compared with the preset installation accuracy threshold, and the deviation value of each control section is calculated. These deviation values ​​are arranged in sequence to form a vector. For example, the deviation of control section 1 is 2mm, the deviation of control section 2 is -3mm, etc., forming a basic deviation vector [2, -3, ...], which reflects the deviation of each control section from the ideal installation state.

[0142] In this embodiment, the preset wind correction factor is: for example, when v = 10 m / s, the preset wind correction factor is 1.25;

[0143] Introducing a three-dimensional coupling factor of "wind speed-wind direction-section location":

[0144] When the angle θ between the wind direction and the arch rib axis is less than or equal to 30°, the wind-induced correction factor is preset to be 1 + 0.03v (v is the wind speed, in m / s).

[0145] When 30°<θ≤60°, the preset wind-induced correction coefficient is 1+0.02v×sinθ (considering the decomposition effect of crosswinds);

[0146] When θ > 60°, the preset wind-induced correction factor = 1 + 0.015v × cosθ (wind load is mainly axial force, and its influence on deflection is weakened);

[0147] For the windward and leeward sections of the hexagonal arch rib (such as the left 1 / 4 span and the right 1 / 4 span), the wind-induced correction coefficients are preset and multiplied by the asymmetry coefficients of 1.2 and 0.8 respectively (considering the uneven distribution of wind loads).

[0148] In this embodiment, a temperature correction factor is used to correct the impact of temperature changes on structural deformation, as temperature variations cause the bridge's structural materials to expand and contract. For example, the correction ratio for structural deformation is set to 0.005 for every 1°C change in temperature. This temperature correction factor varies depending on the characteristics of the structural materials and ambient temperature.

[0149] Arrange 16 temperature sensors (one every 5m along the longitudinal direction of the arch rib, and one on each of the six circumferential surfaces). Calculate:

[0150] Longitudinal temperature gradient ΔT_longitudinal (temperature difference between arch foot and arch crown): When ΔT_longitudinal > 12℃, k_temperature_longitudinal = 1 + 0.02 × ΔT_longitudinal (affects overall deflection);

[0151] Circumferential temperature gradient ΔT_ring (maximum temperature difference at the same cross section): When ΔT_ring > 8℃, k_temperature_ring = 1 + 0.01 × ΔT_ring (affects local bending, only corrects lateral deviation).

[0152] In this embodiment, the basic deviation vector is corrected based on preset wind-induced correction coefficients and temperature correction coefficients to obtain a corrected deviation vector: each element in the basic deviation vector is multiplied by the preset wind-induced correction coefficient and temperature correction coefficient, respectively, taking into account the influence of wind load and temperature on the deviation, to obtain the corrected deviation vector. For example, if the basic deviation vector element is 5, the preset wind-induced correction coefficient is 0.8, and the temperature correction coefficient is 0.9, then after correction, this element becomes 5 × 0.8 × 0.9 = 3.6. All elements after correction form the corrected deviation vector. The corrected deviation vector Δδcorrected = basic deviation vector × preset wind-induced correction coefficient × ktemperature longitudinal × ktemperature ring, realizing refined correction of the spatiotemporal distribution of the temperature field.

[0153] In this embodiment, a multidimensional cable force correction vector is obtained using regularized least squares based on the influence matrix and the correction deviation vector: A multidimensional vector is calculated using the constructed influence matrix and correction deviation vector, employing the Tikhonov regularized least squares method. Each dimension of this multidimensional vector corresponds to a correction value for different suspender cable forces. This minimizes the structural deformation deviation after considering various influencing factors, thereby determining how the cable forces of each suspender need to be adjusted.

[0154] In this embodiment, the multidimensional cable force correction vector is a vector calculated as described above, where each element represents the correction amount for different hanger cable forces. It reflects, from multiple dimensions, the numerical changes that the cable forces of each hanger should undergo to bring the structure back to its ideal state. For example, the multidimensional cable force correction vector [10, -5, 8, ...] indicates that the cable force of the first hanger needs to be increased by 10 kN, and the cable force of the second hanger needs to be reduced by 5 kN, etc.

[0155] In this embodiment, the multidimensional cable force correction vector is filtered by safety constraints, equilibrium constraints, and construction feasibility constraints to obtain the cable force correction amount for all hangers. For each element in the multidimensional cable force correction vector, the cable force correction amount for all hangers is obtained by sequentially filtering and adjusting according to safety constraints (e.g., correction amount for a single hanger ≤ ±180kN (not exceeding 10% of the design cable force), additional limit for hangers on the slant side ≤ ±150kN (controlling arch foot stress ≤ 250MPa)), equilibrium constraints (to make the cable force distribution of each hanger more uniform and reasonable, such as the difference in correction amount between adjacent hangers ≤ 25kN to avoid sudden load changes), and construction feasibility constraints (the correction amount needs to be adapted to the jack tensioning accuracy (0.4 grade), rounded to a multiple of 5kN (e.g., the calculated value of 132kN is adjusted to 130kN)). Finally, the cable force correction amount for all hangers that meet various practical requirements is obtained. For example, if the cable tension correction of a certain suspender is found to exceed the safe range after being filtered by safety constraints, it needs to be adjusted to the safe range; after being filtered by balance constraints, it may be further fine-tuned to ensure that the force on each suspender is more balanced; finally, after being filtered by construction feasibility constraints, corrections that cannot be achieved by existing construction techniques are eliminated, and the final cable tension correction is obtained.

[0156] To determine the required strength of the equilibrium constraints, an equilibrium constraint filtering process is proposed for the multidimensional cable force correction vector. This process includes:

[0157] A finite element model of the construction process of the skew-span steel box arch bridge is generated. In the finite element model, a force flow monitoring section is selected at a preset distance along the longitudinal direction of the arch rib. Multiple feature points are selected in each force flow monitoring section. The axial normal stress direction, radial normal stress direction, and shear stress direction of each feature point are extracted. The force flow direction angle of each feature point is calculated based on the axial normal stress direction, radial normal stress direction, and shear stress direction of each feature point.

[0158] Continuous sections with the same structural or stress characteristics in the finite element model are treated as sections to be analyzed.

[0159] Based on the force flow direction angle of all feature points in the finite element model, the force flow abrupt change section is selected from all sections to be analyzed.

[0160] The strength requirements of the equilibrium constraints are determined based on the distribution location and degree of abrupt changes in all force flow segments.

[0161] The multidimensional cable force correction vector is filtered based on the strength requirements of the equilibrium constraint.

[0162] In this embodiment, a finite element model of the sloping steel box arch bridge during construction is generated: using specialized finite element analysis software, a digital model is constructed based on the bridge's design drawings, material properties, construction techniques, and other information to simulate its mechanical behavior at each stage of construction. This model discretizes the bridge structure into numerous finite elements, and analyzes the overall stress and deformation of the bridge by calculating the interactions between these elements. For example, structural components such as the arch ribs, hangers, and bridge deck are simulated using appropriate element types, and corresponding material parameters and boundary conditions are assigned to obtain a finite element model suitable for analysis.

[0163] In this embodiment, the preset distance is, for example, set to 0.5 meters.

[0164] In this embodiment, the force flow monitoring section is a specific cross section selected in the finite element model of the skew-span steel box arch bridge. These sections are used to monitor the transmission and distribution of forces in the structure.

[0165] In this embodiment, multiple feature points are selected for each force flow monitoring section: on the already determined force flow monitoring section, multiple representative points are selected, and the mechanical parameters of these points can reflect the force characteristics of the section. For example, on a circular force flow monitoring section, edge points, center points, and points distributed at a certain angle are selected as feature points in order to comprehensively obtain force information on the section.

[0166] In this embodiment, the axial normal stress direction, radial normal stress direction, and shear stress direction of each feature point are extracted: For each selected feature point, the directions of the normal stress in the axial direction (along the length of the component), the radial direction (perpendicular to the length of the component), and the shear stress are obtained through the calculation results of the finite element model. The axial normal stress direction of the feature point is to the left along the longitudinal direction of the bridge, the radial normal stress direction is perpendicular to the cross-section and points inward, and the shear stress direction forms a certain angle with a certain coordinate axis.

[0167] In this embodiment, the force flow direction angle of each feature point is calculated based on the axial normal stress direction, radial normal stress direction, and shear stress direction. Using relevant principles and formulas of mechanics of materials, and combining these directions, an angle value is calculated. This angle represents the direction of force flow at the feature point and is called the force flow direction angle. The force flow direction angle can intuitively demonstrate the force transmission trend at the feature point. For example, the force flow direction angle of a feature point can be obtained by dividing the arctangent of the ratio of twice the shear stress to the difference between the axial and radial normal stresses by 2.

[0168] In this embodiment, continuous sections with the same structural or stress characteristics in the finite element model are considered as the sections to be analyzed. In the finite element model of a skew-span steel box arch bridge, the structure is divided, and those parts that are consistent and continuous in structural form (such as all straight arch ribs, hangers with the same cross-sectional shape, etc.) or stress characteristics (such as all subjected to axial tension, similar bending moment distribution, etc.) are considered as a whole for analysis. This whole is the section to be analyzed. For example, an arch rib section with an unchanged cross-sectional shape and mainly subjected to axial pressure within a certain length is defined as a section to be analyzed.

[0169] In this embodiment, the force flow abrupt change zone is defined as a region within the analyzed section where the force flow direction angle changes drastically over a short distance. Such abrupt changes in the force transmission path can lead to localized stress concentration in the structure, posing a potential threat to structural safety. For example, at the connection between the arch rib and the pier, the force flow direction may abruptly change from along the arch rib direction to the pier direction; this region is thus a force flow abrupt change zone.

[0170] In this embodiment, the strength requirement of the equilibrium constraint is determined based on the distribution location and degree of abrupt change of all force flow segments: The required strength for equilibrium constraint on the structure is determined by comprehensively considering the location of all force flow segments with abrupt changes in the finite element model and the severity of the force flow direction angle change. For example:

[0171] If the section with a rapid change in force flow is concentrated in the steel-concrete composite section at the arch foot and the rate of change of direction angle reaches 10° / m, the constraint strength of the force difference between adjacent hangers needs to be increased to ≤3% to suppress abrupt changes;

[0172] When a slight abrupt change occurs in 1 / 4 of the span (change rate 6° / m), the constraint strength can be appropriately relaxed to a force difference ≤5%;

[0173] If the arch crown curve section has no sharp turns, it is only necessary to maintain the conventional constraint strength (force difference ≤ 8%).

[0174] If the section with a rapid change in force flow is concentrated in the steel-concrete composite section at the arch foot and the rate of change of direction angle reaches 10° / m, the constraint strength of the force difference between adjacent hangers needs to be increased to ≤3% to suppress abrupt changes;

[0175] When a slight abrupt change occurs in 1 / 4 of the span (change rate 6° / m), the constraint strength can be appropriately relaxed to a force difference ≤5%;

[0176] If the arch crown curve section has no sharp turns, it is only necessary to maintain the conventional constraint strength (force difference ≤ 8%).

[0177] In this embodiment, the multidimensional cable force correction vector is filtered based on the strength requirements of the equilibrium constraints: each element in the multidimensional cable force correction vector is screened and adjusted according to the previously determined equilibrium constraint strength requirements. This ensures that the corrected cable force distribution can make the structure more uniformly stressed in and around the abrupt force flow transition section, avoiding further deterioration of the force flow distribution due to unreasonable cable forces. For example, if the cable force correction of a certain hanger would exacerbate the stress concentration in the abrupt force flow transition section, and the equilibrium constraint strength requirement is high, then the cable force correction needs to be adjusted to meet the requirement of uniform stress distribution in the structure.

[0178] To identify abrupt force-flow transition zones within the analyzed region, a method based on the force-flow direction angles of all feature points in the finite element model is proposed for selecting such zones. This includes:

[0179] The ratio of the force flow direction angle difference of the feature points at the same position in each group of adjacent force flow monitoring sections to the distance between adjacent sections is taken as the corresponding direction angle change rate.

[0180] The standard deviation of the force flow direction angle of all feature points in the section to be analyzed is defined as the force flow continuity index.

[0181] The ratio of all abrupt changes in orientation angles to all feature points in the section to be analyzed is taken as the proportion of extreme points.

[0182] Based on the rate of change of all directional angles, the force-flow continuity index, and the proportion of extreme points in each analysis segment, the force-flow abrupt change segments are selected from all analysis segments.

[0183] In this embodiment, the abrupt change point of the direction angle refers to the point in the finite element analysis of a skew-span steel box arch bridge where the force flow direction angle changes abruptly and significantly. In continuous structural sections, the force flow direction angle usually changes relatively smoothly, but at certain points, due to changes in structural form, abrupt changes in stress conditions, or other reasons, the direction angle may change drastically. For example, at the connection point between the arch rib and a special structural component, the force flow direction angle may change instantaneously from one direction to another direction with a significant difference; this point is the abrupt change point of the direction angle. It plays an important indicative role in judging anomalies in the transmission of internal forces and areas of local stress concentration within the structure.

[0184] In this embodiment, the abrupt change in force flow is selected from all segments to be analyzed based on the rate of change of all directional angles, the force flow continuity index, and the proportion of extreme points within each segment to be analyzed:

[0185] A higher rate of change means that the direction of force flow changes significantly over a short distance. The lower the force flow continuity index, the less smooth the transmission of force flow within that section, and there may be abrupt changes or concentrations of force. A higher proportion of extreme points means that the angle of force flow direction changes frequently and drastically within that section.

[0186] Screening for abrupt force-flow transitions: Considering the three indicators mentioned above, select segments from all the segments to be analyzed that have a large rate of change of direction angle, a low force-flow continuity index, and a high proportion of extreme points. These segments are considered abrupt force-flow transition segments. For example, if the rate of change of direction angle of a segment to be analyzed is higher than a set threshold, the force-flow continuity index is lower than a specific standard, and the proportion of extreme points exceeds a certain percentage, then it is identified as abrupt force-flow transition segment.

[0187] This invention provides an implementation method for a machine learning-based dynamic control system for the installation accuracy of a slant-span steel box arch bridge, comprising:

[0188] The data acquisition module is used to collect real-time data during the construction of the slanted steel box arch bridge. The real-time data includes real-time stress of the arch ribs, real-time deformation of the arch ribs, real-time ambient wind speed, and real-time hoisting parameters.

[0189] The model building module is used to build a predictive model based on a large amount of historical construction data that can analyze the coupling effect of wind load and hoisting off-center load.

[0190] The coupling analysis module is used to input real-time data into the prediction model to obtain real-time coupling effect analysis results;

[0191] The sequence optimization module is used to output hoisting sequence optimization instructions based on the decision engine and coupling effect analysis results, as well as the preset mechanical constraints of the skew-span steel box arch bridge.

[0192] The scheme generation module is used to generate cable force graded adjustment schemes based on the deviation between the real-time deformation of the arch rib and the preset installation accuracy threshold.

[0193] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for dynamic adjustment of installation accuracy of a skew-span steel box arch bridge based on machine learning, characterized in that, include: Real-time data was collected during the construction of the skew-span steel box arch bridge, including real-time stress of the arch ribs, real-time deformation of the arch ribs, real-time ambient wind speed, and real-time hoisting parameters. A predictive model capable of analyzing the coupling effect of wind load and hoisting off-center load was constructed based on a large amount of historical construction data. Real-time data is input into the prediction model to obtain real-time analysis results of coupling effects; Based on the analysis results of the decision engine and coupling effect, as well as the preset mechanical constraints of the skew-span steel box arch bridge, the hoisting sequence optimization command is output. Based on the deviation between the real-time deformation of the arch rib and the preset installation accuracy threshold, a graded adjustment scheme for cable force is generated. Among them, a predictive model based on a large amount of historical construction data is constructed to analyze the coupling effect of wind load and hoisting eccentric load, including: Simulate the combined working conditions of wind load and hoisting eccentric load from a large amount of historical construction data to obtain a preset set of sample data; All sample data are divided into training and validation sets according to a preset ratio. The neural network model is trained based on the training and validation sets. At the same time, the mean square error between the predicted deformation and the actual deformation is minimized by the Adam optimizer until the mean square error does not exceed the preset error threshold. Then, a prediction model that can analyze the coupling effect of wind load and hoisting off-center load is obtained. The process of building the decision engine also includes: The reward function for the reinforcement learning algorithm is generated based on the actual deformation, allowable deformation, and hoisting delay rate. A decision engine is built based on the reward function of the reinforcement learning algorithm; Based on the decision engine and coupling effect analysis results, as well as the preset mechanical constraints of the skew-span steel box arch bridge, the system outputs hoisting sequence optimization instructions, including: A lifting sequence thread is generated based on each optional lifting sequence; The reward value for each hoisting sequence thread is determined based on the reward function of the reinforcement learning algorithm in the decision engine, and hoisting sequence optimization instructions are generated based on the hoisting sequence thread corresponding to the maximum reward value.

2. The method for dynamic adjustment of installation accuracy of a skew-span steel box arch bridge based on machine learning according to claim 1, characterized in that, Real-time data was collected during the construction of the skew-span steel box arch bridge, including: The real-time axial and radial stresses of the arch ribs during the construction of the skew-span steel box arch bridge are collected by fiber optic grating sensors uniformly arranged along the longitudinal direction of the arch ribs. The three-dimensional coordinates of the arch rib during the construction of the sloping steel box arch bridge are collected in real time by laser trackers set on the piers on both sides of the arch rib, and the lateral bending deformation and axial deviation are calculated based on the real-time collected three-dimensional coordinates as the real-time deformation of the arch rib. The instantaneous wind speed and direction during the construction of the sloping steel box arch bridge are collected by wind speed sensors installed at the top of the arch ribs as real-time environmental wind speed. Real-time segmental physical parameters, real-time hoisting posture parameters, and real-time motion state parameters are collected during the construction of the skew-span steel box arch bridge as real-time hoisting parameters.

3. The method for dynamic adjustment of installation accuracy of a skew-span steel box arch bridge based on machine learning according to claim 1, characterized in that, Based on the deviation between the real-time deformation of the arch rib and the preset installation accuracy threshold, a graded adjustment scheme for cable force is generated, including: The real-time deflection value of the arch rib is determined based on the real-time deformation of the arch rib. Use the preset maximum preset deflection value as the preset installation accuracy threshold; When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold does not exceed the first deviation threshold, the first proportional initial tensioning will be executed as a cable force grade adjustment scheme. When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold exceeds the first deviation threshold but does not exceed the second deviation threshold, the first to second proportional graded tensioning will be performed as a graded cable force adjustment scheme. When the deviation between the real-time deflection value of the arch rib and the preset installation accuracy threshold exceeds the second deviation threshold, the maximum proportional tensioning and the cable force correction of all hangers calculated by the matrix influence method will be re-adjusted as the cable force grade adjustment scheme.

4. The method for dynamic adjustment of installation accuracy of a skew-span steel box arch bridge based on machine learning according to claim 3, characterized in that, The cable force correction for all suspenders was calculated based on the matrix influence method, including: An influence matrix is ​​constructed using the hanger numbers of the slanted steel box arch bridge as row vectors and the key control sections of the arch ribs as column vectors. A basic deviation vector is generated based on the deviation between the real-time deflection values ​​of all key control sections of the arch ribs and the preset installation accuracy threshold. The base deviation vector is corrected based on the preset wind-induced correction coefficient and temperature correction coefficient to obtain the corrected deviation vector; The multidimensional cable force correction vector is solved by regularized least squares method based on the influence matrix and the correction deviation vector; The cable force correction vector is filtered by safety constraints, equilibrium constraints, and construction feasibility constraints to obtain the cable force correction amount for all suspenders.

5. The method for dynamic adjustment of installation accuracy of a skew-span steel box arch bridge based on machine learning according to claim 4, characterized in that, Equilibrium constraint filtering is applied to the multidimensional cable force correction vector, including: A finite element model of the construction process of the skew-span steel box arch bridge is generated. In the finite element model, a force flow monitoring section is selected at a preset distance along the longitudinal direction of the arch rib. Multiple feature points are selected in each force flow monitoring section. The axial normal stress direction, radial normal stress direction, and shear stress direction of each feature point are extracted. The force flow direction angle of each feature point is calculated based on the axial normal stress direction, radial normal stress direction, and shear stress direction of each feature point. Continuous sections with the same structural or stress characteristics in the finite element model are treated as sections to be analyzed. Based on the force flow direction angle of all feature points in the finite element model, the force flow abrupt change section is selected from all sections to be analyzed. The strength requirements of the equilibrium constraints are determined based on the distribution location and degree of abrupt changes in all force flow segments. The multidimensional cable force correction vector is filtered based on the strength requirements of the equilibrium constraint.

6. The method for dynamic adjustment of installation accuracy of a skew-span steel box arch bridge based on machine learning according to claim 5, characterized in that, Based on the force flow direction angles of all feature points in the finite element model, abrupt force flow transition zones were selected from all sections to be analyzed, including: The ratio of the force flow direction angle difference of the feature points at the same position in each group of adjacent force flow monitoring sections to the distance between adjacent sections is taken as the corresponding direction angle change rate. The standard deviation of the force flow direction angle of all feature points in the section to be analyzed is defined as the force flow continuity index. The ratio of all abrupt changes in orientation angles to all feature points in the section to be analyzed is taken as the proportion of extreme points. Based on the rate of change of all directional angles, the force-flow continuity index, and the proportion of extreme points in each analysis segment, the force-flow abrupt change segments are selected from all analysis segments.

7. A machine learning-based dynamic control system for the installation accuracy of a skew-span steel box arch bridge, characterized in that, The method for dynamically adjusting the installation accuracy of a skew-span steel box arch bridge based on machine learning, as described in any one of claims 1 to 6, comprises: The data acquisition module is used to collect real-time data during the construction of the slanted steel box arch bridge. The real-time data includes real-time stress of the arch ribs, real-time deformation of the arch ribs, real-time ambient wind speed, and real-time hoisting parameters. The model building module is used to build a predictive model based on a large amount of historical construction data that can analyze the coupling effect of wind load and hoisting off-center load. The coupling analysis module is used to input real-time data into the prediction model to obtain real-time coupling effect analysis results; The sequence optimization module is used to output hoisting sequence optimization instructions based on the decision engine and coupling effect analysis results, as well as the preset mechanical constraints of the skew-span steel box arch bridge. The scheme generation module is used to generate cable force graded adjustment schemes based on the deviation between the real-time deformation of the arch rib and the preset installation accuracy threshold.

Images