Intelligent beam yard information management system based on BIM and digital twinning

Abstract

The invention relates to the field of railway engineering construction management, and discloses an intelligent beam field information management system based on BIM and digital twinning, which comprises a twinning modeling unit, a data acquisition and feedback unit, an intelligent scheduling unit, a simulation training and migration unit, a visual interaction unit, a cross-field cooperation unit and a prediction analysis and repair unit. A three-dimensional twin model of a beam field is established through fusion of a BIM model and GIS coordinates, and virtual-real synchronization is realized in combination with real-time data collected by the Internet of Things; an improved reinforcement learning algorithm is adopted to generate a beam body production and transfer scheduling scheme, and strategy optimization is carried out through simulation training and transfer learning; the system can display beam body states, equipment tracks and path animations on a visual interface, and supports manual intervention and multi-beam field cooperative control. Application results show that the system can effectively reduce the total beam moving cost, improve the production efficiency and the pedestal utilization rate, and is suitable for intelligent management of a large-scale precast beam field.

Description

Technical Field

[0001] This invention relates to the field of railway engineering construction management, specifically to an intelligent beam yard information management system based on BIM and digital twins. Background Technology

[0002] With the continuous expansion of railway construction, precast beam yards play a crucial role in bridge engineering construction, undertaking core tasks such as beam prefabrication, storage, transportation, and erection. Their production management level directly affects project progress and construction quality. Currently, most beam yards employ independent production management systems, equipment monitoring systems, and construction scheduling systems. Data sources are scattered, and information updates are not timely, making it difficult to achieve dynamic management of the entire life cycle of the beams.

[0003] During beam yard operations, the storage location of beams, the occupancy status of piers, and the transport routes need to be constantly adjusted according to the construction schedule and beam erection plan. Due to the large number of beams, frequent overlap of operating equipment, and complex spatial layout, the traditional method of scheduling based on manual experience suffers from resource conflicts, route duplication, and equipment idleness, resulting in low overall operational efficiency. In addition, existing systems are mostly based on two-dimensional information, lacking a direct representation of the actual spatial layout of the beam yard. This makes it difficult for managers to fully grasp the status of beams and changes in the construction site, affecting the accuracy of decision-making.

[0004] In recent years, BIM technology has been widely used in the field of engineering construction, but most applications are still limited to the design and visualization stage and have failed to be combined with real-time data from the construction site. On the other hand, the dynamic operation characteristics of beam yards require the system to be able to monitor and coordinate the piers, tracks and equipment in real time. However, existing scheduling methods often do not consider spatial constraints and multi-objective balance, and cannot meet the high-frequency decision-making needs in complex construction environments. Therefore, we propose an intelligent beam yard information management system based on BIM and digital twins. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent beam yard information management system based on BIM and digital twins, which solves technical problems such as imperfect beam lifecycle management, low efficiency of platform and equipment resource scheduling, inability to synchronize physical scenes and virtual models in real time, and lack of global optimization in production scheduling.

[0006] The above-mentioned technical objective of the present invention is achieved through the following technical solution: The intelligent beam yard information management system based on BIM and digital twins includes: The twin modeling unit is used to perform parametric modeling of the beam fabrication area, beam storage area, and beam lifting area based on BIM, and to establish a spatial coordinate system of the beam yard in conjunction with GIS; the twin modeling unit also receives beam status, equipment status and environmental parameters collected by the Internet of Things, constructs a digital twin model of the beam yard, and performs topological relationship modeling of beams, pedestals and equipment through graph neural networks; The data acquisition and feedback unit is used to collect, clean, and standardize the beam production status, platform occupancy, equipment operation data, and environmental monitoring data, and synchronize the processed data to the digital twin model. At the same time, it receives the prediction output of the twin model and feeds it back to the physical beam field, realizing closed-loop interaction between the virtual model and the physical beam field. The intelligent scheduling unit is used to optimize the scheduling of beam moving paths and production tasks based on multi-agent reinforcement learning. During the scheduling process, a task and resource game mechanism is introduced to balance the dynamic matching relationship between beam production tasks and equipment resources, and the feasibility of the scheduling results is verified in combination with engineering physical constraints. The simulation training and transfer unit is used to perform multi-process parallel simulation training in the digital twin model, and to realize the transfer and application of simulation strategies to the actual beam field through domain adaptation and meta-learning. The visualization and interaction unit is used to dynamically display the beam position, equipment running trajectory and scheduling scheme in the BIM 3D model, and supports the visualization comparison and manual intervention of candidate layouts generated by the twin model. Cross-site collaborative units are used to share scheduling model parameters among multiple beam yards through federated learning, and combine knowledge graphs to achieve cross-project knowledge transfer and adaptive optimization. The predictive analysis and repair unit is used to predict equipment failures, weather changes, and platform utilization based on historical and real-time data. When an anomaly is detected, it triggers the correction of the digital twin model and dynamically reconstructs the scheduling strategy to achieve system self-learning and adaptive operation.

[0007] Preferably, the twin modeling unit performs integrated calibration of BIM local coordinates and GIS global coordinates, and uses affine transformation to achieve unified positioning of the beam, platform, and track:

[0008] In the formula, For rotation matrix, The translation vector is used to achieve spatial alignment between the physical beam field and the digital twin model. The calibration result serves as the geometric reference for subsequent scheduling and simulation. The calibration results serve as the geometric reference for subsequent scheduling states; The twin modeling unit abstracts beams, platforms, equipment, and track nodes as graph nodes, abstracts traffic relationships and operational constraints as graph edges, and uses graph convolution to dynamically represent the topology to generate state embeddings for scheduling.

[0009] In the formula, To add a self-loop adjacency matrix, This is the degree matrix. It is a nonlinear function, and its output is embedded as the state input of the intelligent scheduling unit; The twin modeling unit introduces attention weights with edge features into the graph representation to reflect the impact of curvature, slope, and congestion on traffic priority:

[0010] In the formula, , in order, are the edges The inverse of curvature, slope, and congestion characteristics. Used to dynamically adjust edge traversability and path priority; The twin modeling unit has the ability to generate self-evolving candidate layouts. The multi-objective function aimed at improving the pedestal utilization and shortening the beam-moving distance is as follows:

[0011] In the formula, To maximize the utilization of the pedestal, The total beam movement distance, To penalize conflicts and satisfy engineering constraints such as load capacity, clearance, and minimum turning radius, candidate layouts are generated for scheduling evaluation.

[0012] Preferably, the data acquisition and feedback unit performs residual detection on the device and environmental samples:

[0013] Abnormal threshold:

[0014] In the formula, For actual measurement, For prediction, For the residual mean and standard deviation; when > The system will trigger security degradation and border blocking signals to be written back to the twin model. The data acquisition and feedback unit injects IoT quantitative indicators into the scheduling reward, forming:

[0015] In the formula, For energy consumption, For the construction period; Risk items

[0016] In the formula, For load, For wind speed, For congestion level, As an immediate reward for the intelligent scheduling unit.

[0017] Preferably, the intelligent scheduling unit employs dual-network deep Q-learning to reduce estimation bias, and the target and update are as follows:

[0018] In the formula, Depend on As an immediate reward for the intelligent scheduling unit, For target network parameters, periodic soft updates are performed; used to optimize beam-moving paths and equipment coordination. The Q-network front-end of the intelligent scheduling unit adopts a residual convolutional structure:

[0019] In the formula, The sequence of convolution, normalization, activation, and convolution is used to stabilize deep network training and extract local spatial patterns of beam fields. The intelligent scheduling unit employs weighted priority replay to enhance the learning of rare conflict samples:

[0020] In the formula, For sample TD error, For smoothing terms, As a priority index, Importance weight, To correct the index; The intelligent scheduling unit introduces a task-resource game to balance task priority and resource consumption, and its benefits are as follows:

[0021] In the formula, For the task Allocate resources binary variables, For resource value, For task cost; It is used to solve for the optimal allocation and is linked with reinforcement learning strategies.

[0022] Preferably, the intelligent scheduling unit uses Nash equilibrium as the stopping criterion:

[0023] In the formula, For the task The optimal strategy at equilibrium To remove the task The allocation iteration ends when all tasks satisfy the above formula, taking into account the strategy combinations of other participants, and outputs a stable schedule; The intelligent scheduling unit applies a monotonic constraint in the multi-agent value aggregation:

[0024]

[0025] In the formula, For the first Local value function of an agent For the global value function, The function is monotonically increasing, ensuring that the local optimal action selection is equivalent to the global optimal action; this ensures the consistency between centralized training and decentralized execution, meaning that the individual greedy action is equivalent to the global greedy action. The intelligent scheduling unit applies a time window non-intersection constraint to operations on the same track segment:

[0026] In the formula, For the task The time intervals for entry and exit on this orbital segment, For the task The time interval; where the time window is calculated from the road segment length and the equipment speed level; used to avoid collisions and congestion.

[0027] Preferably, the simulation training and transfer units are aligned between the simulation domain and the real domain distribution using the maximum mean difference:

[0028] In the formula, For simulation samples, For real samples, It is a feature map used to reduce the bias between the real and virtual domains. The simulation training and transfer unit further employs adversarial domain adaptation:

[0029] In the formula, For feature generator, It serves as a domain discriminator, used to improve cross-domain unidentifiability and transfer robustness. The simulation training and transfer unit employs meta-learning to quickly adapt to weather, shift, and personnel disturbances, and updates accordingly.

[0030] In the formula, For model parameters, The learning rate is used for rapid convergence under new operating conditions. The simulation training and transfer learning units are trained using a joint loss:

[0031] In the formula, To reinforce learning loss, To combat domain loss, Penalties for violations of engineering regulations , , Weights are used to weigh performance, migration, and security.

[0032] Preferably, the cross-field collaborative unit adopts a federated average aggregated multi-beam field local model:

[0033] In the formula, For the first The beam yard is in Wheel parameters, Its sample size; used to formulate a global strategy without sharing the original data; The cross-field cooperative unit introduces near-end regularization for robust convergence in heterogeneous operating conditions:

[0034] In the formula, The regularization coefficient is... These are global parameters for the current round; used to mitigate drift caused by differences in data distribution across different beam yards. The cross-field collaborative unit adds differential privacy noise when uploading parameters:

[0035] In the formula, Set according to the privacy budget; used to prevent sensitive on-site data from being deduced from uploaded parameters.

[0036] Preferably, the visualization interaction unit visualizes the beam movement process in the BIM model through 3D trajectory rendering, using a parameterized trajectory function:

[0037] In the formula, The initial position, The time-varying velocity vector output by the scheduling unit. The trajectory curves are plotted in real time in the model to show the comparison between the actual beam path and the recommended path. The visualization interaction unit allows users to intervene in candidate paths through an interactive interface, defining an intervention correction function:

[0038] In the formula, The correction offset is input by the user; this correction is fed back to the intelligent scheduling unit in real time for secondary simulation and scheme recalculation.

[0039] Preferably, the prediction analysis and repair unit uses a time series prediction model to predict the beam yard operation indicators, based on:

[0040] In the formula, For historical sequence, For time window, This is a fitting function for a long short-term memory network, used to predict future equipment load and platform utilization. The predictive analysis and repair unit identifies anomalies using residual criteria:

[0041] like If it is, it is marked as an exception, and exceptions are used to trigger the repair mechanism; The predictive analysis and repair unit provides repair suggestions through an optimization function:

[0042] In the formula, For the cost of energy consumption, To the cost of delay, These are the balance coefficients; the solutions are... Used to generate adjusted production and scheduling plans.

[0043] Preferably, the topological embedding vector generated by the twin modeling unit is used as the state input of the Q-value function of the intelligent scheduling unit, and the input is used to ensure that the scheduling decision is dynamically consistent with the twin model; The Q-value output by the intelligent scheduling unit is extracted by residual convolution feature extraction and jointly solved with the task-resource game mechanism. The utility is optimized under the condition of multi-task Nash equilibrium to achieve unified decision-making on beam yard resource allocation and path selection. The joint loss function used in the simulation training and transfer unit during transfer learning is designed to ensure alignment between the virtual and real domains while maintaining scheduling performance. The visualization interaction unit uses a trajectory function. The rendered path result is synchronously bound to the candidate path output by the intelligent scheduling unit, and user interaction is used for correction. It is directly fed back to the scheduling network, forming a closed-loop correction mechanism; When the predictive analysis and repair unit detects abnormalities in future indicators, it generates an adjustment plan. The repair effect is then verified by simulation and remapped to the virtual beam yard scene in real time through twin modeling units before being sent to the actual beam yard.

[0044] In summary, the present invention has the following main beneficial effects: This invention deeply integrates BIM modeling, GIS coordinate system, IoT sensing, and digital twin technology to establish a three-dimensional virtual model synchronized with the actual operation of the beam yard, realizing dynamic mapping and visualization of beams, equipment, platforms, and track nodes. The system utilizes twin modeling units to achieve high-precision alignment between virtual and real spaces, providing a unified geometric benchmark for subsequent scheduling and simulation, thereby enabling dynamic linkage of data and status throughout the entire process of beam production, storage, and transportation.

[0045] By combining intelligent scheduling units with multi-agent reinforcement learning and task resource game mechanisms, the system can achieve dynamic matching and global optimal scheduling of beam production tasks and equipment resources under the premise of meeting engineering constraints. This significantly improves equipment utilization and platform turnover efficiency compared to traditional manual scheduling methods. Through simulation training and multi-process parallel training and domain adaptation mechanisms of the transfer unit, the system can quickly verify strategies in a digital twin environment and transfer them to the actual beam yard, reducing on-site trial and error costs and realizing self-learning and self-evolution of scheduling strategies.

[0046] The system utilizes a visualization and interactive unit to achieve 3D display and manual intervention correction of the beam yard's operational status. A cross-site collaboration unit, based on federated learning, enables parameter sharing and experience transfer among multiple beam yards. A predictive analysis and repair unit uses a time series model to identify potential anomalies in advance and dynamically correct strategies, giving the system self-diagnosis and self-repair capabilities. Overall, this invention demonstrates significant technical effectiveness in reducing the total cost of beam relocation, improving production organization efficiency, and ensuring construction safety, providing a replicable technical path for the intelligent construction of large-scale precast beam yards. Attached Figure Description

[0047] Figure 1 This is a system structure block diagram of the present invention; Figure 2 This is a structural diagram of the twin modeling unit of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1 refer to Figure 1 The intelligent beam yard information management system based on BIM and digital twins includes the following steps: The twin modeling unit is used to perform parametric modeling of the beam fabrication area, beam storage area, and beam lifting area based on BIM, and to establish a spatial coordinate system of the beam yard in conjunction with GIS; the twin modeling unit also receives beam status, equipment status and environmental parameters collected by the Internet of Things, constructs a digital twin model of the beam yard, and performs topological relationship modeling of beams, pedestals and equipment through graph neural networks; The data acquisition and feedback unit is used to collect, clean, and standardize the beam production status, platform occupancy, equipment operation data, and environmental monitoring data, and synchronize the processed data to the digital twin model. At the same time, it receives the prediction output of the twin model and feeds it back to the physical beam field, realizing closed-loop interaction between the virtual model and the physical beam field. The intelligent scheduling unit is used to optimize the scheduling of beam moving paths and production tasks based on multi-agent reinforcement learning. During the scheduling process, a task and resource game mechanism is introduced to balance the dynamic matching relationship between beam production tasks and equipment resources, and the feasibility of the scheduling results is verified in combination with engineering physical constraints. The simulation training and transfer unit is used to perform multi-process parallel simulation training in the digital twin model, and to realize the transfer and application of simulation strategies to the actual beam field through domain adaptation and meta-learning. The visualization and interaction unit is used to dynamically display the beam position, equipment running trajectory and scheduling scheme in the BIM 3D model, and supports the visualization comparison and manual intervention of candidate layouts generated by the twin model. Cross-site collaborative units are used to share scheduling model parameters among multiple beam yards through federated learning, and combine knowledge graphs to achieve cross-project knowledge transfer and adaptive optimization. The predictive analysis and repair unit is used to predict equipment failures, weather changes, and platform utilization based on historical and real-time data. When an anomaly is detected, it triggers the correction of the digital twin model and dynamically reconstructs the scheduling strategy to achieve system self-learning and adaptive operation.

[0050] The twin modeling unit is used to establish a synchronized virtual and real model of the beam yard. The system first uses BIM technology for parametric modeling, generating three-dimensional geometric models of the beam fabrication area, beam storage area, and beam lifting area. A geographic coordinate system for the beam yard is constructed through the GIS module to achieve geographic alignment of the model space.

[0051] The spatial mapping relationship uses the affine transformation formula:

[0052] In the formula, This is a rotation matrix, representing the directional mapping from the BIM coordinate system to the GIS coordinate system; This is a translation vector used to correct the difference between the origin and the coordinate system. and These represent local and global coordinate points, respectively. In this coordinate system, beams, equipment, and platforms are abstracted as graph nodes, and travel paths and constraints are abstracted as graph edges. The system uses a graph convolution algorithm to extract topological features.

[0053] In the formula, To add a self-loop adjacency matrix; This is the degree matrix; It is a nonlinear function; the output is embedded as the state input of the intelligent scheduling unit; For the first Layer node embedding; This is the weight matrix; The twin modeling unit introduces attention weights with edge features into the graph representation to reflect the impact of curvature, slope and congestion on traffic priority;

[0054] In the formula, , in order, are the edges The inverse of curvature, slope, and congestion characteristics. Used to dynamically adjust edge traversability and path priority; The twin modeling unit has the ability to generate self-evolving candidate layouts. The multi-objective function aimed at improving the pedestal utilization and shortening the beam-moving distance is as follows:

[0055] In the formula, To maximize the utilization of the pedestal, The total beam movement distance, To penalize conflicts and satisfy engineering constraints such as load capacity, clearance, and minimum turning radius, candidate layouts are generated for scheduling evaluation.

[0056] This unit transforms beam field structural information into computable topological vector embeddings, achieving a unified representation of virtual and physical spaces. Whenever the physical beam field changes, the system automatically updates the topological parameters to maintain synchronization with the digital twin. The twin modeling unit enables real-time mapping between the virtual model and the physical scene, providing a unified data foundation for subsequent scheduling and prediction.

[0057] The data acquisition and feedback unit collects real-time data from the beam yard through IoT devices, including abutment occupancy, equipment status, beam location, and environmental parameters. The collected dataset is represented as follows:

[0058] In the formula, For the first Sensor data, The length of the time series; The system performs anomaly detection and filtering on the collected data, using a residual detection formula:

[0059] Abnormal threshold:

[0060] In the formula, For actual measurement, For prediction, For the residual mean and standard deviation; when The system will trigger security degradation and border blocking signals to be written back to the twin model. The data acquisition unit and the twin modeling unit interact bidirectionally to achieve physical-virtual state synchronization. Updated data is standardized and written to the database for use by the scheduling and simulation modules. This module ensures the real-time performance and data reliability of the twin model, providing reliable input for subsequent scheduling.

[0061] The intelligent scheduling unit is used to generate the transport path and operation sequence of the beam. The intelligent scheduling unit employs dual-network deep Q-learning to reduce estimation bias. The objective and update are as follows:

[0062] In the formula, Depend on As an immediate reward for the intelligent scheduling unit, For target network parameters, periodic soft updates are performed; used to optimize beam-moving paths and equipment coordination. The Q-network front-end of the intelligent scheduling unit adopts a residual convolutional structure:

[0063] In the formula, The sequence of convolution, normalization, activation, and convolution is used to stabilize deep network training and extract local spatial patterns of beam fields; The intelligent scheduling unit employs weighted priority replay to enhance the learning of rare conflict samples:

[0064] In the formula, For sample TD error, For smoothing terms, As a priority index, Importance weight, To correct the index; The intelligent scheduling unit introduces a task-resource game to balance task priority and resource consumption, and its benefits are as follows:

[0065] In the formula, For the task Allocate resources binary variables, For resource value, For task cost; It is used to solve for the optimal allocation and is linked with reinforcement learning strategies.

[0066] The intelligent scheduling unit uses Nash equilibrium as the stopping criterion.

[0067] In the formula, For the task The optimal strategy at equilibrium To remove the task The strategy combination of other participants is used to end the allocation iteration when all tasks satisfy the above formula, and a stable schedule is output. The intelligent scheduling unit applies a monotonic constraint in the multi-agent value aggregation:

[0068] In the formula, For the first Local value function of an agent For the global value function, The function is monotonically increasing, ensuring that the local optimal action selection is equivalent to the global optimal action; this ensures the consistency between centralized training and decentralized execution, meaning that the individual greedy action is equivalent to the global greedy action. The intelligent scheduling unit applies a time window non-intersection constraint to operations on the same track segment:

[0069] In the formula, For the task The time intervals for entry and exit on this orbital segment, For the task The time interval; where the time window is calculated from the road segment length and the equipment speed level; used to avoid collisions and congestion.

[0070] The system updates the Q-value through continuous interaction and simulation to form the optimal path planning strategy. The residual network structure is used for stable convergence, and the experience replay mechanism is used to accelerate learning.

[0071] Achieving the optimal solution for dynamic scheduling under multiple constraints reduces the conflict rate of beam-moving paths and improves operational efficiency.

[0072] The simulation training and transfer unit is used for parallel simulation training in a virtual beam field and to transfer virtual strategies to the real environment. The maximum mean difference is used to align the simulation domain and the real domain distribution.

[0073] In the formula, For simulation samples, For real samples, It is a feature map used to reduce the bias between the real and virtual domains. And an adversarial domain adaptive network is introduced:

[0074] In the formula, For feature generator, It serves as a domain discriminator, used to improve cross-domain unidentifiability and transfer robustness. The simulation training and transfer unit employs meta-learning to quickly adapt to weather, shift, and personnel disturbances, and updates accordingly.

[0075] In the formula, For model parameters, , The learning rate is used for rapid convergence under new operating conditions. The simulation training and transfer learning units are trained using a joint loss:

[0076] In the formula, To reinforce learning loss, To combat domain loss, Penalties for violations of engineering regulations Weights are used to weigh performance, migration, and security.

[0077] The system is trained in parallel in a simulation environment to generate scheduling model parameters; the transfer learning mechanism enables the virtual policy to adapt to the field environment through feature distribution matching, reducing the cost of trial and error in the field and improving the model's generalization and application security.

[0078] The visual interactive unit is used to dynamically display the status of beams, equipment, and paths in a BIM scenario. The system uses WebGL rendering to achieve 3D animation display and supports path editing and manual intervention.

[0079] The visualization interaction unit visualizes the beam movement process in the BIM model through 3D trajectory rendering, using a parameterized trajectory function:

[0080] In the formula, The initial position, The time-varying velocity vector output by the scheduling unit. The trajectory curves are plotted in real time in the model to show the comparison between the actual beam path and the recommended path. The visualization interaction unit allows users to intervene in candidate paths through an interactive interface, defining an intervention correction function:

[0081] In the formula, The correction offset is input by the user; this correction is fed back to the intelligent scheduling unit in real time for secondary simulation and scheme recalculation.

[0082] Users can modify the scheduling plan by dragging or clicking. The system can instantly detect the feasibility of the new path and whether the constraints are met, providing a scheduling environment with enhanced interpretability and interactivity, which is convenient for manual review and on-site execution.

[0083] The cross-field collaborative unit is used for resource collaborative scheduling among multiple beam fields, and the optimal allocation is solved using a multi-agent game framework.

[0084] The cross-field collaborative unit adopts a federated average aggregated multi-beam field local model:

[0085] In the formula, For the first The beam yard is in Wheel parameters, Its sample size; used to formulate a global strategy without sharing the original data; The cross-field cooperative unit introduces near-end regularization for robust convergence in heterogeneous operating conditions:

[0086] In the formula, The regularization coefficient is... These are global parameters for the current round; used to mitigate drift caused by differences in data distribution across different beam yards. The cross-field collaborative unit adds differential privacy noise when uploading parameters:

[0087] In the formula, Set according to the privacy budget; used to prevent sensitive on-site data from being deduced from uploaded parameters.

[0088] The predictive analysis and repair unit uses a time series prediction model to predict the beam yard's operational indicators, based on:

[0089] In the formula, For historical sequence, For time window, This is a fitting function for a long short-term memory network, used to predict future equipment load and platform utilization. The predictive analysis and repair unit identifies anomalies using residual criteria:

[0090] like If it is, it is marked as an exception, and exceptions are used to trigger the repair mechanism; The predictive analysis and repair unit provides repair suggestions through an optimization function:

[0091] In the formula, For the cost of energy consumption, To the cost of delay, These are the balance coefficients; the solutions are... Used to generate adjusted production and scheduling plans.

[0092] Example 2 A smart beam yard information management method based on BIM and digital twins is provided to illustrate the operation flow and collaboration mechanism of the system in Embodiment 1. This method operates based on the system structure described in Embodiment 1 and includes the following steps: For model initialization and twin modeling, the system uses BIM to perform 3D parametric modeling of the beam fabrication area, beam storage area, and beam lifting area, and combines GIS to achieve geographic coordinate calibration, forming a digital twin model corresponding to the actual beam yard. During the initialization phase, basic data such as piers, tracks, and equipment are loaded into this model, serving as a unified geometric benchmark for subsequent scheduling and simulation.

[0093] Real-time data acquisition and synchronization are achieved by continuously collecting beam yard operation data using IoT devices, including beam number, equipment location, environmental parameters, and platform occupancy status. The system periodically performs residual detection and data verification, synchronizing the cleaned and valid data to the twin model in real time, enabling bidirectional dynamic updates between the virtual and physical beam yards.

[0094] The intelligent scheduling and task generation system automatically constructs a task-resource relationship matrix after acquiring the beam yard status. It generates candidate scheduling schemes using reinforcement learning algorithms and balances task priority and equipment utilization by incorporating engineering constraints and game theory mechanisms. Once the scheduling result meets feasibility verification, it outputs the optimal path and work sequence.

[0095] Simulation training and transfer optimization: The system performs parallel simulation training of the scheduling strategy in a virtual twin environment, and uses a transfer learning mechanism to quickly adapt the simulation strategy to the real-world scenario, reducing on-site trial and error. Model parameters under different environments are automatically fine-tuned through a meta-learning algorithm, achieving cross-condition adaptiveness of the strategy.

[0096] The system features 3D visualization and interactive revision, visually displaying beam distribution and equipment movement trajectories within the BIM model. Users can view, pause, or modify scheduling results through the interface. Manually revised paths will automatically trigger system re-simulation and scheme updates, ensuring the scheme is safe, feasible, and meets site conditions.

[0097] Cross-site collaboration and knowledge transfer: When multiple beam yards are operating in parallel, the system shares model parameters based on a federated learning mechanism, without directly transmitting raw data, thus ensuring data security and model consistency. Through knowledge graph comparison, experience transfer and scheduling strategy reuse between different projects are achieved.

[0098] Predictive analysis and dynamic remediation: The system continuously monitors the beam yard's operational indicators and predicts potential anomalies using time-series models. When predicted values ​​deviate from thresholds, remediation strategies are automatically generated, adjusting scheduling schemes and resource allocation to ensure the continuity and stability of the beam yard's operations.

[0099] This method uses a twin model as its core, integrating data acquisition, AI scheduling, simulation optimization, and human interaction to form a closed-loop control system. Through real-time feedback and strategy iteration, it achieves self-learning and self-adaptation in beam yard production management.

[0100] Compared with traditional experience-based scheduling, this embodiment can significantly improve the utilization rate of the beam piers and production coordination, reduce path conflicts and energy consumption, and realize digital, intelligent and visual management of the entire beam yard production process.

[0101] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent beam yard information management system based on BIM and digital twins, characterized by: The twin modeling unit is used to perform parametric modeling of the beam fabrication area, beam storage area, and beam lifting area based on BIM, and to establish a spatial coordinate system of the beam yard in conjunction with GIS; the twin modeling unit also receives beam status, equipment status and environmental parameters collected by the Internet of Things, constructs a digital twin model of the beam yard, and models the topological relationship between beams, pedestals and equipment through graph neural networks; The data acquisition and feedback unit is used to collect, clean, and standardize the beam production status, platform occupancy, equipment operation data, and environmental monitoring data, and synchronize the processed data to the digital twin model. At the same time, it receives the prediction output of the twin model and feeds it back to the physical beam field, realizing closed-loop interaction between the virtual model and the physical beam field. The intelligent scheduling unit is used to optimize the scheduling of beam moving paths and production tasks based on multi-agent reinforcement learning. During the scheduling process, a task and resource game mechanism is introduced to balance the dynamic matching relationship between beam production tasks and equipment resources, and the feasibility of the scheduling results is verified in combination with engineering physical constraints. The simulation training and transfer unit is used to perform multi-process parallel simulation training in the digital twin model, and to realize the transfer and application of simulation strategies to the actual beam field through domain adaptation and meta-learning. The visualization and interaction unit is used to dynamically display the beam position, equipment running trajectory and scheduling scheme in the BIM 3D model, and supports the visualization comparison and manual intervention of candidate layouts generated by the twin model. Cross-site collaborative units are used to share scheduling model parameters among multiple beam yards through federated learning, and combine knowledge graphs to achieve cross-project knowledge transfer and adaptive optimization. The predictive analysis and repair unit is used to predict equipment failures, weather changes, and platform utilization based on historical and real-time data. When an anomaly is detected, it triggers the correction of the digital twin model and dynamically reconstructs the scheduling strategy to achieve system self-learning and adaptive operation.

2. The intelligent beam yard information management system based on BIM and digital twins according to claim 1, characterized in that, The twin modeling unit performs integrated calibration of BIM local coordinates and GIS global coordinates, and uses affine transformation to achieve unified positioning of the beam, platform, and track: In the formula, Let be a rotation matrix. The translation vector is used to achieve spatial alignment between the physical beam field and the digital twin model. The calibration result serves as the geometric reference for subsequent scheduling and simulation. The twin modeling unit abstracts beams, platforms, equipment, and track nodes as graph nodes, abstracts traffic relationships and operational constraints as graph edges, and uses graph convolution to dynamically represent the topology to generate state embeddings for scheduling. In the formula, To add a self-loop adjacency matrix, This is the degree matrix. It is a nonlinear function, and its output is embedded as the state input of the intelligent scheduling unit; The twin modeling unit introduces attention weights with edge features into the graph representation to reflect the impact of curvature, slope, and congestion on traffic priority: In the formula, , in order, are the edges ( The inverse of curvature, slope, and congestion characteristics of ( ). Used to dynamically adjust edge traversability and path priority; The twin modeling unit has the ability to generate self-evolving candidate layouts. The multi-objective function aimed at improving the pedestal utilization and shortening the beam-moving distance is as follows: ; ; In the formula, To maximize the utilization of the pedestal, The total beam movement distance, To penalize conflicts and satisfy engineering constraints such as load capacity, clearance, and minimum turning radius, candidate layouts are generated for scheduling evaluation.

3. The intelligent beam yard information management system based on BIM and digital twins according to claim 2, characterized in that, The data acquisition and feedback unit performs residual detection on equipment and environmental samples: Abnormal threshold: In the formula, For actual measurement, For prediction, For the residual mean and standard deviation; when > The system will trigger security degradation and border blocking signals to be written back to the twin model. The data acquisition and feedback unit injects IoT quantitative indicators into the scheduling reward, forming: ; In the formula, For energy consumption, For the construction period; Risk items ; In the formula, For load, For wind speed, For congestion level, As an immediate reward for the intelligent scheduling unit.

4. The intelligent beam yard information management system based on BIM and digital twins according to claim 3, characterized in that, The intelligent scheduling unit employs dual-network deep Q-learning to reduce estimation bias, with the objective and update being: ; In the formula, Depend on As an immediate reward for the intelligent scheduling unit, For target network parameters, periodic soft updates; Used to optimize beam-moving paths and equipment coordination; The Q-network front-end of the intelligent scheduling unit adopts a residual convolutional structure: In the formula, The sequence of convolution, normalization, activation, and convolution is used to stabilize deep network training and extract local spatial patterns of beam fields. The intelligent scheduling unit employs weighted priority replay to enhance the learning of rare conflict samples: ; ; In the formula, For sample TD error, For smoothing terms, As a priority index, Importance weight, To correct the index; The intelligent scheduling unit introduces a task-resource game to balance task priority and resource consumption, and its benefits are as follows: In the formula, For the task Allocate resources binary variables, For resource value, For task cost; It is used to solve for the optimal allocation and is linked with reinforcement learning strategies.

5. The intelligent beam yard information management system based on BIM and digital twins according to claim 4, characterized in that, The intelligent scheduling unit uses Nash equilibrium as the stopping criterion. ; In the formula, For the task The optimal strategy at equilibrium To remove the task The allocation iteration ends when all tasks satisfy the above formula, taking into account the strategy combinations of other participants, and outputs a stable schedule; The intelligent scheduling unit applies a monotonic constraint in the multi-agent value aggregation: ; In the formula, For the first Local value function of an agent For the global value function, The function is monotonically increasing, ensuring that the local optimal action selection is equivalent to the global optimal action; this ensures the consistency between centralized training and decentralized execution, meaning that the individual greedy action is equivalent to the global greedy action. The intelligent scheduling unit applies a time window non-intersection constraint to operations on the same track segment: In the formula, For the task The time intervals for entry and exit on this orbital segment, For the task Time interval; The time window is calculated based on the road segment length and the equipment speed level; Used to avoid collisions and blockages.

6. The intelligent beam yard information management system based on BIM and digital twins according to claim 5, characterized in that, The simulation training and transfer units align the simulation domain and the real domain distribution using the maximum mean difference: In the formula, For simulation samples, For real samples, It is a feature map used to reduce the bias between the real and virtual domains. The simulation training and transfer unit further employs adversarial domain adaptation: In the formula, For feature generator, It serves as a domain discriminator, used to improve cross-domain unidentifiability and transfer robustness. The simulation training and transfer unit employs meta-learning to quickly adapt to weather, shift, and personnel disturbances, and updates accordingly. ; In the formula, For model parameters, , The learning rate is used for rapid convergence under new operating conditions. The simulation training and transfer learning units are trained using a joint loss: In the formula, To reinforce learning loss, To combat domain loss, Penalties for violations of engineering regulations , , As weight; Used to balance performance, migration, and security.

7. The intelligent beam yard information management system based on BIM and digital twins according to claim 6, characterized in that, The cross-field collaborative unit adopts a federated average aggregated multi-beam field local model: In the formula, For the first The beam yard is in Wheel parameters, Its sample size; Used to form a global strategy without sharing the original data; The cross-field cooperative unit introduces near-end regularization for robust convergence in heterogeneous operating conditions: In the formula, The regularization coefficient is... These are global parameters for the current round; used to mitigate drift caused by differences in data distribution across different beam yards. The cross-field collaborative unit adds differential privacy noise when uploading parameters: In the formula, Set according to the privacy budget; used to prevent sensitive on-site data from being deduced from uploaded parameters.

8. The intelligent beam yard information management system based on BIM and digital twins according to claim 7, characterized in that, The visualization interaction unit visualizes the beam movement process in the BIM model through 3D trajectory rendering, using a parameterized trajectory function: In the formula, The initial position, The time-varying velocity vector output by the scheduling unit. The trajectory curves are plotted in real time in the model to show the comparison between the actual beam path and the recommended path. The visualization interaction unit allows users to intervene in candidate paths through an interactive interface, defining an intervention correction function: In the formula, Corrected offset for user input; The correction is fed back to the intelligent scheduling unit in real time for secondary simulation and scheme recalculation.

9. The intelligent beam yard information management system based on BIM and digital twin as described in claim 8, characterized in that: The predictive analysis and repair unit uses a time series prediction model to predict the beam yard's operational indicators, based on: In the formula, For historical sequence, For time window, This is a fitting function for a long short-term memory network, used to predict future equipment load and platform utilization. The predictive analysis and repair unit identifies anomalies using residual criteria: like If it is, it is marked as an exception, and exceptions are used to trigger the repair mechanism; The predictive analysis and repair unit provides repair suggestions through an optimization function: In the formula, For the cost of energy consumption, To the cost of delay, These are the balance coefficients; the solutions are... Used to generate adjusted production and scheduling plans.

10. The intelligent beam yard information management system based on BIM and digital twin as described in claim 9, characterized in that: The topological embedding vector generated by the twin modeling unit serves as the state input to the Q-value function of the intelligent scheduling unit, and the input is used to ensure that the scheduling decision remains consistent with the dynamics of the twin model. The Q-value output by the intelligent scheduling unit is extracted by residual convolution feature extraction and jointly solved with the task-resource game mechanism. The utility is optimized under the condition of multi-task Nash equilibrium to achieve unified decision-making on beam yard resource allocation and path selection. The joint loss function used in the simulation training and transfer unit during transfer learning is designed to ensure alignment between the virtual and real domains while maintaining scheduling performance. The visualization interaction unit uses a trajectory function. The rendered path result is synchronously bound to the candidate path output by the intelligent scheduling unit, and user interaction is used for correction. It is directly fed back to the scheduling network, forming a closed-loop correction mechanism; When the predictive analysis and repair unit detects abnormalities in future indicators, it generates an adjustment plan. The repair effect is then verified by simulation and remapped to the virtual beam yard scene in real time through twin modeling units before being sent to the actual beam yard.

Images