Beam field production progress monitoring method based on total factor cooperation and related equipment

Abstract

The application discloses a beam yard production progress monitoring method based on full-element cooperation and related equipment, and through real-time collection of multi-source heterogeneous data, cleaning, alignment and fusion are carried out by using a space-time coding algorithm, and a dynamic full-element data set with unified space-time reference is constructed. Based on the data, a multi-agent system including three types of intelligent agents of beam yard production, offline construction and logistics transportation is constructed, each intelligent agent analyzes field data in parallel, extracts features and outputs local state and event log. The central digital twin engine aggregates these information, drives the real-time evolution and visualization of the digital twin model fused with BIM and GIS, and forms global linkage monitoring. Finally, through the double-layer architecture of "local prediction + global collaborative deduction", each intelligent agent performs short-term local prediction, the central intelligent agent performs cross-field interaction simulation and global risk judgment, and a closed loop from real-time monitoring to trend prediction and active early warning is realized.

Description

Technical Field

[0001] This invention relates to the field of industrial production monitoring technology, specifically to a method for monitoring the production progress of beam yards based on full-element collaboration. Background Technology

[0002] Currently, in the construction of large-scale linear projects such as railways and highways, the production progress of precast beam yards is a key factor restricting the overall project schedule. The production and erection of beams at the yard are closely linked to the construction of subgrade, bridges, and tunnels, forming a complex and dynamic engineering system. Traditional schedule management relies mainly on manual inspections, report summaries, and regular meetings, making it difficult to achieve real-time and accurate coordination between different stages. Managers struggle to fully grasp the true impact of beam yard production on subgrade construction, leading to schedule decisions often lagging behind actual on-site changes, thus affecting overall project efficiency and cost control.

[0003] Current technologies for monitoring the production progress of beam yards typically employ several isolated or partially combined methods. These include: 1) static or periodically updated 3D visualization based on Building Information Modeling (BIM); 2) data collection and monitoring of specific locations or processes using IoT devices such as video surveillance and sensors; and 3) inputting and tracking key milestones and their actual completion through project management software. While these methods achieve some degree of information digitization and partial visualization, they often focus on a single dimension or post-event recording. Inconsistent data formats and asynchronous updates between systems create isolated "information silos," failing to form an organic whole covering all elements of "people, machines, materials, methods, and environment."

[0004] The existing technical solutions have significant shortcomings. First, the monitoring depth is insufficient, lacking dynamic correlation analysis of the intrinsic influencing factors of progress, making it difficult to transition from "status display" to "trend prediction," and the early warning function is weak and lagging. Second, the monitoring breadth is limited; there is a lack of real-time, automated data linkage and collaborative analysis between the beam yard progress and the construction progress of various offline sub-projects, making it impossible to effectively identify systemic delay risks caused by overlapping processes and resource competition. Finally, the decision support capability is weak; existing systems mostly remain at the data presentation level and cannot provide simulation-verified optimization solutions for progress adjustments based on multi-source data and intelligent algorithms, and decision-making still relies heavily on personal experience.

[0005] In summary, the core technical problem that existing technologies have failed to solve is: how to break down the data and business barriers between the beam yard and the offline construction, and build a collaborative progress monitoring system that deeply integrates digital models and physical entities, can perceive the status of all elements in real time, intelligently predict progress risks, and provide closed-loop decision support, so as to realize the transformation from passive response to active intervention, and from local optimization to global collaboration, and ensure the efficient and accurate advancement of large-scale linear projects. Summary of the Invention

[0006] Based on the problems raised in the background technology above, the purpose of this invention is to provide a method and related equipment for monitoring the production progress of beam yards based on full-element collaboration. This solves the problems of insufficient monitoring depth, lack of dynamic correlation analysis of the intrinsic influencing factors of progress, difficulty in achieving the leap from "current status display" to "trend prediction", weak and lagging early warning function; limited monitoring breadth, lack of real-time and automated data linkage and collaborative analysis between beam yard progress and the construction progress of various offline sub-sections, and inability to effectively identify systemic delay risks caused by process overlap and resource competition; and weak decision support capability, with existing systems mostly remaining at the data presentation level, unable to provide simulation-verified optimization schemes for progress adjustment based on multi-source data and intelligent algorithms, and decision-making still highly dependent on personal experience.

[0007] This invention is achieved through the following technical solution:

[0008] The first aspect of this invention provides a method for monitoring the production progress of beam yards based on full-element collaboration, comprising the following steps:

[0009] Step S1: Collect multiple types of data through the IoT edge gateway of the beam yard and various construction sites, and generate a spatiotemporally unified dynamic full-element dataset after cleaning and alignment by the spatiotemporal coding algorithm.

[0010] Step S2: Construct a multi-agent system, which includes beam yard production, offline construction and logistics transportation. Each agent parses the corresponding domain subset of the dynamic full-element dataset in parallel and extracts features, and outputs local state vectors and event logs.

[0011] Step S3: Aggregate the output results of each intelligent agent through the central digital twin engine, drive the synchronous evolution and visualization of the digital twin model that integrates BIM and physical scene, and generate a virtual and real synchronized full-element digital twin scene.

[0012] Step S4: Each agent performs local progress prediction based on its local lightweight model, and the central agent aggregates global information and performs cross-domain inference through a deep collaborative analysis model to generate progress-related analysis results.

[0013] Step S5: The central intelligent agent generates a progress adjustment strategy based on the progress correlation analysis results. After verification and evaluation through digital twin scenario simulation, it outputs an optimized strategy set, forming a closed-loop feedback.

[0014] In the above technical solution, by deploying Internet of Things edge gateways at the beam yard and key nodes of each off-site construction, multi-dimensional heterogeneous data such as personnel, equipment, material consumption, and environmental parameters are collected in real time. Subsequently, the spatio-temporal coding algorithm is used to clean, align, and fuse these raw data. This algorithm can assign unified timestamps and spatial coordinates to all data points, and solve the differences in sampling frequencies, data formats, and precisions of different devices, and finally generate a dynamic all-element dataset with consistent spatio-temporal benchmarks and capable of correlation analysis. Through the standardized spatio-temporal coding algorithm, the spatio-temporal benchmark unification and real-time fusion across regions and systems are achieved at the data source, laying a reliable data foundation for subsequent in-depth dynamic correlation analysis and fundamentally breaking the data barriers.

[0015] A multi-agent system including three types of agents: beam yard production, off-site construction, and logistics transportation is constructed. Each agent parses in parallel the subset of the all-element dataset related to its professional field, extracts key features, and outputs a vector representing its local state and a structured log recording key events. A distributed and domain-specialized multi-agent architecture is introduced to achieve parallel and intelligent interpretation of data and generate event logs rich in semantics, providing a structured information source for identifying complex risks such as process intersections and resource competitions.

[0016] Through a central digital twin engine, the local state vectors and event logs from each agent are received and aggregated in real time. The engine uses this information to drive a digital twin model that integrates a refined BIM model and a GIS geographical scene to synchronously evolve and render, thereby generating a visual twin scene that is highly synchronized with the physical construction site in spatio-temporal state and can intuitively display the operating conditions of all elements. A digital twin that can aggregate multi-source real-time data and achieve dynamic mapping and two-way drive between physical entities and digital models is constructed. It fuses the scattered local states into a unified global situation panoramic view, realizing the transformation of monitoring from single-point discreteness to global linkage.

[0017] A two-layer prediction architecture is adopted. Locally at each agent, short-term and local progress predictions are made based on historical and real-time data in its field; at the central agent, all local states and prediction results are aggregated, and a deep collaborative analysis model is run. This model can simulate the interactions and dependencies between beam yard production, logistics transportation, and off-site construction, and conduct cross-domain global progress deduction and risk judgment. Through the two-layer model of "local prediction + global collaborative deduction", the timeliness and granularity of prediction are improved. By simulating the interactions between multi-agents, early and proactive identification of systematic delay risks is achieved, completing the key leap from current situation display to trend prediction and risk warning.

[0018] In one optional embodiment, the dynamic full-element dataset corresponds to a domain subset, which includes: dividing the dynamic full-element dataset into several domain subsets according to each agent domain, and constructing several domain subsets into an agent environment; wherein, the several domain subsets include: a precast beam production full-cycle domain subset, a bridge substructure construction domain subset, and a transportation network domain subset;

[0019] Among them, the environment for constructing a subset of the precast beam production lifecycle domain as a beam yard production intelligence agent includes: constructing a production progress tensor for storing the status of each beam shape in each process, a resource status tensor for storing equipment operation status, and a quality risk matrix for storing deviation values;

[0020] The environment for constructing a subset of bridge offline construction domains into an offline construction intelligent agent includes: digitally reconstructing the physical construction space based on the subset of bridge offline construction domains, and constructing an engineering progress tensor for storing the process status of each construction section, a resource occupancy tensor for storing the status of key resources occupied by each spatial unit, a quality and safety matrix for storing quality and risk, and an environment vector for storing environmental constraints based on the digitally reconstructed physical construction space.

[0021] The environment for constructing a subset of the transportation network domain into a logistics transportation intelligent agent includes: digitally reconstructing the physical space of the transportation network based on the subset of the transportation network domain, and constructing a transportation progress tensor, a vehicle resource tensor, and a road network state tensor based on the digitally reconstructed physical space of the transportation network.

[0022] In one optional embodiment, the beam yard production agent analyzes a subset of the entire precast beam production lifecycle and extracts features, including:

[0023] The flow rate feature and the time deviation feature are extracted from the production progress tensor, and the progress is predicted based on the flow rate feature and the time deviation feature to obtain the progress prediction feature;

[0024] Perform operational state analysis on the resource state tensor to obtain operational state transition quantities and resource utilization quantities. Use the operational state transition quantities and resource utilization quantities to perform resource competition analysis to obtain resource competition topology features.

[0025] The construction quality characteristics are obtained by using the aforementioned quality risk matrix for construction quality analysis.

[0026] The progress prediction features, resource competition topology features, and construction quality features are spatially quantized and encoded to generate a beam yard production state vector.

[0027] In one optional embodiment, the offline construction agent parses a subset of the bridge offline construction domain and extracts features, including:

[0028] The project progress tensor is scanned periodically along the time axis, and spatial position transition detection is performed. Based on the detected spatial position transitions, the process time is analyzed and extracted to obtain the project entity trajectory and progress characteristics.

[0029] Spatial conflict detection is performed on the resource occupancy tensor, and resource competition analysis is performed based on the detected spatial conflicts to obtain the characteristics of construction interference and resource competition.

[0030] Risk assessment is performed using the aforementioned quality and safety matrix to generate risk evolution characteristics;

[0031] The environmental vectors are used to perform environmental assessment and generate environmental response and constraint features.

[0032] The project entity trajectory and progress characteristics, construction interference and resource competition characteristics, risk evolution characteristics, and environmental response and constraint characteristics are spatially quantified and encoded to generate an offline construction state vector.

[0033] In one optional embodiment, the logistics transportation agent parses a subset of the transportation network domain and extracts features, including:

[0034] Scan the transportation progress tensor and perform transportation trajectory transition detection. Extract transportation trajectory features based on the results of the transportation trajectory transition detection.

[0035] Extract vehicle performance features from the vehicle resource tensor;

[0036] The road network capacity is determined and bottleneck sections are identified by the road network state tensor. The dynamic characteristics of the road network are constructed based on the road network capacity and bottleneck sections.

[0037] The transportation trajectory features, vehicle efficiency features, and road network dynamic features are spatially quantized and encoded to generate a logistics transportation state vector.

[0038] In one alternative embodiment, each agent performs local progress prediction based on a local lightweight model, including:

[0039] The beam yard production agent determines the beam yard production actions based on the beam yard production state vector in the output local state vector through a beam yard production state-action mapping mechanism; the beam yard production actions include: plan adjustment actions, resource scheduling actions, and quality intervention actions;

[0040] The offline construction agent determines offline construction actions based on the offline construction state vector in the output local state vector through an offline construction state-action mapping mechanism; the offline construction actions include: progress control and scheduling actions, dynamic resource allocation actions, quality and safety intervention actions, and environmental response and risk mitigation actions.

[0041] The logistics transportation agent determines logistics transportation actions based on the logistics transportation state vector in the output local state vector through a logistics transportation state-action mapping mechanism; the logistics transportation actions include: dynamic path optimization actions and vehicle resource scheduling actions.

[0042] In one optional embodiment, the central agent aggregates global information and performs cross-domain inference through a deep collaborative analysis model, including:

[0043] The received local optimization actions of each agent are structured and parsed, and each local optimization action is parsed into action elements containing information on type, subject, target spatiotemporal window, resource requirement / release and expected impact. A cross-domain action intent network is then constructed based on the action elements.

[0044] Based on the aforementioned cross-domain action intent network, global resource coupling analysis and spatiotemporal interference analysis are performed to detect cross-domain hard conflicts, soft bottlenecks, spatiotemporal interference, and supply and demand chain mismatch risks.

[0045] A cross-domain coupled simulation model is established, with the cross-domain action intent network as the input to the cross-domain coupled simulation model, and a preset uncertainty disturbance is injected. Multi-scenario simulation is run to evaluate the impact of the local optimization action on the global target index and generate progress correlation analysis results.

[0046] A second aspect of the present invention provides a beam yard production progress monitoring system based on full-element collaboration, comprising:

[0047] The data acquisition module is used to collect various types of data through the IoT edge gateway of the beam yard and various construction sites, and generate a dynamic full-element dataset with spatiotemporal uniformity after cleaning and alignment by the spatiotemporal coding algorithm.

[0048] The local optimization module is used to construct a multi-agent system, which includes beam yard production, offline construction and logistics transportation. Each agent parses the corresponding domain subset of the dynamic full-element dataset in parallel and extracts features, and outputs local state vectors and event logs.

[0049] The twin scene module is used to aggregate the output results of each intelligent agent through the central digital twin engine, drive the synchronous evolution and visualization of the digital twin model that integrates BIM and physical scene, and generate a virtual and real synchronized full-element digital twin scene.

[0050] The cross-domain extrapolation module is used by each agent to perform local progress prediction based on its local lightweight model, while the central agent aggregates global information and performs cross-domain extrapolation through a deep collaborative analysis model to generate progress-related analysis results.

[0051] The closed-loop feedback module is used by the central intelligent agent to generate progress adjustment strategies based on the progress-related analysis results. After being verified and evaluated by digital twin scenario simulation, it outputs an optimized strategy set, forming a closed-loop feedback.

[0052] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a method for monitoring the production progress of a beam yard based on full-element collaboration.

[0053] The fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for monitoring the production progress of a beam yard based on full-element collaboration.

[0054] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0055] By adopting a two-layer model of "local prediction + global collaborative inference", the timeliness and granularity of prediction are improved. By simulating the interaction between multiple agents, early and proactive identification of systemic delay risks is achieved, completing a key leap from displaying the current situation to trend prediction and risk warning. Attached Figure Description

[0056] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0057] Figure 1 This is a flowchart illustrating the beam yard production progress monitoring method based on full-element collaboration provided in Embodiment 1 of the present invention.

[0058] Figure 2 This is a schematic diagram of the beam yard production progress monitoring system based on full-element collaboration provided in Embodiment 2 of the present invention;

[0059] Figure 3 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0061] Example 1

[0062] Figure 1 This is a flowchart illustrating the beam yard production progress monitoring method based on full-element collaboration provided in Embodiment 1 of the present invention. Figure 1 As shown, the beam yard production progress monitoring method based on full-element collaboration includes the following steps:

[0063] Step S1: Collect multiple types of data through the IoT edge gateway of the beam yard and various construction sites, and generate a spatiotemporally unified dynamic full-element dataset after cleaning and alignment by the spatiotemporal coding algorithm.

[0064] Step S2: Construct a multi-agent system, which includes beam yard production, offline construction and logistics transportation. Each agent parses the corresponding domain subset of the dynamic full-element dataset in parallel and extracts features, and outputs local state vectors and event logs.

[0065] Step S3: Aggregate the output results of each intelligent agent through the central digital twin engine, drive the synchronous evolution and visualization of the digital twin model that integrates BIM and physical scene, and generate a virtual and real synchronized full-element digital twin scene.

[0066] Step S4: Each agent performs local progress prediction based on its local lightweight model, and the central agent aggregates global information and performs cross-domain inference through a deep collaborative analysis model to generate progress-related analysis results.

[0067] Step S5: The central intelligent agent generates a progress adjustment strategy based on the progress correlation analysis results. After verification and evaluation through digital twin scenario simulation, it outputs an optimized strategy set, forming a closed-loop feedback.

[0068] It should be noted that by deploying IoT edge gateways at the beam yard and various key offline construction nodes, multi-dimensional heterogeneous data such as personnel, equipment, material consumption, and environmental parameters are collected in real time. Subsequently, a spatiotemporal coding algorithm is used to clean, align, and fuse this raw data. This algorithm can assign a unified timestamp and spatial coordinates to all data points and resolve differences in sampling frequency, data format, and accuracy among different devices, ultimately generating a dynamic, full-element dataset with a consistent spatiotemporal benchmark and capable of correlation analysis. Through the standardized spatiotemporal coding algorithm, cross-regional and cross-system spatiotemporal benchmark unification and real-time fusion are achieved at the data source, laying a reliable data foundation for subsequent in-depth dynamic correlation analysis and fundamentally breaking down data barriers.

[0069] A multi-agent system is constructed, which consists of three types of agents: beam yard production, off-site construction, and logistics transportation. Each agent parses the subset of the full-element dataset related to its professional field in parallel, extracts key features, and outputs a vector representing its local state and a structured log recording key events. A distributed and domain-specialized multi-agent architecture is introduced, which realizes the parallel and intelligent interpretation of data and generates semantic-rich event logs, providing a structured information source for identifying complex risks such as process intersections and resource competitions.

[0070] Through a central digital twin engine, the local state vectors and event logs from each agent are received and aggregated in real time. The engine uses this information to drive the synchronous evolution and rendering of a digital twin model that integrates a refined BIM model and a GIS geographical scene, thus generating a visual twin scene that is highly synchronized with the physical construction site in terms of space-time state and can intuitively display the operating conditions of all elements. A digital twin is constructed that can aggregate multi-source real-time data and实现 the dynamic mapping and two-way drive between physical entities and digital models. It integrates scattered local states into a unified global situation panoramic map, achieving the transformation of monitoring from single-point discreteness to global linkage.

[0071] A two-layer prediction architecture is adopted. Locally at each agent, short-term and local progress predictions are made based on historical and real-time data in its field. At the central agent, all local states and prediction results are aggregated, and a deep collaborative analysis model is run. This model can simulate the interactions and dependencies between beam yard production, logistics transportation, and off-site construction, and conduct cross-domain global progress deduction and risk judgment. Through the two-layer model of "local prediction + global collaborative deduction", the timeliness and granularity of prediction are improved. By simulating the interactions between multi-agents, the early and proactive identification of systematic delay risks is achieved, completing the key leap from current situation display to trend prediction and risk warning.

[0072] Multiple types of data are collected through the IoT edge gateways at the beam yard and each off-site construction point, including: device sensing data (device location, operating status, number of work cycles, load data), material identification data (component ID, production batch, quality inspection status, inbound and outbound time), environmental monitoring data (temperature, humidity, wind speed, rainfall sensors deployed at the construction site and API data of weather stations), artificial terminal data (process inspection reports, safety inspection records, working hours records), and third-party system data (BIM design models, construction plans, structural parameters synchronized from the material management system, planned time, inventory quantity).

[0073] After being cleaned and aligned by the spatio-temporal coding algorithm, a spatio-temporally unified dynamic full-element dataset is generated, including:

[0074] Multiple types of data undergo edge preprocessing and time stamp alignment, including: using Kalman filtering and sliding window algorithms at the gateway layer to eliminate sensor jitter data; uniformly encoding multiple types of data into standardized event IDs and timestamps using event semantics; providing microsecond-level time synchronization for all edge gateways to eliminate clock drift between devices; and converting all location data to the unified construction coordinate system in real time.

[0075] The process involves spatiotemporal coding and data fusion of multiple types of data that have undergone edge preprocessing and time-stamp alignment. This includes generating a unique spatiotemporal code for each data record in the multiple types of data that have undergone edge preprocessing and time-stamp alignment, and establishing a correlation relationship. Finally, statistical anomaly detection is used to remove obvious outliers, and the rationality of spatial location is verified through the geometric constraints of the BIM model (e.g., components cannot appear in non-construction areas).

[0076] In one optional embodiment, the dynamic full-element dataset corresponds to a domain subset, which includes: dividing the dynamic full-element dataset into several domain subsets according to each agent domain, and constructing several domain subsets into an agent environment; wherein, the several domain subsets include: a precast beam production full-cycle domain subset, a bridge substructure construction domain subset, and a transportation network domain subset;

[0077] Among them, the environment for constructing a subset of the precast beam production lifecycle domain as a beam yard production intelligence agent includes: constructing a production progress tensor for storing the status of each beam shape in each process, a resource status tensor for storing equipment operation status, and a quality risk matrix for storing deviation values;

[0078] The environment for constructing a subset of bridge offline construction domains into an offline construction intelligent agent includes: digitally reconstructing the physical construction space based on the subset of bridge offline construction domains, and constructing an engineering progress tensor for storing the process status of each construction section, a resource occupancy tensor for storing the status of key resources occupied by each spatial unit, a quality and safety matrix for storing quality and risk, and an environment vector for storing environmental constraints based on the digitally reconstructed physical construction space.

[0079] The environment for constructing a subset of the transportation network domain into a logistics transportation intelligent agent includes: digitally reconstructing the physical space of the transportation network based on the subset of the transportation network domain, and constructing a transportation progress tensor, a vehicle resource tensor, and a road network state tensor based on the digitally reconstructed physical space of the transportation network.

[0080] It should be noted that in this embodiment, the dynamic full-element dataset is divided into several domain subsets according to the three major domains of precast beam production cycle, bridge offline construction, and transportation network. Each subset is constructed as the operating environment of the corresponding intelligent agent. The core function of this mechanism is to realize the paradigm shift from "data-centralized processing" to "domain-distributed intelligence"—by decoupling massive heterogeneous data according to professional domains, the beam yard production intelligent agent, the offline construction intelligent agent, and the transportation network intelligent agent can independently perceive, make decisions, and optimize in their respective exclusive digital environments. At the same time, cross-domain collaboration is achieved through the interaction of local state vectors, thereby supporting the accurate monitoring and real-time control of the full-element, full-chain, and dynamic production progress.

[0081] Specifically, the subset of the entire precast beam production cycle includes: each beam with a unique number (such as T001) as well as equipment such as rebar tying frames, formwork, steam curing sheds, tensioning equipment, grouting equipment, and beam storage areas.

[0082] The environment for constructing a subset of the entire precast beam production lifecycle as a beam yard production agent includes:

[0083] The tensor shape of the production schedule tensor is: number of beams N × number of processes S × state dimension D; where the number of processes is used to record the processes; the state dimension is used to track the spatiotemporal state of each beam in all processes, and is the basis for calculating the production cycle and identifying bottleneck processes.

[0084] The resource state tensor has the following shape: number of resource types T × state dimension M; where the number of resource types represents the resource types of each device, and the state dimension is used to globally control the location, status, and load of all key production resources, serving as the basis for resource conflict detection and dynamic scheduling.

[0085] The quality risk matrix is ​​a sparse correlation matrix, in which rows are associated with specific beam IDs and columns are associated with quality inspection items and deviation records. It is used to structure and store the full-dimensional quality data of each beam, supporting data-driven quality traceability, risk warning and process optimization.

[0086] An environment for constructing a subset of the transportation network domain into a logistics transportation intelligent agent includes:

[0087] To construct an operational environment for intelligent construction agents in the field of bridge subgrade construction, the first step is to digitally reconstruct the physical construction space. Based on high-precision BIM models and real-scene scanned point cloud data, the continuous three-dimensional construction site is discretized into a spatial unit grid system with hierarchical relationships and unique codes. Each unit not only defines its geometric boundaries and spatial coordinates, but also clarifies its structural attributes, adjacency relationships, and the macroscopic construction section to which it belongs. This establishes a digital twin foundation that is accurately mapped to the physical world and is computable and addressable, providing a unified coordinate framework for the spatial association of all subsequent state data.

[0088] On this digital spatial foundation, a core engineering schedule tensor is constructed to characterize the spatiotemporal evolution of the construction process. This tensor has a four-dimensional structure of "construction segment × sub-item × process × state," aiming to microscopically track the complete lifecycle of each engineering entity in the process flow. Its state dimension integrates multi-source information such as planned and actual engineering quantities, process start and end times, duration, and executing teams, enabling the intelligent agent to accurately identify transitions between processes, calculate net working time, and analyze efficiency deviations, thereby extracting deep characteristics of entity trajectories and schedule dynamics from massive schedule data.

[0089] The concurrently constructed resource occupancy tensor and quality and safety matrix complete environmental modeling from the dimensions of resource constraints and risks, respectively. The resource occupancy tensor, with its "spatial unit × resource type × state" structure, depicts in real-time the distribution, occupancy status, and load of key resources such as tower cranes and pump trucks in the digital space. It serves as the core basis for detecting spatial conflicts, analyzing resource competition patterns, and predicting scheduling bottlenecks. The quality and safety matrix, as a dynamically correlated sparse data structure, couples and correlates the measured quality data, safety risk indicators, and process control records of each construction segment. It supports the quantitative assessment and trend prediction of risks such as abnormal concrete strength development and systematic deviations in prestressing, achieving the explicitness and calculability of quality and safety risks.

[0090] Finally, by introducing a multi-dimensional environmental vector, external dynamic constraints are incorporated into the environmental perception system. This vector continuously feeds in real-time meteorological monitoring data, geological sensor readings, short-term early warnings and forecasts, and surrounding environmental monitoring information, providing the agent with key environmental conditions such as temperature, rainfall, wind speed, and groundwater level. Combined with the established progress, resource, and quality models, the agent can assess the impact of the environment on construction suitability, deduce chain response patterns such as "strong winds - high-altitude work stoppage," and thus generate forward-looking environmental response and risk mitigation strategies, forming a complete digital closed loop of the construction environment covering "space-progress-resources-quality-environment."

[0091] To construct a logistics transportation intelligent agent operating environment within the transportation network domain, a systematic digital reconstruction of the physical space of the transportation network is first required. On this digital road network foundation, a core transportation progress tensor is constructed to accurately characterize the spatiotemporal execution status of each transportation task. This tensor adopts a three-dimensional structure of "number of transportation tasks × number of transportation stages × state dimension," aiming to track the entire chain from the production of precast beams, their en route transportation, to on-site delivery. Its state dimension integrates real-time information such as planned and actual start and end times, current location coordinates, mileage traveled, remaining distance, estimated delays, current load, and cargo status. This enables the intelligent agent to dynamically evaluate the execution efficiency of each transportation trip, predict arrival times, and identify abnormal delays, thereby achieving global visualization and refined control of logistics progress.

[0092] The concurrently constructed vehicle resource tensor and road network state tensor enrich the environmental model from two dimensions: mobile vehicle supply and infrastructure capacity, respectively. The vehicle resource tensor, with a structure of "number of vehicles × state dimension," records in real-time the location, speed, load status, fuel consumption, driver fatigue coefficient, vehicle health, and current task assignment of each transport vehicle. This serves as a crucial basis for capacity assessment, task allocation, route planning, and maintenance early warning. The road network state tensor, as a dynamically updated network layer, with a structure of "number of road segments × state dimension," continuously aggregates multi-source information from traffic flow monitoring, variable message signs, weather services, and temporary traffic control announcements. It reflects in real-time the traffic speed, congestion index, accident status, weather impact, and temporary closures of each road segment.

[0093] By deeply integrating and analyzing the transportation progress tensor, vehicle resource tensor, and road network state tensor, the intelligent logistics agent gains a panoramic environmental awareness. The agent can not only dynamically plan the optimal route for individual trips based on real-time road conditions, but also predict road network bottlenecks, balance fleet load, avoid regional congestion from a global perspective, and initiate emergency replanning in extreme weather or emergencies. This upgrades transportation scheduling from passive task execution to a proactive, predictive, and networked resource optimization and allocation process, ultimately constructing an intelligent logistics transportation environment that covers all elements of "task-vehicle-road network" and supports real-time response and collaborative decision-making.

[0094] In one optional embodiment, the beam yard production agent analyzes a subset of the entire precast beam production lifecycle and extracts features, including:

[0095] The flow rate feature and the time deviation feature are extracted from the production progress tensor, and the progress is predicted based on the flow rate feature and the time deviation feature to obtain the progress prediction feature;

[0096] Perform operational state analysis on the resource state tensor to obtain operational state transition quantities and resource utilization quantities. Use the operational state transition quantities and resource utilization quantities to perform resource competition analysis to obtain resource competition topology features.

[0097] The construction quality characteristics are obtained by using the aforementioned quality risk matrix for construction quality analysis.

[0098] The progress prediction features, resource competition topology features, and construction quality features are spatially quantized and encoded to generate a beam yard production state vector.

[0099] It is important to note that the core of the intelligent agent for beam yard production in analyzing subsets and extracting features across the entire precast beam production cycle lies in constructing a quantitative cognitive system that comprehensively reflects the dynamic operation of the production system through in-depth mining and coupled analysis of multi-source heterogeneous data. This process begins with a refined analysis of the production schedule tensor. By calculating process throughput, actual production cycle time, flow rate index, and time-series time deviations, delay coefficients, and conflict indices, the agent can not only depict the real-time flow rate of beam segments on the production line but also identify key deviation patterns such as process imbalances, systemic timeouts, and queue backlogs. Based on these flow rate and time-series deviation characteristics, the agent further implements schedule prediction by estimating the remaining construction period of beams under fabrication, predicting the batch completion time distribution, and calculating the delivery risk entropy, which reflects the overall schedule uncertainty. This closely links the current micro-dynamics with the macro-delivery risk, providing a quantitative basis for forward-looking scheduling decisions.

[0100] Furthermore, the beam yard production intelligence agent deconstructs the resource state tensor to deeply analyze the operational health and utilization efficiency of production resources. This includes analyzing equipment state sequences to calculate state transition matrices, identifying key patterns such as frequent start-stop cycles and potential fault precursors, and evaluating mean time between failures (MTBF) and repair efficiency. Simultaneously, by calculating effective utilization rate, fragmentation index, and functional area synergy, the beam yard production intelligence agent can quantify the deep characteristics of equipment time utilization, revealing scheduling efficiency and resource coordination levels. By comprehensively utilizing these operational state transition quantities and resource utilization quantities, the beam yard production intelligence agent conducts resource competition analysis. By constructing equipment demand heatmaps, detecting resource dependency deadlocks, and identifying persistent bottlenecks, it ultimately forms a clear resource competition topology map, accurately locating key resource nodes that constrain the production process and their spatiotemporal distribution.

[0101] For feature extraction in the quality control dimension, the beam yard production agent focuses on the quality risk matrix and conducts in-depth construction quality analysis. This process employs a combination of mechanistic models and data-driven methods. For example, it uses maturity theory to calculate the equivalent age of concrete, fits strength development curves, and identifies curing anomalies; it analyzes prestressing tension curves, calculates the dual-control deviation of stress and elongation, assesses tension synchronization and pressure holding stability, and infers grouting fullness. These analyses go beyond judging a single result; they aim to capture the development trends of quality parameters, the stability of the process, and potential systematic deviations, thereby extracting deep construction quality features that can provide early warning of potential defects and support process optimization.

[0102] Finally, through a spatial quantization coding mechanism, the complex features extracted from the three dimensions of schedule, resources, and quality—including predictive schedule risks, topological resource competition patterns, and trending construction quality status—are fused and encoded into a high-dimensional, structured beam yard production state vector. This state vector is not a simple accumulation of features, but rather maps them to a computable state space through a unified mathematical framework, providing accurate and complete input for subsequent state assessment and intelligent decision-making.

[0103] In one optional embodiment, the offline construction agent parses a subset of the bridge offline construction domain and extracts features, including:

[0104] The project progress tensor is scanned periodically along the time axis, and spatial position transition detection is performed. Based on the detected spatial position transitions, the process time is analyzed and extracted to obtain the project entity trajectory and progress characteristics.

[0105] Spatial conflict detection is performed on the resource occupancy tensor, and resource competition analysis is performed based on the detected spatial conflicts to obtain the characteristics of construction interference and resource competition.

[0106] Risk assessment is performed using the aforementioned quality and safety matrix to generate risk evolution characteristics;

[0107] The environmental vectors are used to perform environmental assessment and generate environmental response and constraint features.

[0108] The project entity trajectory and progress characteristics, construction interference and resource competition characteristics, risk evolution characteristics, and environmental response and constraint characteristics are spatially quantified and encoded to generate an offline construction state vector.

[0109] It should be noted that the analysis and feature extraction of a subset of bridge construction data by the offline construction intelligent agent is a process of systematically building digital cognition from multi-source heterogeneous data. This process begins with deep temporal analysis of the project progress tensor. The offline construction intelligent agent scans cycle by cycle to detect spatial position transition events of construction entities in the process dimension, thereby accurately depicting the complete life cycle trajectory of each structure from start to completion. Based on these transition events, the offline construction intelligent agent further removes ineffective time such as environmental waiting and resource waiting, analyzes the net working time and process efficiency index, and finally integrates them to form the trajectory and progress characteristics of engineering entities that can reflect micro-operational efficiency and macro-progress rhythm, providing a quantitative basis for understanding the true driving force of the construction process.

[0110] Meanwhile, the offline construction intelligence agent focuses on spatial interference and resource competition issues during construction by deconstructing the resource occupancy tensor. It maps ongoing processes to digital spatial units, monitors conflicts in multiple processes requesting the same physical space in real time, and quantifies the frequency and duration of these conflicts. Furthermore, by constructing a spatiotemporal heatmap of resource demand, the offline construction intelligence agent identifies high-competition areas and resource bottleneck periods, and analyzes how resource shortages propagate along the process chain as schedule delays. This allows it to extract resource competition characteristics that reveal the root causes of construction interference and systemic resource contradictions, accurately pinpointing key nodes that restrict project flow.

[0111] In terms of risk perception, the offline construction intelligence agent utilizes a quality and safety matrix for proactive risk assessment. It quantifies the uncertainty and unpredictability of the overall progress status by calculating the distribution entropy value of schedule deviations; identifies potential quality degradation and systemic process deviations by monitoring the trends of concrete strength development curves and prestressing tension deviations; and dynamically calculates a safety risk index by combining risk operation exposure, historical accident rates, and the implementation status of control measures. These analyses collectively generate risk evolution characteristics covering the three levels of progress, quality, and safety, enabling early warning and quantitative assessment of potential hazards.

[0112] Finally, the offline construction agent comprehensively considers external environmental constraints, analyzes environmental vectors, identifies meteorological and geological warnings, and links them with specific construction activities to extract typical "environment-response" chain patterns such as "strong winds - suspension of high-altitude operations." Simultaneously, it calculates a comprehensive construction suitability index to quantify the intensity of environmental constraints on current operations. Ultimately, through a spatial quantification coding mechanism, the offline construction agent integrates the complex features extracted from the four dimensions of progress trajectory, resource competition, risk evolution, and environmental response into a high-dimensional, structured offline construction state vector. This vector concisely encapsulates the complete operational status, efficiency bottlenecks, and risk profile of the construction site in a specific spatiotemporal context, providing accurate and complete perceptual input for subsequent intelligent decision-making and adaptive control.

[0113] In one optional embodiment, the logistics transportation agent parses a subset of the transportation network domain and extracts features, including:

[0114] Scan the transportation progress tensor and perform transportation trajectory transition detection. Extract transportation trajectory features based on the results of the transportation trajectory transition detection.

[0115] Extract vehicle performance features from the vehicle resource tensor;

[0116] The road network capacity is determined and bottleneck sections are identified by the road network state tensor. The dynamic characteristics of the road network are constructed based on the road network capacity and bottleneck sections.

[0117] The transportation trajectory features, vehicle efficiency features, and road network dynamic features are spatially quantized and encoded to generate a logistics transportation state vector.

[0118] It should be noted that the intelligent logistics agent's analysis of subsets of the transportation network begins with a deep scan of the transportation progress tensor and trajectory transition detection. By continuously monitoring the state transitions of each transportation task during loading, in-yard transfer, mainline transportation, and unloading, the intelligent logistics agent accurately identifies key transition events such as "departure" and "delay," and simultaneously records the precise location and spatiotemporal dwell information of vehicles within the digital road network. Based on these detection results, the intelligent logistics agent extracts refined transportation trajectory features, including net travel time, waiting decomposition, path sequence, and speed profile, thereby depicting the real dynamic flow of goods within the transportation network and providing a core basis for evaluating transportation efficiency and reliability.

[0119] Meanwhile, the intelligent logistics and transportation agent focuses on the efficiency analysis of transportation vehicles, extracting key vehicle efficiency characteristics through in-depth deconstruction of vehicle resource tensors. This includes calculating vehicle time utilization, load utilization, and mileage utilization to quantify asset utilization efficiency; monitoring the evolution of status data of key components such as hydraulic systems, tires, and engines to build failure probability prediction models and achieve forward-looking health assessments; and correlating driver behavior to analyze speed stability, operational smoothness, and route adherence. These characteristics collectively constitute a comprehensive profile of the overall operational efficiency and potential risks of the transportation fleet, providing data support for capacity optimization and maintenance scheduling.

[0120] At the infrastructure level, intelligent logistics agents utilize road network state tensors to dynamically perceive network traffic conditions. They calculate the actual capacity and service level of each road segment in real time, quickly identifying dynamic bottlenecks caused by congestion, accidents, or weather, and assessing their impact range and dissipation time. Based on this, the intelligent logistics agent not only constructs dynamic characteristics reflecting the instantaneous load and vulnerability of the road network but also proactively generates and evaluates the cost and feasibility of alternative routes. This process transforms the static road network topology into a living system model that can be evaluated in real time and predictably avoided, forming the decision-making basis for dynamic route optimization.

[0121] Ultimately, the intelligent logistics agent integrates and standardizes the heterogeneous features extracted from the three dimensions of transportation trajectory, vehicle efficiency, and road network dynamics through a unified spatial quantization coding framework. It maps and integrates microscopic vehicle positions, mesoscopic path states, and macroscopic network bottleneck information into a high-dimensional, structured logistics transportation state vector. This vector concisely encapsulates the operational efficiency, resource health, and network connectivity of the entire logistics transportation system under specific spatiotemporal conditions, providing accurate and comprehensive situational awareness input for the intelligent agent's subsequent global resource scheduling, real-time path optimization, and risk emergency response.

[0122] In one alternative embodiment, each agent performs local progress prediction based on a local lightweight model, including:

[0123] The beam yard production agent determines the beam yard production actions based on the beam yard production state vector in the output local state vector through a beam yard production state-action mapping mechanism; the beam yard production actions include: plan adjustment actions, resource scheduling actions, and quality intervention actions;

[0124] The offline construction agent determines offline construction actions based on the offline construction state vector in the output local state vector through an offline construction state-action mapping mechanism; the offline construction actions include: progress control and scheduling actions, dynamic resource allocation actions, quality and safety intervention actions, and environmental response and risk mitigation actions.

[0125] The logistics transportation agent determines logistics transportation actions based on the logistics transportation state vector in the output local state vector through a logistics transportation state-action mapping mechanism; the logistics transportation actions include: dynamic path optimization actions and vehicle resource scheduling actions.

[0126] It should be noted that each agent performs local progress prediction based on a local lightweight model. The essence of this mechanism lies in decoupling the globally complex engineering system into multiple highly cohesive and loosely coupled autonomous domains. Each agent achieves a closed-loop "perception-analysis-decision" cycle within its own professional domain. The beam yard production agent, the offline construction agent, and the logistics and transportation agent serve as the "digital brains" of the three core links of precast beam manufacturing, on-site structural construction, and material flow, respectively. They continuously receive state inputs from their respective digital environments and use lightweight prediction models (such as time series analysis, queuing theory, or machine learning models) to quickly extrapolate local processes in the short future. This distributed prediction architecture effectively reduces the computational complexity of global modeling, improves the real-time response of the system, and enables each subdomain to make agile and forward-looking judgments on internal disturbances.

[0127] After completing state perception and progress prediction, each intelligent agent transforms quantitative cognition into specific control commands based on its built-in "state-action mapping mechanism." The beam yard production agent, based on its beam yard production state vector, generates planned adjustment actions (such as rearranging production sequences), resource scheduling actions (such as allocating templates and pedestals), and quality intervention actions (such as triggering stricter inspections of specific processes) through the mapping mechanism. These actions directly affect the production line control system, aiming to optimize production cycle time, resolve resource conflicts, and mitigate quality risks, ensuring the stable and timely output of precast beams.

[0128] The decision-making of offline construction intelligence agents is more comprehensive and complex, with their offline construction state vector-driven mapping mechanism outputting a multi-dimensional action set. This includes progress control and scheduling actions based on the spatiotemporal relationships of work processes (such as start / stop commands and work sequence adjustments), dynamic resource allocation actions for key equipment and spatial units (such as the reallocation of tower crane service areas), quality and safety intervention actions based on risk evolution characteristics (such as issuing special rectification or enhanced monitoring commands), and environmental response and risk mitigation actions in response to meteorological and geological changes (such as activating emergency plans and adjusting construction techniques). These actions together constitute the dynamic control of all elements of the construction site—"people, machines, materials, methods, and environment"—to ensure safe, high-quality, and efficient progress in complex operating environments.

[0129] The intelligent logistics transportation agent focuses on the smoothness and reliability of the transportation network. Its logistics transportation state vector is mainly transformed into two types of optimization actions through a mapping mechanism: First, dynamic route optimization, which, based on real-time road network status, traffic restrictions, and transportation tasks, replans the shortest or most cost-effective route for each trip, and even dynamically arranges the overall route of the fleet to avoid regional congestion; second, vehicle resource scheduling, which dynamically assigns transportation tasks, directs vehicle relocation, arranges maintenance windows, or allocates escort resources according to vehicle efficiency characteristics and task requirements. These actions aim to maximize capacity utilization, minimize transportation delays and cargo damage risks, and ensure that key components such as precast beams can accurately match the "cycle time requirements" of on-site construction, thereby effectively connecting the two major links of production and construction.

[0130] Furthermore, a reward function is designed to iteratively optimize each agent, including:

[0131] The reward function for the beam yard production agent aims to cultivate a behavioral pattern that pursues efficiency, stability, and high quality. Its core reward signals are closely designed around the core value stream of the production system: the agent receives a basic reward for producing qualified beam segments and additional rewards for on-time delivery. Simultaneously, by introducing continuous rewards for production line process balancing, efficient utilization of key resources, and stability of the quality process, the agent is incentivized to maintain a smooth and reliable production rhythm and proactively prevent defects. The function also incorporates strong binding penalties, negatively incentivizing inefficient and risky behaviors such as production safety accidents, drastic plan fluctuations due to poor scheduling, and long-term inventory backlogs. This guides the agent to maximize output and on-time performance while simultaneously considering resource costs and operational stability, achieving a dynamic balance among multiple objectives: quality, efficiency, and cost.

[0132] The reward function of the offline construction agent focuses on achieving a challenging balance between safety, schedule, and resources in complex and dynamic environments. This function assigns the highest weight to safety objectives, building a strong positive feedback loop for safety behavior by providing continuous rewards for every accident-free operating cycle and establishing incentives for successfully completing high-risk tasks. Simultaneously, it imposes a devastating penalty on any safety incident, sufficient to reset the training process, thus establishing an inviolable safety red line. Within the safety boundary, the function strongly drives the agent to optimize construction sequences to shorten the overall project duration through critical path milestone rewards and schedule catch-up rewards. Furthermore, by proactively resolving resource conflicts, load balancing, and adaptively adjusting to environmental disturbances, it guides the agent to intelligently allocate scarce time, space, and equipment resources, thereby achieving an optimal trade-off between schedule advancement and cost control while ensuring absolute safety and project quality.

[0133] The reward function for intelligent logistics agents aims to achieve timely, economical, and safe cargo delivery in uncertain road networks. With on-time delivery as the core reward source, the function directly incentivizes agents to optimize route planning and vehicle scheduling to combat delays. Building on this, it drives operational efficiency improvements by rewarding lower-cost route selection and higher vehicle occupancy rates, and ensures the safety of precast beams worth tens of millions of dollars in transit by rewarding lossless delivery and proactive risk avoidance. Simultaneously, the function imposes penalties for traffic violations, policy infractions, cargo damage, and high proportions of empty runs as hard constraints, ensuring that all efficiency-optimizing behaviors are based on compliance and safety, thereby shaping a smart, efficient, and prudent intelligent transportation scheduling strategy.

[0134] In one optional embodiment, the central agent aggregates global information and performs cross-domain inference through a deep collaborative analysis model, including:

[0135] The received local optimization actions of each agent are structured and parsed, and each local optimization action is parsed into action elements containing information on type, subject, target spatiotemporal window, resource requirement / release and expected impact. A cross-domain action intent network is then constructed based on the action elements.

[0136] Based on the aforementioned cross-domain action intent network, global resource coupling analysis and spatiotemporal interference analysis are performed to detect cross-domain hard conflicts, soft bottlenecks, spatiotemporal interference, and supply and demand chain mismatch risks.

[0137] A cross-domain coupled simulation model is established, with the cross-domain action intent network as the input to the cross-domain coupled simulation model, and a preset uncertainty disturbance is injected. Multi-scenario simulation is run to evaluate the impact of the local optimization action on the global target index and generate progress correlation analysis results.

[0138] It should be noted that the deep collaborative analysis model of the central agent begins with the refined decoding of local actions and the construction of a global intent network. This process first parses the local optimization actions from the beam yard, offline construction, and logistics transportation agents into standardized "action elements," extracting their core semantics such as type, executing entity, target spatiotemporal window, resource requirements, and expected impact. By arranging these action elements along a timeline and establishing logical dependencies between them (e.g., "beam production → transportation → erection"), the central agent constructs a panoramic "cross-domain action intent network." This network is no longer an isolated set of instructions, but rather a predetermined trajectory map depicting the interweaving of all planned activities in time, space, and resource dimensions over a future period, providing a structured data foundation for subsequent conflict detection and impact assessment.

[0139] Based on this intent network, the central agent initiates in-depth detection of global conflicts and bottlenecks. By establishing a unified global resource pool view, it overlays and compares dispersed resource demands, accurately identifying cross-domain hard conflicts (such as multiple processes vying for the same tower crane) and soft bottlenecks that reveal system vulnerabilities. Simultaneously, within a unified three-dimensional geographic information space, it overlays and analyzes construction-occupied areas and transportation routes to detect potential physical space interference and safety conflicts. Furthermore, the agent analyzes the matching degree between the beam yard's supply rhythm and on-site consumption demand, predicts the risk of supply-demand mismatch, and simulates how local risks such as severe weather propagate along the supply chain and amplify into global delays. This step aims to reveal the systemic contradictions and risks that may arise after the aggregation of locally optimal decisions.

[0140] To overcome the limitations of static analysis, the central agent then initiates a simulation-based dynamic evaluation process. It uses the constructed cross-domain action intent network as the core driving input, injecting it into a coupled simulation model that integrates simplified behavioral logic from various subdomains (production line, process network, traffic flow). By introducing uncertainties such as equipment failure and traffic congestion, and running hundreds of Monte Carlo simulations, the model can extrapolate the joint effects of various local action sets under complex real-world interactions and random disturbances. The simulation output is no longer a single prediction, but a probability distribution of the global project duration, overall cost, resource load, and cross-domain inventory levels, thereby generating schedule-related analysis results containing measures of uncertainty regarding the future project status.

[0141] Finally, based on the problems and opportunities revealed by the simulation evaluation, the central agent acts as a collaborative optimization engine. It analyzes the simulation results, identifies "key regulatory levers" that can achieve greater global benefits with smaller local costs, and generates higher-order collaborative strategies and constraints accordingly—such as issuing global priority directives, proposing cross-domain resource swaps, or negotiating time windows—rather than directly taking over control. By feeding these collaborative suggestions back to the local agents and guiding them to replan under new constraints, the central agent initiates multiple rounds of "suggestion-replanning-reevaluation" iterations. This process aims to guide decentralized decisions to converge to a Pareto optimal frontier, that is, to seek a balanced solution that collaboratively optimizes global objectives such as total project duration, total cost, and overall risk, while respecting the core interests of each subdomain.

[0142] Example 2

[0143] Figure 2 This is a schematic diagram of the beam yard production progress monitoring system based on full-element collaboration provided in Embodiment 2 of the present invention. Figure 2 As shown, the beam yard production progress monitoring system based on full-element collaboration includes:

[0144] The data acquisition module is used to collect various types of data through the IoT edge gateway of the beam yard and various construction sites, and generate a dynamic full-element dataset with spatiotemporal uniformity after cleaning and alignment by the spatiotemporal coding algorithm.

[0145] The local optimization module is used to construct a multi-agent system, which includes beam yard production, offline construction and logistics transportation. Each agent parses the corresponding domain subset of the dynamic full-element dataset in parallel and extracts features, and outputs local state vectors and event logs.

[0146] The twin scene module is used to aggregate the output results of each intelligent agent through the central digital twin engine, drive the synchronous evolution and visualization of the digital twin model that integrates BIM and physical scene, and generate a virtual and real synchronized full-element digital twin scene.

[0147] The cross-domain extrapolation module is used by each agent to perform local progress prediction based on its local lightweight model, while the central agent aggregates global information and performs cross-domain extrapolation through a deep collaborative analysis model to generate progress-related analysis results.

[0148] The closed-loop feedback module is used by the central intelligent agent to generate progress adjustment strategies based on the progress-related analysis results. After being verified and evaluated by digital twin scenario simulation, it outputs an optimized strategy set, forming a closed-loop feedback.

[0149] Example 3

[0150] Figure 3 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention, as shown below. Figure 3 As shown, the electronic device includes a processor 21, a memory 22, an input device 23, and an output device 24; the number of processors 21 in the computer device can be one or more. Figure 3 Taking a processor 21 as an example; the processor 21, memory 22, input device 23, and output device 24 in an electronic device can be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.

[0151] The memory 22, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules. The processor 21 executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory 22, thereby realizing the beam yard production progress monitoring method based on full-element collaboration in Embodiment 1.

[0152] The memory 22 may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a given function; the data storage area may store data created based on terminal usage. Furthermore, the memory 22 may include high-speed random access memory and non-volatile memory, such as at least one disk storage device, flash memory, or other non-volatile solid-state storage device. In some instances, the memory 22 may further include memory remotely located relative to the processor 21, which can be connected to the electronic device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0153] Input device 23 can be used to receive user input such as ID and password. Output device 24 is used to output the network configuration page.

[0154] Example 4

[0155] Embodiment 4 of the present invention also provides a computer-readable storage medium, wherein the computer-executable instructions, when executed by a computer processor, are used to implement the beam yard production progress monitoring method based on full-element collaboration as provided in Embodiment 1.

[0156] The storage medium containing computer-executable instructions provided in the embodiments of the present invention is not limited to the method operation provided in Embodiment 1, but can also execute related operations in the beam yard production progress monitoring method based on full-element collaboration provided in any embodiment of the present invention.

[0157] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for monitoring the production progress of beam yards based on full-element collaboration, characterized in that, Includes the following steps: Step S1: Collect multiple types of data through the IoT edge gateway of the beam yard and various construction sites, and generate a spatiotemporally unified dynamic full-element dataset after cleaning and alignment by the spatiotemporal coding algorithm. Step S2: Construct a multi-agent system, which includes beam yard production, offline construction and logistics transportation. Each agent parses the corresponding domain subset of the dynamic full-element dataset in parallel and extracts features, and outputs local state vectors and event logs. Step S3: Aggregate the output results of each intelligent agent through the central digital twin engine, drive the synchronous evolution and visualization of the digital twin model that integrates BIM and physical scene, and generate a virtual and real synchronized full-element digital twin scene. Step S4: Each agent performs local progress prediction based on its local lightweight model, and the central agent aggregates global information and performs cross-domain inference through a deep collaborative analysis model to generate progress-related analysis results. Step S5: The central intelligent agent generates a progress adjustment strategy based on the progress correlation analysis results. After verification and evaluation through digital twin scenario simulation, it outputs an optimized strategy set, forming a closed-loop feedback.

2. The beam yard production progress monitoring method based on full-element collaboration according to claim 1, characterized in that, The dynamic full-element dataset corresponds to a domain subset, including: dividing the dynamic full-element dataset into several domain subsets according to each agent domain, and constructing several domain subsets into an agent environment; wherein, the several domain subsets include: a precast beam production full-cycle domain subset, a bridge sub-construction domain subset, and a transportation network domain subset; Among them, the environment for constructing a subset of the precast beam production lifecycle domain as a beam yard production intelligence agent includes: constructing a production progress tensor for storing the status of each beam shape in each process, a resource status tensor for storing equipment operation status, and a quality risk matrix for storing deviation values; The environment for constructing a subset of bridge offline construction domains into an offline construction intelligent agent includes: digitally reconstructing the physical construction space based on the subset of bridge offline construction domains, and constructing an engineering progress tensor for storing the process status of each construction section, a resource occupancy tensor for storing the status of key resources occupied by each spatial unit, a quality and safety matrix for storing quality and risk, and an environment vector for storing environmental constraints based on the digitally reconstructed physical construction space. The environment for constructing a subset of the transportation network domain into a logistics transportation intelligent agent includes: digitally reconstructing the physical space of the transportation network based on the subset of the transportation network domain, and constructing a transportation progress tensor, a vehicle resource tensor, and a road network state tensor based on the digitally reconstructed physical space of the transportation network.

3. The beam yard production progress monitoring method based on full-element collaboration according to claim 2, characterized in that, The intelligent agent for beam yard production analyzes a subset of the entire precast beam production lifecycle and extracts features, including: The flow rate feature and the time deviation feature are extracted from the production progress tensor, and the progress is predicted based on the flow rate feature and the time deviation feature to obtain the progress prediction feature; Perform operational state analysis on the resource state tensor to obtain operational state transition quantities and resource utilization quantities. Use the operational state transition quantities and resource utilization quantities to perform resource competition analysis to obtain resource competition topology features. The construction quality characteristics are obtained by using the aforementioned quality risk matrix for construction quality analysis. The progress prediction features, resource competition topology features, and construction quality features are spatially quantized and encoded to generate a beam yard production state vector.

4. The beam yard production progress monitoring method based on full-element collaboration according to claim 3, characterized in that, The offline construction agent parses a subset of the bridge's offline construction domain and extracts features, including: The project progress tensor is scanned periodically along the time axis, and spatial position transition detection is performed. Based on the detected spatial position transitions, the process time is analyzed and extracted to obtain the project entity trajectory and progress characteristics. Spatial conflict detection is performed on the resource occupancy tensor, and resource competition analysis is performed based on the detected spatial conflicts to obtain the characteristics of construction interference and resource competition. Risk assessment is performed using the aforementioned quality and safety matrix to generate risk evolution characteristics; The environmental vectors are used to perform environmental assessment and generate environmental response and constraint features. The project entity trajectory and progress characteristics, construction interference and resource competition characteristics, risk evolution characteristics, and environmental response and constraint characteristics are spatially quantified and encoded to generate an offline construction state vector.

5. The beam yard production progress monitoring method based on full-element collaboration according to claim 4, characterized in that, The logistics transportation intelligent agent parses a subset of the transportation network domain and extracts features, including: Scan the transportation progress tensor and perform transportation trajectory transition detection. Extract transportation trajectory features based on the results of the transportation trajectory transition detection. Extract vehicle performance features from the vehicle resource tensor; The road network capacity is determined and bottleneck sections are identified by the road network state tensor. The dynamic characteristics of the road network are constructed based on the road network capacity and bottleneck sections. The transportation trajectory features, vehicle efficiency features, and road network dynamic features are spatially quantized and encoded to generate a logistics transportation state vector.

6. The beam yard production progress monitoring method based on full-element collaboration according to claim 5, characterized in that, Each agent performs local progress prediction based on a local lightweight model, including: The beam yard production agent determines the beam yard production actions based on the beam yard production state vector in the output local state vector through a beam yard production state-action mapping mechanism; the beam yard production actions include: plan adjustment actions, resource scheduling actions, and quality intervention actions; The offline construction agent determines offline construction actions based on the offline construction state vector in the output local state vector through an offline construction state-action mapping mechanism; the offline construction actions include: progress control and scheduling actions, dynamic resource allocation actions, quality and safety intervention actions, and environmental response and risk mitigation actions. The logistics transportation agent determines logistics transportation actions based on the logistics transportation state vector in the output local state vector through a logistics transportation state-action mapping mechanism; the logistics transportation actions include: dynamic path optimization actions and vehicle resource scheduling actions.

7. The beam yard production progress monitoring method based on full-element collaboration according to claim 1, characterized in that, The central intelligent agent aggregates global information and performs cross-domain inferences through a deep collaborative analysis model, including: The received local optimization actions of each agent are structured and parsed, and each local optimization action is parsed into action elements containing information on type, subject, target spatiotemporal window, resource requirement / release and expected impact. A cross-domain action intent network is then constructed based on the action elements. Based on the aforementioned cross-domain action intent network, global resource coupling analysis and spatiotemporal interference analysis are performed to detect cross-domain hard conflicts, soft bottlenecks, spatiotemporal interference, and supply and demand chain mismatch risks. A cross-domain coupled simulation model is established, with the cross-domain action intent network as the input to the cross-domain coupled simulation model, and a preset uncertainty disturbance is injected. Multi-scenario simulation is run to evaluate the impact of the local optimization action on the global target index and generate progress correlation analysis results.

8. A beam yard production progress monitoring system based on full-element collaboration, used to implement the beam yard production progress monitoring method based on full-element collaboration as described in any one of claims 1 to 7, characterized in that, The beam yard production progress monitoring system includes: The data acquisition module is used to collect various types of data through the IoT edge gateway of the beam yard and various construction sites, and generate a dynamic full-element dataset with spatiotemporal uniformity after cleaning and alignment by the spatiotemporal coding algorithm. The local optimization module is used to construct a multi-agent system, which includes beam yard production, offline construction and logistics transportation. Each agent parses the corresponding domain subset of the dynamic full-element dataset in parallel and extracts features, and outputs local state vectors and event logs. The twin scene module is used to aggregate the output results of each intelligent agent through the central digital twin engine, drive the synchronous evolution and visualization of the digital twin model that integrates BIM and physical scene, and generate a virtual and real synchronized full-element digital twin scene. The cross-domain extrapolation module is used by each agent to perform local progress prediction based on its local lightweight model, while the central agent aggregates global information and performs cross-domain extrapolation through a deep collaborative analysis model to generate progress-related analysis results. The closed-loop feedback module is used by the central agent to generate progress adjustment strategies based on progress-related analysis results. After verification and evaluation through digital twin scenario simulation, it outputs an optimized strategy set, forming a closed-loop feedback.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the beam yard production progress monitoring method based on full-element collaboration as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the beam yard production progress monitoring method based on full-element collaboration as described in any one of claims 1 to 7.

Images