Three-dimensional visual construction management system and method

Abstract

The invention discloses a three-dimensional visual construction management system and method, and the method comprises the following steps: S1, obtaining construction project information, and constructing a three-dimensional digital modeling environment of a construction region; s2, arranging a sensor network at a construction site, and forming a dynamic load data set; s3, generating a preprocessed dynamic load data set; s4, constructing a gating circulation unit network prediction model by using the preprocessed dynamic load data set, wherein the gating circulation unit network prediction model is used for simulating and predicting the dynamic response of a construction structure under the action of an earthquake; s5, performing global search and optimization on the key hyper-parameters of the gating circulation unit network prediction model by using a grey wolf optimization algorithm to obtain an optimized gating circulation unit network prediction model; and S6, based on a three-dimensional visualization display result, carrying out risk grade division and early warning processing on the dynamic response of the construction structure, and generating construction management decision support information. According to the invention, a feasible intelligent management means is provided for engineering construction in the earthquake region.

Description

Technical Field

[0001] The present invention relates to the technical field of construction management, and in particular to a three-dimensional visual construction management system and method. Background Art

[0002] With the rapid advancement of urban infrastructure construction, large-scale construction projects are facing increasingly complex structural design requirements and changing construction environment factors. In earthquake-prone areas, how to ensure the dynamic stability, safety and response control capabilities of the structure during construction has become a key issue in project management.

[0003] In existing technologies, structural seismic response analysis mainly relies on numerical simulation methods or static safety margin calculations. The core algorithms are mostly linear modeling, static bearing capacity assessment, or simplified dynamic analysis under preset conditions, which are difficult to reflect the nonlinear evolution of the structural stress state during the actual construction process. At the same time, due to the model's insufficient adaptability to changes in input data, existing methods are difficult to process multi-source time series data such as wind load, seismic intensity, and structural acceleration response collected in real time, resulting in practical problems in construction management such as response prediction lag, inaccurate identification of stress concentration areas, and weak risk warning mechanisms. In addition, some existing studies have attempted to introduce neural networks or optimization algorithms, but most of them remain at the structural modeling level and fail to integrate with real-time data. They also lack a linkage mechanism with a three-dimensional visualization system, making it impossible to form a "prediction-display-decision-making" closed-loop management process.

[0004] In terms of risk assessment and early warning processing of structural response prediction results, the currently commonly used rule system is based on empirical thresholds or manually set levels. This not only lacks adaptability, but also makes it difficult to conduct refined management based on the response characteristics under different geographical environments, structural configurations and dynamic load conditions. As a result, the timeliness and accuracy of the early warning results are poor, affecting the risk prevention and control efficiency and structural safety assurance capabilities during construction.

[0005] In summary, there is an urgent need for a comprehensive technical approach that can integrate intelligent prediction algorithms, three-dimensional modeling visualization, and construction risk identification to solve the core problems of untimely response prediction, inaccurate risk judgment, and incomplete decision support in current construction management. Summary of the Invention

[0006] One purpose of the present invention is to propose a three-dimensional visual construction management system and method, which provides a practical and feasible intelligent management means for engineering construction in earthquake zones.

[0007] A three-dimensional visualization construction management method according to an embodiment of the present invention includes the following steps:

[0008] S1. Obtain construction project information and build a 3D digital modeling environment for the construction area;

[0009] S2. Deploy a sensor network at the construction site to collect real-time wind load data, seismic intensity data, and structural acceleration response data to form a dynamic load data set;

[0010] S3. Perform unified data cleaning, noise filtering, missing value filling, and time series reconstruction on the dynamic load dataset to generate a preprocessed dynamic load dataset;

[0011] S4. Utilize the preprocessed dynamic load dataset to construct a gated cyclic unit network prediction model to simulate and predict the dynamic response of the construction structure under earthquake action;

[0012] S5. Apply the Gray Wolf Optimization Algorithm to perform a global search and optimization of the key hyperparameters of the GRU network prediction model, obtaining an optimized GRU network prediction model.

[0013] S6. Map the structural response prediction results output by the optimized gated cyclic unit network prediction model into a three-dimensional digital modeling environment to achieve real-time dynamic simulation and three-dimensional visualization of the construction structure under earthquake action. Based on the three-dimensional visualization results, the dynamic response of the construction structure is classified into risk levels and early warning processing is performed to generate construction management decision support information.

[0014] Optionally, S1 includes the following steps:

[0015] S11. Obtaining three-dimensional structural information of the construction project, using a three-dimensional laser scanning device or an unmanned aerial vehicle remote sensing system to collect point cloud data of the three-dimensional structural shape of the construction project to form a point cloud set P;

[0016] S12. Obtain geographic environmental information of the construction area, including topographic information, geological structure information, and surface cover type, and construct an environmental dataset G consisting of multiple geographic information units;

[0017] S13. Based on the point cloud set P and the environment dataset G, digital modeling is used to perform multi-source data registration and fusion to construct a 3D digital modeling environment M. d , 3D digital modeling environment M d Represents the mapping relationship M through the function d =f(P,G) to obtain the integrated 3D model of building structure and geographical environment, where function f is the spatial matching and feature reconstruction mapping between point cloud data and geographic information units;

[0018] S14. The constructed 3D digital modeling environment M dTopological structure organization and mesh division are carried out to form a computable geometric model for response simulation and visual rendering, so that the three-dimensional digital modeling environment has the structural characteristics of node definition, boundary closure and attribute binding.

[0019] Optionally, S2 includes the following steps:

[0020] S21. Install wind load monitoring sensors at key locations on the construction structure. These sensors collect real-time data on wind loads generated by wind speed, direction, and impact area.

[0021] S22. Deploy seismic monitoring sensors in the construction area to collect real-time seismic intensity information, which is represented by horizontal and vertical acceleration components of the ground surface, to form seismic intensity data.

[0022] S23. Installing a structural response acceleration sensor within the construction structure to monitor the acceleration response of the structure under earthquake and wind loads, thereby generating structural acceleration response data.

[0023] S24. Synchronize the wind load data, earthquake intensity data, and structural acceleration response data in time and format, and fuse them to generate a complete dynamic load data set D load :

[0024]

[0025] Where t represents the unified time index in the dynamic load data set, and v is the alignment result of the time axis of various sensor data. t Indicates the wind speed at time point t, which is used to reflect the wind load intensity, θ t Indicates the wind direction angle at time point t, which is used to determine the direction of wind load. t Indicates the wind action area at time point t, which is used to correspond to the size of the force area of the wind load on the structure. They represent the components of the earthquake acceleration in the X, Y, and Z directions at time point t, is the structural acceleration response data of the pth monitoring point at time t, and T is the unified valid time series set.

[0026] Wind speed (v t ) is used to reflect the wind flow speed per unit time and is the key physical quantity that determines the intensity of wind load;

[0027] Wind direction angle (θ t ) is used to determine the direction of wind load and is the main parameter of wind vector direction;

[0028] Wind action area (A t) represents the effective area of the structure surface exposed to the wind at that point in time and is an important spatial factor affecting the total wind force.

[0029] The three jointly describe the dynamic characteristics of wind load at a certain point in time, and are collectively referred to as wind load action data.

[0030] At the same time, the earthquake acceleration component They represent the surface acceleration response of the structure in the X, Y, and Z spatial directions under earthquake action, respectively, and constitute the seismic intensity data. They are used to characterize the instantaneous inertial impact of seismic wave input on the structure and are one of the basic excitation inputs for structural response prediction.

[0031] Optionally, S3 includes the following steps:

[0032] S31. Dynamic load data set D load Perform data integrity checks and remove data records with duplicate timestamps, missing fields, or illegal values to form a preliminary cleansed dynamic load data set;

[0033] S32. Smooth the high-frequency noise data in the preliminary cleaned dynamic load dataset, set the window size, and perform noise smoothing on any numerical field at each time point t. Interpolate and fill any missing values in the smoothed data at time points t. Reconstruct the dynamic load dataset after missing value filling to a unified time series, remap all fields to a standard time axis, and perform linear resampling on any misaligned data points to form a time series dynamic load dataset.

[0034] S35. Normalize all fields contained in the time series dynamic load dataset and convert any field into a normalized value to obtain the preprocessed dynamic load dataset D pre .

[0035] Optionally, S4 includes the following steps:

[0036] S41. Using the pre-processed dynamic load data set D pre As input data, a gated recurrent unit network prediction model is constructed to predict the dynamic response of construction structures under earthquake action:

[0037]

[0038] Among them, h t represents the output hidden state of the gated recurrent unit network prediction model at time point t, which is used to describe the internal response state of the construction structure when subjected to the external earthquake load at time point t. t-1 represents the output hidden state of the gated recurrent unit network prediction model at time point t-1, z tTo update the gate control factor, represents the candidate response state at the current time step, ° represents the Hadamard element-wise product operator;

[0039] S42. Using the gated recurrent unit network prediction model, the pre-processed dynamic load data set is trained using a sliding window method. Based on the current moment and historical response, the response state of the structure at multiple future time steps is predicted, and the structural dynamic response prediction sequence H is output. pred ;

[0040] S43. In the process of constructing the input of the gated recurrent unit network prediction model, considering the difference in the importance of different types of load inputs to the structural response in the construction structure, a multi-dimensional dynamic response attention mechanism is introduced to calculate the input feature x at each time point t. t Each characteristic component Dynamically assign attention weights to construct attention-weighted input features

[0041]

[0042] in, Represents the feature component of the input feature in the i-th feature dimension, represents the attention weight of the input feature on the structural response at time point t, d is the total number of dimensions of the input feature, represents the impact strength score of the i-th input feature on the hidden state of the structural response at time point t, represents the impact strength score of the p-th input feature on the hidden state of the structural response at time point t, W e 、b e are the weight matrix and bias vector of relevance attention respectively;

[0043] S44. Input features after attention weighting Replace the original input feature x t Input to the gated recurrent unit network prediction model, all the updates to the gate control factor z t , candidate response status The calculations are based on Optimize the response capability and prediction sensitivity of the gated cyclic unit network prediction model to sudden seismic motion or severe wind loads.

[0044] Optionally, S5 includes the following steps:

[0045] S51. Construct the gray wolf optimized parameter space of the key hyperparameter Θ involved in the gated recurrent unit network prediction model, and map each set of parameter combinations to the position X of the gray wolf population individual. uand input the position into the improved Grey Wolf optimization algorithm driven by three-dimensional structural response;

[0046] S52. In each iteration, calculate the structural response prediction error corresponding to the u-th gray wolf individual and define the structural response sensitivity factor SRSF u =(s u,1 ,s u,2 ,…,s u,d ):

[0047]

[0048] Among them, s u,j represents the structural response sensitivity factor corresponding to the jth key hyperparameter dimension of the uth gray wolf individual, θ u,j represents the value of the u-th gray wolf in the j-th key hyperparameter dimension, f(X u ) is the structural response prediction error function, which is used to characterize the influence of different parameter disturbances on the prediction accuracy;

[0049] S53. Align the predicted structural response results with the actual structural response values in space, construct a spatiotemporal structural error distribution function, and map the spatiotemporal structural error distribution function back to the 3D digital modeling environment M. d The spatial structure unit in the 3D structure is constructed to construct the spatial mapping factor SMF (x, y, z) to reflect the response prediction error contribution of different regions of the 3D structure and define the response visualization coupling fitness function f'(X u ):

[0050]

[0051] Where, λ is the spatial error weighting coefficient;

[0052] S54. Based on the error decreasing trend in the gray wolf optimization iteration, define the gradient-guided dynamic step size factor γ o and the structural response sensitivity factor SRSF u , spatial mapping factor SMF and gradient-guided dynamic step factor γ o Integrate it into the gray wolf position update to build a response-driven improved gray wolf update mechanism:

[0053]

[0054] Among them, X u (d+1) is the position vector of the u-th gray wolf individual at the d+1th iteration, representing the updated key hyperparameter combination, X u (d) is the position vector of the u-th gray wolf individual at the d-th iteration, representing the current key hyperparameter combination, X hThey represent the positions of the current optimal, suboptimal, and third-optimal individuals, respectively. h = α, β, δ are the three types of leader individuals in the gray wolf algorithm, which are the individual indices of the global optimal α, suboptimal β, and third-optimal δ, respectively. w h is the corresponding guidance weight;

[0055] S55. Iterate and update S51-S54 until convergence to obtain the optimal set of key hyperparameters Θ opt , apply the optimal set of key hyperparameters to the gated recurrent unit network prediction model, and construct the optimized gated recurrent unit network prediction model GRU opt .

[0056] Optionally, S6 includes the following steps:

[0057] S61. The optimized gated recurrent unit network prediction model GRU opt The output structural response prediction sequence H pred Match to 3D digital modeling environment M by time index d The corresponding structural unit in is used to construct the structural response mapping function R(x, y, z, t), which represents the predicted response intensity at the three-dimensional coordinate point (x, y, z) at time t;

[0058] S62. Perform color coding and deformation-driven conversion on the structural response mapping function R(x, y, z, t), convert the response intensity value into dynamic deformation form, vibration frequency visualization, and stress intensity heat map in the three-dimensional structural model, and dynamically render it in conjunction with the time axis to generate a three-dimensional visualization animation of the structural response;

[0059] S63. Construct a response threshold classification system, set structural response intensity classification intervals, and classify the risk level of the response prediction values of each unit in the three-dimensional structural model;

[0060] S64. Based on the 3D visualization animation of structural responses and the risk classification results, construction management decision support information is generated, including a report on the overall safety status of the current construction structure, a spatial distribution map of the risk levels of each structural unit, visual early warning prompts and response trend predictions for key nodes, as well as corresponding construction recommendations and scheduling optimization strategies.

[0061] Optionally, the risk level classification rules of the response threshold classification system are as follows:

[0062] Safe area: The value of the structural response mapping function is less than or equal to the risk classification critical value θ1 of the structural response strength, and the unit structural deformation and unit stress are both lower than 40% of the structural bearing standard. The corresponding structural unit is classified as a safe area;

[0063] Area of concern: The value of the structural response mapping function is greater than θ1 and less than or equal to the risk classification critical value of the structural response intensity θ2, and the unit structural deformation and unit stress are between 40% and 70% of the structural bearing standard. The corresponding structural unit is classified as an area of concern;

[0064] Warning area: When the value of the structural response mapping function is greater than θ2 and less than or equal to the risk classification critical value θ3 of the structural response strength, and the unit structural deformation and unit stress are between 70% and 90% of the structural bearing standard, the corresponding structural unit is classified as a warning area, and the system automatically prompts to suspend construction and generates response optimization suggestions;

[0065] Dangerous area: When the value of the structural response mapping function is greater than the risk classification critical value θ3 of the structural response strength, and the unit structural deformation and unit stress exceed 90% of the structural bearing standard, the corresponding structural unit is classified as a dangerous area. The system automatically triggers the early warning mechanism and recommends the implementation of emergency control and construction suspension measures.

[0066] A three-dimensional visualization construction management system is applied to a three-dimensional visualization construction management method, comprising:

[0067] The data acquisition module is used to obtain construction project information, build a three-dimensional digital modeling environment for the construction area, and collect real-time data including wind load, earthquake intensity and structural acceleration response to form a dynamic load data set;

[0068] A data preprocessing module is used to preprocess the dynamic load data set to generate a preprocessed dynamic load data set;

[0069] Structural response prediction module, which is used to construct a gated recurrent unit network prediction model using the preprocessed dynamic load data set;

[0070] The parameter optimization module is used to apply the Gray Wolf Optimization Algorithm to perform global search and optimization on key hyperparameters in the GRU network prediction model to obtain the optimized GRU network prediction model.

[0071] A 3D visualization simulation module is used to map the structural response prediction results output by the optimized gated recurrent unit network prediction model into a 3D digital modeling environment to achieve 3D visualization of the structural response;

[0072] The risk identification and decision support module is used to analyze the structural response mapping function of each structural unit based on the three-dimensional visualization display results, divide the risk level according to the relative relationship between the response intensity and the structural bearing standard, and generate the risk level assessment results of the construction structure.

[0073] The beneficial effects of the present invention are:

[0074] (1) The present invention introduces a structural response sensitivity factor in the gray wolf optimization process, dynamically weights the gradient sensitivity of each parameter dimension according to the structural response prediction error function, effectively guides the search direction to converge to the key hyperparameter dimension that has the greatest impact on the prediction accuracy, strengthens the responsiveness control of the update gate, reset gate, and candidate state core structure, and improves the adaptability and optimization efficiency of the model under nonlinear seismic response conditions.

[0075] (2) The present invention spatially maps and aligns the multidimensional time series response results output by the gated recurrent unit network with the structural units in the three-dimensional digital modeling environment by defining a structural response mapping function, and constructs real-time deformation rendering, thermal map display and dynamic animation based on stress intensity and deformation amplitude response indicators, which significantly enhances the interpretability of the construction structure behavior, breaks through the limitations of traditional two-dimensional chart display methods, and realizes the full-time and full-domain visualization of the structure's response under seismic loads, thereby improving engineers' perception and judgment accuracy of complex construction conditions.

[0076] (3) Based on the structural response prediction results, the present invention integrates the three indicators of response amplitude, unit stress, and structural deformation to establish a refined classification standard including four risk levels, and links with the three-dimensional visualization system to realize dynamic risk classification, distribution map generation and management suggestion output at the structural unit level. It breaks through the extensive safety control method based on experience thresholds in traditional construction management. The system can automatically output construction scheduling suggestions and emergency response measures according to the risk level, greatly improving the response speed and adaptability of the construction site, and is superior to the existing system in terms of risk identification timeliness before and after earthquake disturbances. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0078] Figure 1 This is a flow chart of a three-dimensional visual construction management system and method proposed by the present invention. DETAILED DESCRIPTION

[0079] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0080] refer to Figure 1 , a 3D visual construction management method, comprising the following steps:

[0081] S1. Obtain construction project information and build a 3D digital modeling environment for the construction area;

[0082] S2. Deploy a sensor network at the construction site to collect real-time wind load data, seismic intensity data, and structural acceleration response data to form a dynamic load data set;

[0083] S3. Perform unified data cleaning, noise filtering, missing value filling, and time series reconstruction on the dynamic load dataset to generate a preprocessed dynamic load dataset;

[0084] S4. Utilize the preprocessed dynamic load dataset to construct a gated cyclic unit network prediction model to simulate and predict the dynamic response of the construction structure under earthquake action;

[0085] S5. Apply the Gray Wolf Optimization Algorithm to perform a global search and optimization of the key hyperparameters of the GRU network prediction model, obtaining an optimized GRU network prediction model.

[0086] S6. Map the structural response prediction results output by the optimized gated cyclic unit network prediction model into a three-dimensional digital modeling environment to achieve real-time dynamic simulation and three-dimensional visualization of the construction structure under earthquake action. Based on the three-dimensional visualization results, the dynamic response of the construction structure is classified into risk levels and early warning processing is performed to generate construction management decision support information.

[0087] In this embodiment, S1 includes the following steps:

[0088] S11. Obtaining three-dimensional structural information of the construction project, using a three-dimensional laser scanning device or an unmanned aerial vehicle remote sensing system to collect point cloud data of the three-dimensional structural shape of the construction project to form a point cloud set P;

[0089] S12. Obtain geographic environmental information of the construction area, including topographic information, geological structure information, and surface cover type, and construct an environmental dataset G consisting of multiple geographic information units;

[0090] S13. Based on the point cloud set P and the environment dataset G, digital modeling is used to perform multi-source data registration and fusion to construct a 3D digital modeling environment M. d , 3D digital modeling environment M d Represents the mapping relationship M through the function d =f(P,G) to obtain the integrated 3D model of building structure and geographical environment, where function f is the spatial matching and feature reconstruction mapping between point cloud data and geographic information units;

[0091] S14. The constructed 3D digital modeling environment M d Topological structure organization and mesh division are carried out to form a computable geometric model for response simulation and visual rendering, so that the three-dimensional digital modeling environment has the structural characteristics of node definition, boundary closure and attribute binding.

[0092] The formation of a computational geometry model for responsive simulation and visualization rendering specifically includes:

[0093] Topological structure organization: closing the boundaries of the structure contours generated by point cloud reconstruction, eliminating redundant points, and confirming the connection relationships to ensure that each three-dimensional structural unit has clear geometric boundaries, connection surface information, and topological adjacency relationships.

[0094] Node and edge definition: define the key points in the structural model as calculation nodes, and build an edge set between every two connected nodes for subsequent physical property binding and response path analysis.

[0095] Structural domain division and mesh generation, based on the differences in structural complexity and response characteristics, use the finite volume method or tetrahedron partitioning algorithm to divide the structure into multiple mesh units. Each unit is a finite substructure and has the characteristic of being able to be discretely solved.

[0096] Property binding: Each structural unit is bound to physical material properties, boundary constraints, and mechanical load transfer paths, providing basic parameters for subsequent response analysis models.

[0097] This implementation method spatially aligns and integrates point cloud data with geographic information data to construct a three-dimensional digital modeling environment with node definitions and attribute bindings, significantly improving the integrated expression capabilities between construction structures and geographic environments, and providing a highly realistic and computable spatial foundation for subsequent earthquake response simulation and visualization.

[0098] In this embodiment, S2 includes the following steps:

[0099] S21. Install wind load monitoring sensors at key locations on the construction structure. These sensors collect real-time data on wind loads generated by wind speed, direction, and impact area.

[0100] S22. Deploy seismic monitoring sensors in the construction area to collect real-time seismic intensity information, which is represented by horizontal and vertical acceleration components of the ground surface, to form seismic intensity data.

[0101] S23. Installing a structural response acceleration sensor within the construction structure to monitor the acceleration response of the structure under earthquake and wind loads, thereby generating structural acceleration response data.

[0102] S24. Synchronize the wind load data, earthquake intensity data, and structural acceleration response data in time and format, and fuse them to generate a complete dynamic load data set D load :

[0103]

[0104] Where t represents the unified time index in the dynamic load data set, and v is the alignment result of the time axis of various sensor data. t Indicates the wind speed at time point t, which is used to reflect the wind load intensity, θ t Indicates the wind direction angle at time point t, which is used to determine the direction of wind load. t Indicates the wind action area at time point t, which is used to correspond to the size of the force area of the wind load on the structure. They represent the components of the earthquake acceleration in the X, Y, and Z directions at time point t, is the structural acceleration response data of the pth monitoring point at time t, and T is the unified valid time series set.

[0105] This implementation method constructs a structural dynamic load dataset by unifying the time and format of three types of sensor data: wind load, seismic motion, and structural response. This provides high-quality input for time series modeling and response prediction, and improves the stability of the data-driven model and the perception accuracy of load state changes.

[0106] In this embodiment, S3 includes the following steps:

[0107] S31. Dynamic load data set D load Perform data integrity checks and remove data records with duplicate timestamps, missing fields, or illegal values to form a preliminary cleansed dynamic load data set;

[0108] S32. Smooth the high-frequency noise data in the preliminary cleaned dynamic load dataset, set the window size, and perform noise smoothing on any numerical field at each time point t. Interpolate and fill any missing values in the smoothed data at time points t. Reconstruct the dynamic load dataset after missing value filling to a unified time series, remap all fields to a standard time axis, and perform linear resampling on any misaligned data points to form a time series dynamic load dataset.

[0109] S35. Normalize all fields contained in the time series dynamic load dataset and convert any field into a normalized value to obtain the preprocessed dynamic load dataset D pre .

[0110] This implementation method performs integrity detection, missing complementation, and time series reconstruction on multi-source dynamic load data, and forms a structured input set through normalization preprocessing, which significantly reduces sensor data noise and redundant interference, and improves the training quality and generalization ability of the prediction model.

[0111] In this embodiment, S4 includes the following steps:

[0112] S41. Using the pre-processed dynamic load data set D pre As input data, a gated recurrent unit network prediction model is constructed to predict the dynamic response of construction structures under earthquake action:

[0113]

[0114] Among them, h t represents the output hidden state of the gated recurrent unit network prediction model at time point t, which is used to describe the internal response state of the construction structure when subjected to the external earthquake load at time point t. t-1 represents the output hidden state of the gated recurrent unit network prediction model at time point t-1, z t To update the gate control factor, represents the candidate response state at the current time step, Represents the Hadamard element-wise product operator;

[0115] S42. Using the gated recurrent unit network prediction model, the pre-processed dynamic load data set is trained using a sliding window method. Based on the current moment and historical response, the response state of the structure at multiple future time steps is predicted, and the structural dynamic response prediction sequence H is output. pred ;

[0116] S43. In the process of constructing the input of the gated recurrent unit network prediction model, considering the difference in the importance of different types of load inputs to the structural response in the construction structure, a multi-dimensional dynamic response attention mechanism is introduced to calculate the input feature x at each time point t. t Each characteristic component Dynamically assign attention weights to construct attention-weighted input features

[0117]

[0118] in, Represents the feature component of the input feature in the i-th feature dimension, represents the attention weight of the input feature on the structural response at time point t, d is the total number of dimensions of the input feature, represents the impact strength score of the i-th input feature on the hidden state of the structural response at time point t, represents the impact strength score of the p-th input feature on the hidden state of the structural response at time point t, W e 、b e are the weight matrix and bias vector of relevance attention respectively;

[0119] The formula's mechanism enables the model to automatically identify and amplify the most critical input features for earthquake response prediction, such as sudden changes in wind speed or Z-axis acceleration. Compared to raw input, attention-weighted input significantly improves the model's ability to perceive sudden, multi-source dynamic load characteristics, thereby improving response prediction accuracy.

[0120] S44. Input features after attention weighting Replace the original input feature x t Input to the gated recurrent unit network prediction model, all the updates to the gate control factor z t , candidate response status The calculations are based on Optimize the response capability and prediction sensitivity of the gated cyclic unit network prediction model to sudden seismic motion or severe wind loads.

[0121] This embodiment constructs a gated recurrent unit network prediction model with an attention mechanism. The system can assign response weights to different load types, improve the model's ability to identify key structural responses, and at the same time improve the model's prediction accuracy and controllability of structural responses under multi-time series dynamic loads.

[0122] In this embodiment, S5 includes the following steps:

[0123] S51. Construct the gray wolf optimized parameter space of the key hyperparameter Θ involved in the gated recurrent unit network prediction model, and map each set of parameter combinations to the position X of the gray wolf population individual. u and input the position into the improved Grey Wolf optimization algorithm driven by three-dimensional structural response;

[0124] S52. In each iteration, calculate the structural response prediction error corresponding to the u-th gray wolf individual and define the structural response sensitivity factor SRSF u =(s u,1 ,s u,2 ,…,s u,d ):

[0125]

[0126] Among them, s u,j represents the structural response sensitivity factor corresponding to the jth key hyperparameter dimension of the uth gray wolf individual, θ u,j represents the value of the u-th gray wolf in the j-th key hyperparameter dimension, f(X u ) is the structural response prediction error function, which is used to characterize the influence of different parameter disturbances on the prediction accuracy;

[0127] This formula quantifies the local variation trend of the model's prediction error with respect to each parameter dimension, giving the Grey Wolf algorithm a differentiated search orientation. Compared to traditional uniform weight update strategies, this method focuses more on key structural control parameters, improving convergence speed and the ability to escape local optima.

[0128] S53. Align the predicted structural response results with the actual structural response values in space, construct a spatiotemporal structural error distribution function, and map the spatiotemporal structural error distribution function back to the 3D digital modeling environment M. d The spatial structure unit in the 3D structure is constructed to construct the spatial mapping factor SMF (x, y, z) to reflect the response prediction error contribution of different regions of the 3D structure and define the response visualization coupling fitness function f'(X u ):

[0129]

[0130] Where, λ is the spatial error weighting coefficient;

[0131] The formula is an error reconstruction mechanism that introduces three-dimensional spatial response contributions. By binding the prediction error to the structural position mapping, it guides the algorithm to prioritize reducing the response uncertainty in areas with high error incidence during the optimization process. The mechanism strengthens the error-sensitive modeling of high-risk structural units, enabling the model to have more accurate response prediction capabilities in earthquake-prone areas.

[0132] S54. Based on the error decreasing trend in the gray wolf optimization iteration, define the gradient-guided dynamic step size factor γ o and the structural response sensitivity factor SRSF u , spatial mapping factor SMF and gradient-guided dynamic step factor γ o Integrate it into the gray wolf position update to build a response-driven improved gray wolf update mechanism:

[0133] X u (d+1)=X u (d)+γ o ·∑ h＝α,β,δ w h ·(X h -X u (d)) SRSF u ;

[0134] Among them, X u (d+1) is the position vector of the u-th gray wolf individual at the d+1th iteration, representing the updated key hyperparameter combination, X u (d) is the position vector of the u-th gray wolf individual at the d-th iteration, representing the current key hyperparameter combination, X hThey represent the positions of the current optimal, suboptimal, and third-optimal individuals, respectively. h = α, β, δ are the three types of leader individuals in the gray wolf algorithm, which are the individual indices of the global optimal α, suboptimal β, and third-optimal δ, respectively. w h is the corresponding guidance weight.

[0135] This update method significantly improves the model's stable convergence performance under nonlinear and complex loads and enhances its ability to control uncertainty in the structural response error space. In simulation experiments, predictive performance improved by 8.5% and convergence time was shortened by nearly 25% under the same dataset. The method demonstrates a superior balance between global exploration and local convergence in multi-objective optimization scenarios.

[0136] S55. Iterate and update S51-S54 until convergence to obtain the optimal set of key hyperparameters Θ opt , apply the optimal set of key hyperparameters to the gated recurrent unit network prediction model, and construct the optimized gated recurrent unit network prediction model GRU opt .

[0137] This implementation method introduces a structural response sensitivity factor in the gray wolf optimization process, dynamically weights the gradient sensitivity of each parameter dimension according to the structural response prediction error function, effectively guides the search direction to converge toward the key hyperparameter dimension that has the greatest impact on the prediction accuracy, strengthens the responsiveness control of the update gate, reset gate, and candidate state core structure, and improves the adaptability and optimization efficiency of the model under nonlinear seismic response conditions.

[0138] In this embodiment, S6 includes the following steps:

[0139] S61. The optimized gated recurrent unit network prediction model GRU opt The output structural response prediction sequence H pred Match to 3D digital modeling environment M by time index d The corresponding structural unit in is used to construct the structural response mapping function R(x, y, z, t), which represents the predicted response intensity at the three-dimensional coordinate point (x, y, z) at time t;

[0140] S62. Perform color coding and deformation-driven conversion on the structural response mapping function R(x, y, z, t), convert the response intensity value into dynamic deformation form, vibration frequency visualization, and stress intensity heat map in the three-dimensional structural model, and dynamically render it in conjunction with the time axis to generate a three-dimensional visualization animation of the structural response;

[0141] The three-dimensional visualization animation of the structural response is generated by expanding the continuous time frames of the structural response mapping function R(x, y, z, t), combining vertex displacement-driven mesh deformation with color-coded heat map expression, and adopting a frame rendering method based on timeline control to generate a structural response visualization animation that can be dynamically played over time.

[0142] S63. Construct a response threshold classification system, set structural response intensity classification intervals, and classify the risk level of the response prediction values of each unit in the three-dimensional structural model;

[0143] S64. Based on the 3D visualization animation of structural responses and the risk classification results, construction management decision support information is generated, including a report on the overall safety status of the current construction structure, a spatial distribution map of the risk levels of each structural unit, visual early warning prompts and response trend predictions for key nodes, as well as corresponding construction recommendations and scheduling optimization strategies.

[0144] This implementation defines a structural response mapping function to spatially align the multidimensional time series response results output by the gated recurrent unit network with the structural units in the three-dimensional digital modeling environment. It also constructs real-time deformation rendering, heat map display, and dynamic animation based on response indicators such as stress intensity and deformation amplitude. This significantly enhances the interpretability of the construction structure behavior, breaks through the limitations of traditional two-dimensional graphical display methods, and realizes the full-time and full-domain visualization of the structure's response under seismic loads, improving engineers' perception and judgment accuracy of complex construction conditions.

[0145] In this embodiment, the risk level classification rules of the response threshold classification system are as follows:

[0146] Safe area: The value of the structural response mapping function is less than or equal to the risk classification critical value θ1 of the structural response strength, and the unit structural deformation and unit stress are both lower than 40% of the structural bearing standard. The corresponding structural unit is classified as a safe area;

[0147] Area of concern: The value of the structural response mapping function is greater than θ1 and less than or equal to the risk classification critical value of the structural response intensity θ2, and the unit structural deformation and unit stress are between 40% and 70% of the structural bearing standard. The corresponding structural unit is classified as an area of concern;

[0148] Warning area: When the value of the structural response mapping function is greater than θ2 and less than or equal to the risk classification critical value θ3 of the structural response strength, and the unit structural deformation and unit stress are between 70% and 90% of the structural bearing standard, the corresponding structural unit is classified as a warning area, and the system automatically prompts to suspend construction and generates response optimization suggestions;

[0149] Dangerous area: When the value of the structural response mapping function is greater than the risk classification critical value θ3 of the structural response strength, and the unit structural deformation and unit stress exceed 90% of the structural bearing standard, the corresponding structural unit is classified as a dangerous area. The system automatically triggers the early warning mechanism and recommends the implementation of emergency control and construction suspension measures.

[0150] Based on the structural response prediction results, this implementation method integrates three indicators: response amplitude, unit stress, and structural deformation to establish a refined classification standard with four risk levels. It also works in conjunction with the three-dimensional visualization system to achieve dynamic risk classification, distribution map generation, and management suggestion output at the structural unit level. This breaks through the extensive safety control method based on empirical thresholds in traditional construction management. The system can automatically output construction scheduling suggestions and emergency response measures based on risk levels, greatly improving the response speed and adaptability of the construction site, and is superior to existing systems in terms of risk identification timeliness before and after earthquake disturbances.

[0151] A three-dimensional visualization construction management system is applied to a three-dimensional visualization construction management method, comprising:

[0152] The data acquisition module is used to obtain construction project information, build a three-dimensional digital modeling environment for the construction area, and collect real-time data including wind load, earthquake intensity and structural acceleration response to form a dynamic load data set;

[0153] A data preprocessing module is used to preprocess the dynamic load data set to generate a preprocessed dynamic load data set;

[0154] Structural response prediction module, which is used to construct a gated recurrent unit network prediction model using the preprocessed dynamic load data set;

[0155] The parameter optimization module is used to apply the Gray Wolf Optimization Algorithm to perform global search and optimization on key hyperparameters in the GRU network prediction model to obtain the optimized GRU network prediction model.

[0156] A 3D visualization simulation module is used to map the structural response prediction results output by the optimized gated recurrent unit network prediction model into a 3D digital modeling environment to achieve 3D visualization of the structural response;

[0157] The risk identification and decision support module is used to analyze the structural response mapping function of each structural unit based on the three-dimensional visualization display results, divide the risk level according to the relative relationship between the response intensity and the structural bearing standard, and generate the risk level assessment results of the construction structure.

[0158] Example 1:

[0159] In October 2024, a municipal traffic interchange project located in Ayi District entered the main structure construction phase. The project is located at the intersection of the basin seismic belt. The construction area has complex geology and is surrounded by densely populated areas and the intersection of main roads. The structural stability and safety requirements during the construction process are extremely high. The project structure design includes 4 cross-line bridges, 1 semi-underground box culvert-type passage and 4 high pier node structures. The construction period is planned to be 14 months. When the project entered the foundation construction to the main pier formwork stage, the construction management party adopted the present invention to continuously monitor, predict and manage the structure at different stages.

[0160] Forty-two multi-dimensional sensor nodes were deployed at the construction site, located at the main piers, bridge piers, reinforcement cages, steel structure installation sections, and foundation slabs. These sensors include seismic sensors (sampling at 24 Hz), multi-directional accelerometers (supporting X / Y / Z triaxial data acquisition), wind load direction finders, and concrete stress sensors. Data is collected every 10 seconds, 24 hours a day.

[0161] During the construction of the 3D digital modeling environment, the project team used drones equipped with LiDAR to collect construction structure point clouds. This was combined with construction drawings to generate a structural point cloud set. Furthermore, a 3D geographic information set was constructed using local topographic information. Data fusion and topological reconstruction were used to create a 3D visual modeling environment, which was then linked to actual construction numbers using structural units.

[0162] On the fifth day of construction, a minor earthquake measuring 3.4 magnitude struck the Chengdu area, with its epicenter 22 km from the construction site. The system captured seismic data in real time, extracting peak horizontal acceleration of 0.072g and peak vertical acceleration of 0.034g. The peak structural acceleration response occurred at main pier G-07, with a maximum response of 0.109g. This acceleration data was automatically linked to the structural mapping function of the construction model to generate a thermal deformation diagram, which was visualized on the construction management platform.

[0163] The system then initiated training for the gated cyclic unit network structural response prediction model, using earthquake monitoring data, environmental load data, and construction structure phase response data from the first three months of the project as training samples. The total sample size was 28,800 entries, including input fields such as wind speed, wind direction, earthquake horizontal and vertical acceleration, concrete temperature gradient, and structural deformation. The output fields were structural displacement response and stress response.

[0164] In this method, the key hyperparameters of the gated recurrent unit network are globally optimized using the Gray Wolf Optimization algorithm. During the optimization process, the optimization path is dynamically adjusted based on the structural response gradient sensitivity factor, ultimately converging to an error value of 0.0028 and achieving a model accuracy of 94.2%. Using the same data set, a traditional long-short-term memory neural network was used for training, achieving an optimal model prediction accuracy of 86.3% with 70 training rounds, while the method of this invention only required 43 rounds.

[0165] The 3D Structural Response Visualization module maps the predicted response results, dynamically displaying the structural response trends over the next three minutes, including deformation zone expansion, displacement path trends, and stress range. The system automatically marks nodes whose response strength exceeds the critical value as "warning zones," with G-07 and G-08 added to the warning list. The project construction management platform then automatically generates construction recommendations based on this system: delaying the concrete pour, optimizing steel support angles, and strengthening temporary support node connections to avoid secondary disaster risks.

[0166] In addition, based on the structural risk level assessment results generated by the system, the system was able to accurately identify 12 medium- and high-risk component units. In subsequent post-earthquake investigations and tests, 9 structural units actually had abnormal stress concentration and loose connections. The early warning identification accuracy rate reached 75%, which is significantly better than the traditional early warning method based on experience judgment.

[0167] In summary, this embodiment effectively verifies the accuracy of the present invention in predicting structural responses, the timeliness of risk assessment, and the intelligence of construction management decision support in large and complex construction projects in seismically active areas. It solves the problems of delayed response prediction and the inability to link visual display and early warning strategies in traditional methods, and provides a practical and feasible intelligent management method for engineering construction in earthquake-prone areas.

[0168] The above are only preferred specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with this technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solutions and inventive concepts of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A three-dimensional visual construction management method, characterized in that: The steps include: S1. Obtain construction project information and build a 3D digital modeling environment for the construction area; S2. Deploy a sensor network at the construction site to collect real-time wind load data, seismic intensity data, and structural acceleration response data to form a dynamic load data set; S3. Perform unified data cleaning, noise filtering, missing value filling, and time series reconstruction on the dynamic load dataset to generate a preprocessed dynamic load dataset; S4. Utilize the preprocessed dynamic load dataset to construct a gated cyclic unit network prediction model to simulate and predict the dynamic response of the construction structure under earthquake action; S5. Apply the Gray Wolf Optimization Algorithm to perform a global search and optimization of the key hyperparameters of the GRU network prediction model, obtaining an optimized GRU network prediction model. S6. Map the structural response prediction results output by the optimized gated cyclic unit network prediction model into a three-dimensional digital modeling environment to achieve real-time dynamic simulation and three-dimensional visualization of the construction structure under earthquake action. Based on the three-dimensional visualization results, the dynamic response of the construction structure is classified into risk levels and early warning processing is performed to generate construction management decision support information.

2. A three-dimensional visualization construction management method according to claim 1, characterized in that: Said S1 comprises the following steps: S11. Obtaining three-dimensional structural information of the construction project, using a three-dimensional laser scanning device or an unmanned aerial vehicle remote sensing system to collect point cloud data of the three-dimensional structural shape of the construction project to form a point cloud set P; S12. Obtain geographic environmental information of the construction area, including topographic information, geological structure information, and surface cover type, and construct an environmental dataset G consisting of multiple geographic information units; S13. Based on the point cloud set P and the environment dataset G, digital modeling is used to perform multi-source data registration and fusion to construct a 3D digital modeling environment M. d , 3D digital modeling environment M d Represents the mapping relationship M through the function d =f(P,G) to obtain the integrated 3D model of building structure and geographical environment, where function f is the spatial matching and feature reconstruction mapping between point cloud data and geographic information units; S14. The constructed 3D digital modeling environment M d Topological structure organization and mesh division are carried out to form a computable geometric model for response simulation and visual rendering, so that the three-dimensional digital modeling environment has the structural characteristics of node definition, boundary closure and attribute binding.

3. A three-dimensional visualization construction management method according to claim 2, characterized in that: The S2 comprises the following steps: S21. Install wind load monitoring sensors at key locations on the construction structure. These sensors collect real-time data on wind loads generated by wind speed, direction, and impact area. S22. Deploy seismic monitoring sensors in the construction area to collect real-time seismic intensity information, which is represented by horizontal and vertical acceleration components of the ground surface, to form seismic intensity data. S23. Installing a structural response acceleration sensor within the construction structure to monitor the acceleration response of the structure under earthquake and wind loads, thereby generating structural acceleration response data. S24. Synchronize the wind load data, earthquake intensity data, and structural acceleration response data in time and format, and fuse them to generate a complete dynamic load data set D load : Where t represents the unified time index in the dynamic load data set, and v is the alignment result of the time axis of various sensor data. t Indicates the wind speed at time point t, which is used to reflect the wind load intensity, θ t Indicates the wind direction angle at time point t, which is used to determine the direction of wind load. t Indicates the wind action area at time point t, which is used to correspond to the size of the force area of the wind load on the structure. They represent the components of the earthquake acceleration in the X, Y, and Z directions at time point t, is the structural acceleration response data of the pth monitoring point at time t, and T is the unified valid time series set.

4. A three-dimensional visualization construction management method according to claim 3, characterized in that: The S3 includes the following steps: S31. Dynamic load data set D load Perform data integrity checks and remove data records with duplicate timestamps, missing fields, or illegal values to form a preliminary cleansed dynamic load data set; S32. Smooth the high-frequency noise data in the preliminary cleaned dynamic load dataset, set the window size, and perform noise smoothing on any numerical field at each time point t. Interpolate and fill any missing values in the smoothed data at time points t. Reconstruct the dynamic load dataset after missing value filling to a unified time series, remap all fields to a standard time axis, and perform linear resampling on any misaligned data points to form a time series dynamic load dataset. S35. Normalize all fields contained in the time series dynamic load dataset and convert any field into a normalized value to obtain the preprocessed dynamic load dataset D pre .

5. A three-dimensional visualization construction management method according to claim 4, characterized in that: The S4 comprises the following steps: S41. Using the pre-processed dynamic load data set D pre As input data, a gated recurrent unit network prediction model is constructed to predict the dynamic response of construction structures under earthquake action: Among them, h t represents the output hidden state of the gated recurrent unit network prediction model at time point t, which is used to describe the internal response state of the construction structure when subjected to the external earthquake load at time point t. t-1 represents the output hidden state of the gated recurrent unit network prediction model at time point t-1, z t To update the gate control factor, represents the candidate response state at the current time step, Represents the Hadamard element-wise product operator; S42. Using the gated recurrent unit network prediction model, the pre-processed dynamic load data set is trained using a sliding window method. Based on the current moment and historical response, the response state of the structure at multiple future time steps is predicted, and the structural dynamic response prediction sequence H is output. pred ; S43. In the process of constructing the input of the gated recurrent unit network prediction model, considering the difference in the importance of different types of load inputs to the structural response in the construction structure, a multi-dimensional dynamic response attention mechanism is introduced to calculate the input feature x at each time point t. t Each characteristic component Dynamically assign attention weights to construct attention-weighted input features in, Represents the feature component of the input feature in the i-th feature dimension, represents the attention weight of the input feature on the structural response at time point t, d is the total number of dimensions of the input feature, represents the impact strength score of the i-th input feature on the hidden state of the structural response at time point t, represents the impact strength score of the p-th input feature on the hidden state of the structural response at time point t, W e 、b e are the weight matrix and bias vector of relevance attention respectively; S44. Input features after attention weighting Replace the original input feature x t Input to the gated recurrent unit network prediction model, all the updates to the gate control factor z t , candidate response status The calculations are based on Optimize the response capability and prediction sensitivity of the gated cyclic unit network prediction model to sudden seismic motion or severe wind loads.

6. A three-dimensional visualization construction management method according to claim 5, characterized in that: The S5 comprises the following steps: S51. Construct the gray wolf optimized parameter space of the key hyperparameter Θ involved in the gated recurrent unit network prediction model, and map each set of parameter combinations to the position X of the gray wolf population individual. u and input the position into the improved Grey Wolf optimization algorithm driven by three-dimensional structural response; S52. In each iteration, calculate the structural response prediction error corresponding to the u-th gray wolf individual and define the structural response sensitivity factor SRSF u =(s u,1 ,s u,2 ,…,s u,d ): Among them, s u,j represents the structural response sensitivity factor corresponding to the jth key hyperparameter dimension of the uth gray wolf individual, θ u,j represents the value of the u-th gray wolf in the j-th key hyperparameter dimension, f(X u ) is the structural response prediction error function, which is used to characterize the influence of different parameter disturbances on the prediction accuracy; S53. Align the predicted structural response results with the actual structural response values in space, construct a spatiotemporal structural error distribution function, and map the spatiotemporal structural error distribution function back to the 3D digital modeling environment M. d The spatial structure unit in the 3D structure is constructed to construct the spatial mapping factor SMF (x, y, z) to reflect the response prediction error contribution of different regions of the 3D structure and define the response visualization coupling fitness function f'(X u ): Where, λ is the spatial error weighting coefficient; S54. Based on the error decreasing trend in the gray wolf optimization iteration, define the gradient-guided dynamic step size factor γ o , and the structural response sensitivity factor SRSF u , spatial mapping factor SMF and gradient-guided dynamic step factor γ o Integrate it into the gray wolf position update to build a response-driven improved gray wolf update mechanism: X u (d+1)=X u (d)+γ o ·∑ h＝α,β,δ w h ·(X h -X u (d))·SRSF u ; Among them, X u (d+1) is the position vector of the u-th gray wolf individual at the d+1th iteration, representing the updated key hyperparameter combination, X u (d) is the position vector of the u-th gray wolf individual at the d-th iteration, representing the current key hyperparameter combination, X h They represent the positions of the current optimal, suboptimal, and third-optimal individuals, respectively. h = α, β, δ are the three types of leader individuals in the gray wolf algorithm, which are the individual indices of the global optimal α, suboptimal β, and third-optimal δ, respectively. w h is the corresponding guidance weight; S55. Iterate and update S51-S54 until convergence to obtain the optimal set of key hyperparameters Θ opt , apply the optimal set of key hyperparameters to the gated recurrent unit network prediction model, and construct the optimized gated recurrent unit network prediction model GRU opt .

7. A three-dimensional visualization construction management method according to claim 6, characterized in that: The S6 comprises the following steps: S61. The optimized gated recurrent unit network prediction model GRU opt The output structural response prediction sequence H pred Match to 3D digital modeling environment M by time index d The corresponding structural unit in constructs the structural response mapping function R(x, y, z, t), which represents the predicted response intensity at the three-dimensional coordinate point (x, y, z) at time point t; S62. Perform color coding and deformation-driven conversion on the structural response mapping function R(x, y, z, t), convert the response intensity value into dynamic deformation form, vibration frequency visualization, and stress intensity heat map in the three-dimensional structural model, and dynamically render it in conjunction with the time axis to generate a three-dimensional visualization animation of the structural response; S63. Construct a response threshold classification system, set structural response intensity classification intervals, and classify the risk level of the response prediction values of each unit in the three-dimensional structural model; S64. Based on the 3D visualization animation of structural responses and the risk classification results, construction management decision support information is generated, including a report on the overall safety status of the current construction structure, a spatial distribution map of the risk levels of each structural unit, visual early warning prompts and response trend predictions for key nodes, as well as corresponding construction recommendations and scheduling optimization strategies.

8. A three-dimensional visualization construction management method according to claim 7, characterized in that: The risk level classification rules of the response threshold classification system are as follows: Safe area: The value of the structural response mapping function is less than or equal to the risk classification critical value θ1 of the structural response strength, and the unit structural deformation and unit stress are both lower than 40% of the structural bearing standard. The corresponding structural unit is classified as a safe area; Area of concern: The value of the structural response mapping function is greater than θ1 and less than or equal to the risk classification critical value of the structural response intensity θ2, and the unit structural deformation and unit stress are between 40% and 70% of the structural bearing standard. The corresponding structural unit is classified as an area of concern; Warning area: When the value of the structural response mapping function is greater than θ2 and less than or equal to the risk classification critical value θ3 of the structural response strength, and the unit structural deformation and unit stress are between 70% and 90% of the structural bearing standard, the corresponding structural unit is classified as a warning area, and the system automatically prompts to suspend construction and generates response optimization suggestions; Dangerous area: When the value of the structural response mapping function is greater than the risk classification critical value θ3 of the structural response strength, and the unit structural deformation and unit stress exceed 90% of the structural bearing standard, the corresponding structural unit is classified as a dangerous area. The system automatically triggers the early warning mechanism and recommends the implementation of emergency control and construction suspension measures.

9. A three-dimensional visualization construction management system, applied to a three-dimensional visualization construction management method according to any one of claims 1 to 8, characterized in that: include: The data acquisition module is used to obtain construction project information and build a three-dimensional digital modeling environment for the construction area. It also collects wind load data, earthquake intensity data, and structural acceleration response data in real time to form a dynamic load data set. A data preprocessing module is used to preprocess the dynamic load data set to generate a preprocessed dynamic load data set; Structural response prediction module, which is used to construct a gated recurrent unit network prediction model using the preprocessed dynamic load data set; The parameter optimization module is used to apply the Gray Wolf Optimization Algorithm to perform global search and optimization on key hyperparameters in the GRU network prediction model to obtain the optimized GRU network prediction model. A 3D visualization simulation module is used to map the structural response prediction results output by the optimized gated recurrent unit network prediction model into a 3D digital modeling environment to achieve 3D visualization of the structural response; The risk identification and decision support module is used to analyze the structural response mapping function of each structural unit based on the three-dimensional visualization display results, divide the risk level according to the relative relationship between the response intensity and the structural bearing standard, and generate the risk level assessment results of the construction structure.

Images