Intelligent construction method and system based on digital twinning

Abstract

The invention relates to the technical field of computer simulation, in particular to an intelligent construction method and system based on digital twinning. Comprising the steps of obtaining multi-source time sequence sensing data, loading the multi-source time sequence sensing data to a BIM model to generate an initial digital twinborn model, embedding a causal association network of construction risk factors and progress nodes into the initial digital twinborn model, and calculating directed weighted association strength of node variables to target variables; based on the causal association network, utilizing an LSTM algorithm to calculate dynamic influence weights of construction risk factors on progress nodes in real time, and generating a risk progress coupling situation prediction curve; when the prediction curve deviates from a threshold value, optimizing a construction path through process topology reconstruction, and pushing an operation instruction with space-time constraint to a terminal; the problems that in an existing intelligent construction technology, a digital twinborn model is insufficient in correlation description of construction risks and progress, and the accurate prediction capability of the risk progress coupling situation is lacked are solved.

Description

Technical Field

[0001] This invention relates to the field of computer simulation technology, and in particular to a smart construction method and system based on digital twins. Background Technology

[0002] In the field of smart construction, construction risk and schedule management are crucial for ensuring project quality and efficiency. With the digital transformation of the construction industry, traditional management methods are no longer sufficient to meet the increasing complexity and dynamism of the demands. Therefore, how to achieve effective collaborative management of construction risks and schedules has become an important research direction in the field of smart construction technology.

[0003] Currently, the main measures taken to address construction risk and schedule management issues include experience-based risk assessment, static scheduling, and simple information management systems. For example, risk assessment is conducted based on expert experience, schedules are developed, and project management software is used for basic schedule tracking and resource allocation. However, these methods often lack a deep understanding of the dynamic interaction between risk and schedule during construction and the ability to respond in real time.

[0004] While existing technologies provide a basic framework for risk and schedule management to some extent, several problems and shortcomings remain. First, current risk assessment methods rely too heavily on subjective experience and lack data-driven quantitative analysis, resulting in inaccurate risk identification and response. Second, existing schedule management methods are mostly static plans, making it difficult to adapt to dynamic changes during construction, leading to untimely schedule adjustments and impacting project efficiency. Furthermore, existing information management systems often focus only on data recording and display, lacking the ability for in-depth data analysis and intelligent decision support, making it difficult to achieve coordinated optimization of risk and schedule.

[0005] Therefore, developing a smart construction risk and schedule collaborative management method and system based on digital twins has significant practical implications and application value. Summary of the Invention

[0006] To address the aforementioned issues, existing smart construction technologies suffer from insufficient characterization of the relationship between construction risks and schedules using digital twin models. They struggle to dynamically quantify the real-time impact of risk factors on schedule milestones, lack accurate predictive capabilities of the risk-schedule coupling dynamic, and fail to reflect the correlation between risk diffusion and schedule delays. Furthermore, the correction of virtual-to-real deviations relies on preset rules, lacking adaptive strategies based on deviation types and model self-learning iterative mechanisms. This invention provides a smart construction method and system based on digital twins.

[0007] Firstly, the present invention provides a smart construction method based on digital twins, which adopts the following technical solution: A smart construction method based on digital twins includes: S1. Acquire multi-source time-series sensor data, load it into the BIM model to generate an initial digital twin model, embed a causal relationship network between construction risk factors and progress nodes in the initial digital twin model, and calculate the directed weighted correlation strength of node variables with target variables. S2. Based on the causal relationship network, the LSTM algorithm is used to calculate the dynamic impact weight of construction risk factors on schedule nodes in real time, and generate a risk-schedule coupling situation prediction curve. S3. When the predicted curve deviates from the threshold, the construction path is optimized by process topology reconstruction, and an operation command with time and space constraints is pushed to the terminal. S4. Identify the deviation type between the physical building and the twin model through multimodal sensing fusion technology, call the corresponding correction strategy, and use reinforcement learning to update the weights of the causal association network; S5. Extract core evolution data from the construction phase, encapsulate it into a digital twin gene chain along the timeline, and support reverse tracing and forward optimization of operation and maintenance decisions.

[0008] Furthermore, the multi-source time-series sensing data includes structural deformation time-series data collected by a millimeter-wave radar drone, material stress time-series data collected by a distributed fiber optic sensor, personnel operation trajectory time-series data collected by a UWB positioning system, and temperature, humidity, and wind time-series data collected by an environmental monitoring station.

[0009] Furthermore, the method for constructing the causal relationship network is as follows: Define node variable set ,in The nodal variable set represents construction risk factors, including material fatigue, process conflict rate, environmental interference intensity, and frequency of personnel errors. Define the target variable set ,in The target variable set, which represents the progress node, includes the completion time of each sub-project, resource turnover cycle, and acceptance pass rate. The strength of the causal relationship between nodal variables and the target variable is calculated by Granger causality test, and a directed weighted network is constructed. Among them, the nodal variables are calculated using the Granger causality test. right The strength of causal association is expressed by the formula: in The regression coefficients are lagged by k. For the maximum lag order, with Construct a directed weighted network for edge weights , For edge set, For a set of nodes.

[0010] Furthermore, the LSTM algorithm is an improved LSTM algorithm, which includes: A risk diffusion coefficient moment is introduced into the LSTM hidden layer to dynamically correct the weights of risk factors on schedule nodes at different construction stages. The formula for calculating the hidden state is as follows: ,in, Let be the hidden layer state at time t. This is the output gate at time t. Let t represent the cell state at time t. The weight matrix of the input data, This is the Sigmoid activation function.

[0011] Furthermore, the optimization of the construction path through process topology reconstruction includes: Break down the process into sub-task sets ,in Construct a directed acyclic graph for the r-th specific subtask in the set. , A graph describing the dependencies between subtasks. This is a set of dependencies between subtasks; Simulate the resource conflict probability of parallel subtasks Delete those that meet the requirements The edge, The preset threshold; Calculate the critical path length of the remaining paths ,choose The corresponding path is to optimize the topology. The shortest critical path length among all feasible paths.

[0012] Furthermore, the step of pushing operation instructions with spatiotemporal constraints to the terminal includes: Based on the three-dimensional spatial coordinates of absolute positioning of the digital twin model, the start and end times of the operation are deduced from the progress node time, and the safety threshold of operation accuracy is determined based on the material performance parameters. Three-dimensional spatial coordinates ,in, These are the vertex coordinates of the corresponding component in the digital twin model. The coordinates of the entity location collected by the sensor; Time window satisfy ,in, This is the start time of the operation. This is the end time of the operation. For progress milestone times. This is the buffer coefficient. This refers to the critical path length. Operational accuracy threshold ,in, For the material's yield strength, To the maximum allowable deformation, For the strength constraint safety factor, This is the safety factor for deformation constraints.

[0013] Furthermore, the invocation of the corresponding correction strategy includes: Two-dimensional contour feature point set extracted by binocular vision laser point cloud 3D model point set Registration is performed, expressed by the formula: ,in, Let R be the geometric parameter deviation between the entity and the model, R be the rotation matrix, and t be the translation vector. For the 3D model point set acquired by laser point cloud, This is a set of two-dimensional contour feature points extracted by binocular vision. If geometric parameters deviate Then, the point cloud registration algorithm is invoked to adjust the coordinates of the virtual component. The threshold for determining geometric deviation; If physical parameters deviate Then update the material mechanics parameter matrix. , The threshold for determining physical deviation; If timing deviation This increases the weight of the corresponding risk factor in the causal relationship network. ,in, For learning rate, This is the threshold for determining timing deviation.

[0014] Furthermore, updating the weights of the causal association network using reinforcement learning includes: Using the deviation correction effect as the reward function signal, the weight coefficients of directed edges in the network are dynamically adjusted. Expressed as a formula: ,in, This is the reward value for the deviation correction effect. This refers to the residual deviation after correction. The weight update formula is: ,in, For the updated weights, As the current weight, For learning rate, The gradient represents the weights and is used to adjust the direction.

[0015] Furthermore, the digital twin gene chain encapsulated along the timeline includes, Material property degradation curve ,in, For initial performance, The attenuation coefficient is... For time; Deviation sequence between actual and planned time for key processes: ,in, The deviation between the actual time and the planned time for the k-th process is given. The actual time spent on the k-th process. The planned time for the k-th process, This represents the total number of processes.

[0016] Furthermore, the support for operation and maintenance decisions includes: During the operation and maintenance phase, the digital twin gene chain is imported into the operation and maintenance digital twin model as the initial parameter for building condition assessment. Based on the gene chain analysis of the factors affecting the current state of the building during the construction process, combined with real-time operation and maintenance monitoring data, the structural aging trend and equipment maintenance cycle are predicted. Based on the prediction results, an operation and maintenance optimization plan is generated, enabling reverse tracing and forward optimization of operation and maintenance decisions based on construction phase data.

[0017] The reverse tracing includes, based on the location of structural damage detected during the operation and maintenance period, linking the material performance decay curve and process deviation sequence of the corresponding area during the construction stage through gene chain correlation. The positive optimization includes correcting the equipment maintenance cycle model parameters based on the process deviation sequence in the gene chain to generate a preventive maintenance plan.

[0018] Secondly, a smart construction system based on digital twins includes: The data acquisition module is configured to acquire multi-source time-series sensor data, load it into the BIM model to generate an initial digital twin model, embed a causal relationship network between construction risk factors and progress nodes in the initial digital twin model, and calculate the directed weighted correlation strength of node variables with target variables. The situation prediction module is configured to, based on the causal relationship network, use the LSTM algorithm to calculate the dynamic impact weight of construction risk factors on schedule nodes in real time, and generate a risk-schedule coupled situation prediction curve. The process optimization instruction module is configured to optimize the construction path through process topology reconstruction and push operation instructions with spatiotemporal constraints to the terminal when the predicted curve deviates from the threshold. The deviation identification and correction module is configured to identify the deviation type between the physical building and the twin model through multimodal sensing fusion technology, call the corresponding correction strategy, and update the weights of the causal association network using reinforcement learning. The gene chain operation and maintenance module is configured to extract core evolution data during the construction phase, encapsulate it into a digital twin gene chain along a timeline, and support reverse tracing and forward optimization of operation and maintenance decisions.

[0019] Thirdly, the present invention provides a computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned smart construction method based on digital twins.

[0020] Fourthly, the present invention provides a terminal device, including a processor and a computer-readable storage medium, wherein the processor is used to implement various instructions; the computer-readable storage medium is used to store multiple instructions, the instructions being adapted to be loaded and executed by the processor to provide the aforementioned smart construction method based on digital twins.

[0021] In summary, the present invention has the following beneficial technical effects: 1. The present invention proposes a smart construction method and system based on digital twins, which improves the synergy between risk control and schedule scheduling. By constructing a dynamic twin model containing causal relationships, this technical solution can dynamically quantify the real-time impact mechanism of risk factors on schedule nodes, effectively improving the synergy between risk control and schedule scheduling.

[0022] 2. This invention enhances the accurate prediction capability of risk-schedule coupling. By utilizing a time-series deep learning algorithm, this technical solution can calculate the dynamic impact weight of construction risk factors on schedule nodes in real time, generate a risk-schedule coupling prediction curve, thereby enhancing the accurate prediction capability of risk-schedule coupling and improving the accuracy of decision-making.

[0023] 3. This invention achieves adaptive correction of virtual-real discrepancies. This technical solution adopts AI visual recognition and laser point cloud fusion technology to compare the geometric and physical parameters of the physical building and the dynamic twin model in real time, and automatically calls the corresponding correction strategy based on the type of discrepancy. By updating the association weights of the causal association network through reinforcement learning, the model can achieve self-iteration and ensure the long-term consistency between the model and the physical building.

[0024] 4. This invention establishes a full-cycle data inheritance mechanism. By constructing a full-cycle digital twin gene chain, this technical solution can extract the core evolution data of the dynamic twin model during the construction phase, and analyze the influencing factors of the construction process on the current state of the building during the operation and maintenance phase through the gene chain. This supports the reverse tracing and forward optimization of operation and maintenance decisions, thereby solving the problem of data separation between the construction and operation and maintenance phases and providing in-depth support for operation and maintenance decisions based on construction process data. Attached Figure Description

[0025] Figure 1 This is the method flow of Embodiment 1 of the present invention; Figure 2 This is a block diagram of Embodiment 2 of the present invention; Detailed Implementation The present invention will be further described in detail below with reference to the accompanying drawings.

[0026] Example 1 Reference Figure 1 This embodiment of a smart construction method based on digital twins includes: S1. Acquire multi-source time-series sensor data, load it into the BIM model to generate an initial digital twin model, embed a causal relationship network between construction risk factors and progress nodes in the initial digital twin model, and calculate the directed weighted correlation strength of node variables with target variables. The multi-source time-series sensing data includes structural deformation time-series data collected by millimeter-wave radar drones, material stress time-series data collected by distributed fiber optic sensors, personnel operation trajectory time-series data collected by UWB positioning systems, and temperature, humidity, and wind time-series data collected by environmental monitoring stations.

[0027] Millimeter-wave radar drone: A quadcopter drone equipped with a 24GHz millimeter-wave radar is used to inspect the main structure of the building every 2 hours, collect time-series data on structural deformation, with a sampling frequency of 10Hz and a data format of three-dimensional coordinate point cloud sequence with an accuracy of ±2mm. The focus is on monitoring the displacement changes of shear walls, beams and columns.

[0028] Distributed fiber optic sensor: Distributed fiber optic cables are laid along the steel reinforcement of the floor slab, with one monitoring point set every 10m. The material stress time series data is collected in real time by an optical time domain reflectometer with a sampling frequency of 1Hz. The data format is a stress value sequence with a range of 0-500MPa, recording the stress changes of concrete and steel reinforcement.

[0029] UWB positioning system: Eight UWB positioning base stations are deployed at the construction site, covering a radius of 50m. Construction workers wear positioning tags to collect time-series data of personnel operation trajectories. The sampling frequency is 5Hz, the data format is a three-dimensional coordinate sequence, the positioning accuracy is ±30cm, and the movement path and dwell time of personnel on each work surface are recorded.

[0030] Environmental monitoring stations: One environmental monitoring station is set up at each of the four corners of the construction site (east, south, west, and north) to collect temperature and humidity data at a sampling frequency of 0.5Hz and an accuracy of ±0.5℃ and ±2%RH, as well as wind data at a sampling frequency of 1Hz and a range of 0-10. The data is transmitted wirelessly to the edge computing node via LoRa.

[0031] The collected multi-source data were spatiotemporally aligned. The UAV inspection time was used as the reference time axis, and the fiber stress data, UWB trajectory data and environmental data were matched by timestamps. Kalman filtering was used to remove noise points in the UAV point cloud data, and wavelet transform was used to denoise the fiber stress data to ensure that the data signal-to-noise ratio was ≥30dB.

[0032] Based on the preprocessed multi-source data, the process of loading the data into the BIM model to generate an initial digital twin model is performed. The initial BIM model is a three-dimensional structural model in Revit format, which includes the geometric parameters (dimensions, coordinates) of building components, material properties (concrete strength grade C30, steel reinforcement type HRB400) and construction sequence information. The model accuracy reaches LOD400.

[0033] Data fusion and model building: Geometric registration of UAV point cloud data with BIM model is performed. Using ICP algorithm, with the building foundation axis as the reference, the three-dimensional coordinate system of point cloud data and BIM model is unified with an error of ≤5mm, generating a real-scene three-dimensional model containing real-time structural morphology.

[0034] After receiving the pre-processed time-series data, the fiber optic stress data, UWB trajectory data, and environmental data are associated with the corresponding locations in the BIM model by component ID through edge computing nodes. For example, the stress data of the 3rd floor slab is bound to the "3rd floor slab" component in the model to build a dynamically updated initial digital twin model. The model update frequency is synchronized with the data acquisition frequency, up to 1Hz.

[0035] The method for constructing the causal relationship network is as follows: S11. Define the set of node variables ,in The nodal variable set represents construction risk factors, including material fatigue, process conflict rate, environmental interference intensity, and frequency of personnel errors. Define node variable set The quantification methods for each variable are as follows: Material fatigue, based on stress data collected by fiber optic sensors, is calculated using the formula... calculate, For real-time stress, The initial stress is designed with a value range of [0,1].

[0036] Process conflict rate, calculated based on the duration of overlapping operations using UWB trajectory data, is determined by... calculate, The value ranges from [0,1] to the time of overlapping work by personnel in different processes.

[0037] Environmental disturbance intensity, integrating temperature, humidity and wind data, according to The value range is [0,1].

[0038] The frequency of human error is counted by combining AI video recognition with UWB positioning to determine the number of times operations were not performed according to specifications. Calculate, with a value range of [0,1].

[0039] S12. Define the target variable set ,in The target variable set, which represents the progress node, includes the completion time of each sub-project, resource turnover cycle, and acceptance pass rate. Define the target variable set The quantification methods for each variable are as follows: The completion time of each sub-project is recorded in days, showing the difference between the actual completion time and the planned time for sub-projects such as "5th floor of main structure" and "3rd floor of masonry works". Unit: days.

[0040] Resource turnover cycle, statistical analysis of the idle time of reusable materials such as templates and scaffolding, according to Calculate, with a value range of [0,1].

[0041] Acceptance pass rate, according to Calculate, with a value range of [0,1].

[0042] S13. Calculate the causal relationship strength between nodal variables and target variables using Granger causality tests, and construct a directed weighted network; Among them, the nodal variables are calculated using the Granger causality test. right The strength of causal association is expressed by the formula: in The regression coefficients are lagged by k. For the maximum lag order, with Construct a directed weighted network for edge weights , For edge set, For a set of nodes.

[0043] Data preprocessing, for the set of node variables and target variable set The time series data were tested for stationarity using the ADF test. Non-stationary data were processed by first-order differencing until stationarity was achieved, with a p-value < 0.05.

[0044] Lag order determination: The maximum lag order p=3 is determined using the AIC information criterion, which means analyzing the impact of risk factors in the first 3 time steps on the current schedule node.

[0045] Causal strength calculation: The causal association strength between nodal variables and target variables is calculated using the Granger causality test formula. For each target variable... Constructing regression equations , ( (For the error term), the lag is solved using the least squares method. regression coefficients of order And calculate the goodness of fit. ,by As a node variable For target variable The strength of the causal relationship, i.e., the weight value.

[0046] Directed weighted network construction: The node set consists of 7 nodes, calculated as follows. Construct a directed weighted network with edge weights. For example, the calculation yields... (Environmental disturbance intensity) on Weight of (Completion Time of Sub-projects) Then add a route from the network. point to The directed edges with a weight of 0.62 visually reflect the significant impact of environmental disturbances on the construction progress.

[0047] By refining the equipment deployment and preprocessing procedures for multi-source data acquisition, the real-time performance and accuracy of the initial digital twin model are ensured. By quantifying node variables and target variables and standardizing the implementation steps of Granger causality tests, the causal relationship network can accurately depict the dynamic causal relationship between construction risks and progress, providing a reliable correlation basis for subsequent risk-progress coupling situation prediction, and realizing the full-process technology implementation from "data acquisition" to "correlation modeling".

[0048] S2. Based on the causal relationship network, the LSTM algorithm is used to calculate the dynamic impact weight of construction risk factors on schedule nodes in real time, and generate a risk-schedule coupling situation prediction curve. The LSTM algorithm is an improved LSTM algorithm, which includes: A risk diffusion coefficient moment is introduced into the LSTM hidden layer to dynamically correct the weights of risk factors on schedule nodes at different construction stages. The formula for calculating the hidden state is as follows: ,in, Let be the hidden layer state at time t. This is the output gate at time t. Let t represent the cell state at time t. The weight matrix of the input data, This is the Sigmoid activation function.

[0049] The construction and input data processing of the improved LSTM algorithm were carried out during the main structure construction phase of this high-rise residential project, specifically for floors 3-10. Based on the causal relationship network constructed in step S1, which includes 4 risk factor nodes and 3 progress nodes, an improved LSTM algorithm model was built. The specific implementation is as follows: The model structure design of the improved LSTM algorithm adopts a three-layer architecture of "input layer - hidden layer - output layer". The input layer receives a temporally sequenced set of node variables. The data has dimensions [T, 4], where T is the time step, taking the most recent 7 days of time-series data, with 12 sampling points per day (T=84), and 4 represents the number of risk factors: material fatigue, process conflict rate, environmental interference intensity, and frequency of human error. Two hidden layers are used, each containing 64 neurons. The core improvement is embedding a risk diffusion coefficient matrix in the first hidden layer. The dimension is [4,3]. The rows correspond to risk factors, and the columns correspond to construction stages: foundation stage, main structure stage, and decoration stage. It is used to dynamically adjust the impact weight of risk factors on schedule nodes at different stages. Output layer: Outputs risk-schedule coupling status parameters, including risk diffusion speed, in units of risk index / hour and schedule lag probability, with a dimension of [T,2].

[0050] Input data preprocessing, specifically the input set of node variables. Time series data are standardized using the min-max standardization formula. The data of each risk factor are mapped to the [0,1] interval. For example, the original value of material fatigue is 0-1 and remains unchanged after standardization, and the environmental disturbance intensity is 0-10 and standardized to 0-1.

[0051] The causal relationship network node set combined with step S1 Historical construction data from the past three months was extracted, containing 2,000 complete time-series samples. These samples were divided into a training set of 1,600 sets and a validation set of 400 sets at an 8:2 ratio for model training and accuracy verification.

[0052] Risk diffusion coefficient matrix The construction of the project is determined based on the characteristics of each construction phase, using the following methods. element value , indicating the first The influence weight of risk factors in the corresponding construction stage, main structure stage (current stage): material fatigue. Process conflict rate The impact is significant, therefore , ; Environmental disturbance intensity and frequency of personnel errors The impact is secondary, therefore , Matrix dynamic update mechanism: When the construction phase changes, such as entering the decoration phase, the matrix is ​​updated in real time through the causal relationship network of step S1. Element values, such as the renovation stage The weighting was increased to 0.30 to ensure that the risk factor weights were adapted to the characteristics of the stage.

[0053] The specific implementation of hidden state computation, the hidden state computation formula of the improved LSTM is as follows: The actual values ​​and calculation logic of each parameter are as follows: The output gate at time t is calculated using the Sigmoid function. The weight matrix The dimensions are [64, 128], with an offset. It is a 64-dimensional vector with an initial value of 0.1.

[0054] Let be the cell state at time t, updated through the forget gate and the input gate, as shown in the formula: Forgotten Gate and input gate All calculations are performed using the Sigmoid function to ensure the effective transmission of long-term dependency information.

[0055] The weight matrix for the input data has dimensions [4, 64] and is optimized iteratively using the training set data. The initial values ​​follow a normal distribution N(0, 0.01).

[0056] The Sigmoid activation function is used to map input data to the [0,1] interval, thereby enhancing the non-linear characteristics of the influence of risk factors.

[0057] Dynamic impact weight calculation and risk-schedule coupling situation prediction curve generation Model training and optimization use the mean absolute error (MAE) to evaluate prediction accuracy, as shown in the formula: ,in, This represents the actual progress deviation. These are predicted values.

[0058] During training, the Adam optimizer was used with a learning rate of 0.001 and a decay rate of 0.9. The training was iterated for 500 rounds, and the results were validated every 10 rounds. Training was stopped when the MAE of the validation set decreased to less than 0.01 for 5 consecutive rounds. The final training set MAE was 0.03 and the validation set MAE was 0.05.

[0059] The dynamic influence weights are calculated in real time. After the model training is completed, the multi-source time-series data from step S1 is received in real time, and the input is updated every 30 minutes. The dynamic impact weights of construction risk factors on schedule milestones are calculated through the following steps: For input data Standardize the data to a mean of 0 and a standard deviation of 1 to ensure consistency with the distribution of the training data.

[0060] The improved LSTM hidden layer is used to calculate and output each risk factor. For progress nodes ( The real-time weight value of ).

[0061] 3. Generation of Coupled Situation Prediction Curve: Based on dynamic influence weights, a risk-progress coupled situation prediction curve is generated. The specific implementation is as follows: Curve dimension definition: The horizontal axis represents time (next 7 days), the left side of the vertical axis represents the risk diffusion speed (unit: risk index / day), and the right side of the vertical axis represents the probability of progress lag (%), which are represented by different colored curves.

[0062] Curve generation logic: When the weight of v3 (environmental disturbance intensity) increases (such as when continuous rainfall is detected), the model predicts that the risk spread rate increases from 0.5 exponential / day to 1.2 exponential / day, and the synchronous output progress lag probability increases from 15% to 45%, forming a prediction curve that links "risk-progress".

[0063] Visual output: The prediction curve is displayed in real time through the digital twin platform, and the risk threshold line is marked. The risk diffusion speed is greater than 1.0 index / day and the progress warning line is marked. The lag probability is greater than 30%, which provides an intuitive decision-making basis for the process optimization of step S3.

[0064] Through dynamic optimization of the improved LSTM algorithm, the dynamic correlation weight calculation between construction risk factors and schedule nodes was realized. Compared with the traditional LSTM algorithm, the generated coupling situation prediction curve can accurately depict the linkage relationship of "risk diffusion-schedule lag", providing a reliable quantitative basis for subsequent process topology reconstruction and resource scheduling.

[0065] S3. When the predicted curve deviates from the threshold, the construction path is optimized by process topology reconstruction, and an operation command with time and space constraints is pushed to the terminal. The optimization of construction paths through process topology reconstruction includes: S31. Break down the process into sub-task sets. ,in Construct a directed acyclic graph for the r-th specific subtask in the set. , A graph describing the dependencies between subtasks. This is a set of dependencies between subtasks; Based on the construction process requirements, the main process of "5-6 floor slab construction" is broken down into 6 independently executable sub-tasks, forming a sub-task set. , : Installation of formwork for 5-story floor slabs requires a carpentry team and formwork materials; : Reinforcing steel tying for 5-story floor slabs requires a team of steelworkers and steel reinforcement materials; : 5 layers of pipeline pre-embedding, requires an electrician team + pipeline materials; : Installation of formwork for 6th floor slabs, same as Resource requirements; : Reinforcement binding of the 6th floor slab, same Resource requirements; Five layers of concrete pouring are required, necessitating a concrete pump truck and vibration equipment. Based on construction process constraints, a directed acyclic graph is constructed. edge set Indicates subtask dependencies: → Reinforcing bars can only be tied after the formwork is installed; →q3, Pipelines can only be embedded after the steel reinforcement is tied; → Concrete can only be poured after the pipeline is pre-buried; → The installation of the 6th layer template can begin when the 5th layer template is 50% complete.

[0066] S32. Simulate the resource conflict probability of parallel subtasks Delete those that meet the requirements The edge, The preset threshold; The construction site is limited to one tower crane for material hoisting, one carpentry team (8 people), and one steelwork team (10 people). Resource conflicts mainly manifest as conflicts in the use of the tower crane and conflicts in the manpower of the work teams.

[0067] Conflict probability calculation logic: The resource conflict probability of parallel subtasks is calculated using the resource load rate method. :like and Shared tower crane resources, probability of conflict (Load rate = Subtask occupancy time / Resource availability time); for example (6-layer template installation) and (Reinforcing steel binding on 5 floors) requires material hoisting by tower crane. Calculations show that when both are carried out concurrently, the tower crane load rate reaches 0.85. .

[0068] Conflicting edge deletion rules, with a preset conflict probability threshold. ,when Delete the corresponding edge. For example, as mentioned above. and The conflict probability is 0.85 > 0.7, so the edge should be deleted. → Adjusted to Start after completion To avoid conflicts over tower crane resources.

[0069] S33. Calculate the critical path length of the remaining paths. ,choose The corresponding path is to optimize the topology. The shortest critical path length among all feasible paths.

[0070] After removing the conflicting edges, the remaining feasible paths are: Path 1: → → → Path 2: → → → The critical path is defined as the longest-running path among all paths.

[0071] The parallel relationships between subtasks are adjusted by recalculating using a process topology reconstruction algorithm, such as... and Parallel execution, utilizing resources from different work groups, the new path is → →( ∥ → → The critical path length is shortened, and selection is made. The corresponding approach is to optimize the topology, thereby shortening the construction period.

[0072] The process of pushing operation instructions with spatiotemporal constraints to the terminal includes: S34. Based on the three-dimensional spatial coordinates of absolute positioning of the digital twin model, the start and end times of the operation are deduced from the progress node time, and the safety threshold of operation accuracy is determined based on the material performance parameters; Three-dimensional spatial coordinates ,in, These are the vertex coordinates of the corresponding component in the digital twin model. The coordinates of the entity location collected by the sensor; S35. Time Window satisfy ,in, This is the start time of the operation. This is the end time of the operation. For progress milestone times. This is the buffer coefficient. This refers to the critical path length. S36. Operational accuracy threshold ,in, For the material's yield strength, To the maximum allowable deformation, For the strength constraint safety factor, This is the safety factor for deformation constraints.

[0073] By reconstructing the process topology, resource conflicts such as tower cranes and manpower are effectively identified and eliminated, and the critical path construction period is shortened. By accurately calculating three-dimensional coordinates, time windows and accuracy thresholds, and strictly matching the accuracy thresholds with material performance and specification requirements, a precise execution basis is provided for subsequent construction, realizing closed-loop management of "risk warning - path optimization - instruction implementation", which significantly improves construction efficiency and safety.

[0074] S4. Identify the deviation type between the physical building and the twin model through multimodal sensing fusion technology, call the corresponding correction strategy, and use reinforcement learning to update the weights of the causal association network; The invocation of the corresponding correction strategy includes: S41. By extracting a set of two-dimensional contour feature points using binocular vision laser point cloud 3D model point set Registration is performed, expressed by the formula: ,in, Let R be the geometric parameter deviation between the entity and the model, R be the rotation matrix, and t be the translation vector. For the 3D model point set acquired by laser point cloud, This is a set of two-dimensional contour feature points extracted by binocular vision. S42. If the geometric parameters deviate Then, the point cloud registration algorithm is invoked to adjust the coordinates of the virtual component. The threshold for determining geometric deviation; S43. If physical parameters deviate Then update the material mechanics parameter matrix. , The threshold for determining physical deviation; S44. If timing deviation This increases the weight of the corresponding risk factor in the causal relationship network. ,in, For learning rate, This is the threshold for determining timing deviation.

[0075] The step of updating the weights of the causal association network using reinforcement learning includes: S45. Using the deviation correction effect as the reward function signal, dynamically adjust the weight coefficients of directed edges in the network. Expressed as a formula: ,in, This is the reward value for the deviation correction effect. This refers to the residual deviation after correction. S46. The weight update formula is: ,in, For the updated weights, As the current weight, For learning rate, The gradient represents the weights and is used to adjust the direction.

[0076] By combining binocular vision with laser point cloud fusion technology, the adaptability and long-term stability of digital twin models are effectively improved, providing reliable support for precise control of the construction process.

[0077] S5. Extract core evolution data from the construction phase, encapsulate it into a digital twin gene chain along the timeline, and support reverse tracing and forward optimization of operation and maintenance decisions.

[0078] The time-axis-encapsulated digital twin gene chain includes, S51. Material property degradation curve ,in, For initial performance, The attenuation coefficient is... For time; For the concrete and steel reinforcement materials of a 5-story floor slab, a material performance degradation curve was constructed based on sensor data and laboratory test results during the construction phase. Parameters determined, initial performance The initial compressive strength of the concrete was determined to be A=30MPa through 28-day test blocks; the initial yield strength of the reinforcing steel was A=335MPa. The attenuation coefficient λ was obtained based on stress attenuation data collected by distributed fiber optic sensors during the construction phase, with tests conducted quarterly. The concrete yield strength was determined by least squares fitting. steel bars Steel bars have better corrosion resistance than concrete.

[0079] Curve generation, based on the time axis t, for example, generating a degradation curve for 0-50 years, such as the 5-year performance curve of concrete. The error between the actual measured value (27.3 MPa) and the actual measured value is less than 1%, ensuring the accuracy of the curve.

[0080] S52. Deviation sequence between actual and planned time for key processes: ,in, The deviation between the actual time and the planned time for the k-th process is given. The actual time spent on the k-th process. The planned time for the k-th process, This represents the total number of processes.

[0081] For key construction processes such as formwork installation, rebar tying, and concrete pouring, record the deviation sequence between actual and planned time consumption. , Six key processes that significantly impact structural safety were selected, forming a deviation sequence of N=6, specifically defined as follows: Installation of 5-layer templates, planned for 8 hours, actually took 10 hours. ; 5 layers of rebar tying, planned for 12 hours, actually took 16 hours. ; Five layers of concrete were poured, planned for 6 hours, actually took 5 hours. The deviation values ​​for the remaining processes are as follows: , , .

[0082] The deviation sequence is stored on blockchain nodes according to the construction timeline and is associated with the material performance data and personnel operation records of the corresponding process to form a complete digital twin gene chain.

[0083] The support for operation and maintenance decisions includes: S53. During the operation and maintenance phase, the digital twin gene chain is imported into the operation and maintenance digital twin model as the initial parameter for building status assessment; For model initialization and operation phases, the digital twin model is built based on the Revit Operation Edition BIM model. The digital twin gene chain encapsulated during the construction phase, including material performance degradation curves and process deviation sequences, is imported into the model database via API as initial parameters for condition assessment. For example, the concrete degradation curves of a 5-story slab... The "5-story floor slab" component attribute is bound to the model, replacing the empirical parameters of the traditional operation and maintenance model. Data association is established, linking gene chain data with sensors deployed during the operation and maintenance phase, such as floor slab strain gauges and elevator vibration sensors, to create a real-time monitoring data association index, ensuring the spatiotemporal consistency between historical construction data and current operation and maintenance data.

[0084] S54. Based on the gene chain analysis of the factors affecting the current state of the building during the construction process, and combined with real-time operation and maintenance monitoring data, predict the structural aging trend and equipment maintenance cycle; Influencing factors were identified, and real-time monitoring data during the operation and maintenance phase was used. A maximum crack width of 0.2mm was detected in the 5th floor slab, and the elevator failure rate was 15% higher than the design value. Pearson correlation analysis was conducted to determine the correlation between the genetic chain data and the current state: floor slab cracks and... The correlation between excessive rebar tying time and initial concrete strength A=30MPa reached 0.82, indicating they were identified as major influencing factors; elevator abnormal noise and The elevator shaft construction exceeded the time limit, with a correlation of 0.76, which is related to the cumulative verticality deviation of the shaft during the construction phase.

[0085] Aging trend prediction based on material performance degradation curves Based on operation and maintenance monitoring data, the grey prediction model GM(1,1) is used to predict that the concrete strength of the floor slab will decrease to 24MPa and the crack width may increase to 0.3mm in the next 3 years, providing a quantitative basis for the formulation of operation and maintenance plans.

[0086] S55. Generate an operation and maintenance optimization plan based on the prediction results, and realize the reverse tracing and forward optimization of operation and maintenance decisions based on the data of the construction phase.

[0087] S56. The reverse tracing includes, based on the location of structural damage detected during the operation and maintenance period, linking the material performance decay curve and process deviation sequence of the corresponding area during the construction stage through gene chain correlation; S57. The forward optimization includes correcting the equipment maintenance cycle model parameters based on the process deviation sequence in the gene chain to generate a preventive maintenance plan.

[0088] By using digital twin gene chains to achieve full-cycle inheritance of construction and operation and maintenance data, the prediction of material performance degradation curves is improved, and reverse tracing can accurately locate the root cause of structural damage during the construction stage, reducing operation and maintenance troubleshooting time. Forward optimization makes the equipment maintenance cycle adapt to the characteristics of construction deviations, thereby reducing the failure rate.

[0089] Example 2 This embodiment provides a smart construction system based on digital twins, including: The data acquisition module is configured to acquire multi-source time-series sensor data, load it into the BIM model to generate an initial digital twin model, embed a causal relationship network between construction risk factors and progress nodes in the initial digital twin model, and calculate the directed weighted correlation strength of node variables with target variables. The situation prediction module is configured to, based on the causal relationship network, use the LSTM algorithm to calculate the dynamic impact weight of construction risk factors on schedule nodes in real time, and generate a risk-schedule coupled situation prediction curve. The process optimization instruction module is configured to optimize the construction path through process topology reconstruction and push operation instructions with spatiotemporal constraints to the terminal when the predicted curve deviates from the threshold. The deviation identification and correction module is configured to identify the deviation type between the physical building and the twin model through multimodal sensing fusion technology, call the corresponding correction strategy, and update the weights of the causal association network using reinforcement learning. The gene chain operation and maintenance module is configured to extract core evolution data during the construction phase, encapsulate it into a digital twin gene chain along a timeline, and support reverse tracing and forward optimization of operation and maintenance decisions.

[0090] A computer-readable storage medium storing a plurality of instructions adapted for loading and execution by a processor of a terminal device of the aforementioned digital twin-based smart construction method.

[0091] A terminal device includes a processor and a computer-readable storage medium, the processor being used to implement various instructions; the computer-readable storage medium being used to store multiple instructions adapted to be loaded and executed by the processor as described in the digital twin-based smart construction method.

[0092] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A smart construction method based on digital twinning, characterized in that, The method comprises the following steps: acquiring multi-source time-series sensing data, loading the data into a BIM model to generate an initial digital twin model, embedding a causal correlation network of construction risk factors and progress nodes in the initial digital twin model, and calculating the directional and weighted correlation strength of node variables to target variables; based on the causal correlation network, using an LSTM algorithm to dynamically calculate the influence weight of construction risk factors on progress nodes in real time, and generating a risk progress coupling situation prediction curve; when the prediction curve deviates from a threshold value, optimizing the construction path by process topology reconstruction, and pushing operation instructions with space-time constraints to a terminal; identifying the deviation type of the entity building and the twin model through multi-modal sensing fusion technology, calling corresponding correction strategies, and updating the weight of the causal correlation network using reinforcement learning; extracting core evolution data at the construction stage, encapsulating the data into a digital twin gene chain according to a time axis, and supporting reverse tracing and forward optimization of operation and maintenance decisions. 2.The digital-twin-based smart construction method of claim 1, wherein, The method for constructing the causal correlation network comprises the following steps: Definition of node variable set wherein represents a construction risk factor, and the node variable set includes material fatigue degree, process conflict rate, environmental disturbance intensity, and personnel error frequency; Define target variable set wherein representing progress nodes, the target variable set includes subproject completion time, resource turnover period, and acceptance pass rate; calculating the causal correlation strength of node variables to target variables through Granger causality test, and constructing a directional and weighted network. 3.The digital-twin-based smart construction method of claim 1, wherein, The LSTM algorithm is an improved LSTM algorithm, which comprises the following steps: The risk diffusion coefficient matrix is introduced into the LSTM hidden layer to dynamically correct the influence weight of risk factors on the progress node in different construction stages. The calculation formula of the hidden state is: wherein, is the hidden layer state at time t, is the output gate at time t, is the cell state at time t, is the weight matrix of the input data, is the Sigmoid activation function. 4.The digital-twin-based smart construction method of claim 1, wherein, the process topology reconstruction for optimizing the construction path comprises the following steps: Decompose the process into a set of subtasks wherein constructing a directed acyclic graph for the rth specific subtask in the set , is a graph describing the subtask dependencies, is a set of dependencies between subtasks; Simulating resource conflict probability of parallel subtasks , deleting edges satisfying , is a preset threshold value; Computing a critical path length of a remaining path , selecting corresponding paths are optimized topologies, is the shortest critical path length among all feasible paths. 5.The digital-twin-based smart construction method of claim 1, wherein, the pushing of operation instructions with space-time constraints to the terminal comprises the following steps: based on the three-dimensional space coordinates of the digital twin model absolute positioning, the operation start time and end time obtained by time reverse tracing of the progress nodes, and the operation precision safety threshold determined based on the material performance parameters; Three-dimensional spatial coordinates wherein, is a vertex coordinate of a corresponding component in the digital twin model, is an entity positioning coordinate collected by the sensor; Time window Satisfies wherein, is an operation start time, is an operation end time, is a progress node time, is a buffer coefficient, is a critical path length; operational precision threshold wherein, is the material yield strength, is the maximum allowable deformation, is the strength constraint safety factor, is the deformation constraint safety factor. 6.The digital-twin-based smart construction method of claim 1, wherein, the calling of corresponding correction strategies comprises the following steps: By binocular vision extraction of two-dimensional contour feature point set With laser point cloud three-dimensional model point set Registration, expressed as a formula: , wherein is the geometric parameter deviation value of the entity and the model, R is the rotation matrix, t is the translation vector, is the three-dimensional model point set collected by the laser point cloud, is the two-dimensional contour feature point set extracted by binocular vision; if the geometric parameter deviation then call the point cloud registration algorithm to adjust the virtual component coordinates, is the determination threshold of the geometric deviation; If the physical parameter deviation , then update the material mechanics parameter matrix , is the determination threshold of the physical deviation; if the timing deviation then the weight of the corresponding risk factor in the causal correlation network is increased wherein, is a learning rate, is a determination threshold of the timing deviation. 7.The digital-twin-based smart construction method of claim 1, wherein, the updating of the weight of the causal correlation network using reinforcement learning comprises the following steps: taking the deviation correction effect as a reward function signal, and dynamically adjusting the weight coefficient of the directional edges in the network, is expressed by the formula: wherein, is a reward value of the bias correction effect, is the residual bias after correction; The weight update formula is: wherein, is the updated weight, is the current weight, is the learning rate, denotes the gradient of the weight, which is used to adjust the direction. 8.The digital-twin-based smart construction method of claim 1, wherein, the encapsulation of the data into a digital twin gene chain according to a time axis comprises the following steps: material performance decay curve wherein, is initial performance, is decay coefficient, is time; deviation sequence of actual time consumption and planned time consumption of key processes: wherein, is the deviation of actual time consumption and planned time consumption of the kth process, is the actual time consumption of the kth process, is the planned time consumption of the kth process, is the total number of processes. 9.The digital-twin-based smart construction method of claim 1, wherein, the support for operation and maintenance decisions comprises the following steps: in the operation and maintenance stage, the digital twin gene chain is introduced into an operation and maintenance digital twin model as initial parameters for building state evaluation; based on the gene chain, the influence factors of the construction process on the current state of the building are analyzed, and the structure aging trend and equipment maintenance cycle are predicted in combination with real-time monitoring data in the operation and maintenance stage; according to the prediction result, an operation and maintenance optimization scheme is generated, so that the reverse tracing and forward optimization of the construction stage data on the operation and maintenance decisions are realized; the reverse tracing comprises the following steps: according to the structure damage position detected in the operation and maintenance stage, the material performance attenuation curve and the process deviation sequence of the corresponding area in the construction stage are associated through the gene chain; the forward optimization comprises the following steps: based on the process deviation sequence in the gene chain, the equipment maintenance cycle model parameters are corrected, and a preventive maintenance scheme is generated.

10. A smart construction system based on digital twinning, characterized by, The method comprises the following steps: a data acquisition module configured to acquire multi-source time-series sensing data, load the data into a BIM model to generate an initial digital twin model, embed a causal correlation network of construction risk factors and progress nodes in the initial digital twin model, and calculate the directional and weighted correlation strength of node variables to target variables; a situation prediction module configured to, based on the causal correlation network, use an LSTM algorithm to dynamically calculate the influence weight of construction risk factors on progress nodes in real time, and generate a risk progress coupling situation prediction curve; The process optimization instruction module is configured to optimize the construction path through process topology reconstruction when the predicted curve deviates from a threshold value, and push the operation instruction with space-time constraints to the terminal; The deviation identification correction module is configured to identify the deviation type of the entity building and the twin model through a multi-modal sensor fusion technology, call a corresponding correction strategy, and update the weight of the causal association network using reinforcement learning; The gene chain operation and maintenance module is configured to extract core evolution data in the construction phase, encapsulate it into a digital twin gene chain according to a time axis, and support reverse tracing and forward optimization of operation and maintenance decisions.

Images