Wharf steel structure construction progress dynamic management and control and resource allocation system

Abstract

The invention relates to the field of engineering management, and discloses a wharf steel structure construction progress dynamic management and control and resource allocation system, which comprises a real-time data perception and fusion module for collecting field data and an initial plan and generating a fusion data stream; the dynamic construction digital twinborn construction and evolution module is used for updating the digital twinborn and a dynamic empirical risk layer in real time based on the fusion data flow; the multi-agent self-adaptive decision engine is used for generating a self-adaptive construction scheme based on digital twinborn simulation deduction and feeding back a risk signal; and the man-machine cooperation and closed-loop execution module is used for visualizing the scheme for manual decision making and issuing an instruction to a site. According to the method, the physical field data sensed in real time is fused with the digital model, the dynamic construction digital twin capable of being synchronously updated with the physical construction field state is constructed, and the technical effect that digital information and the physical entity state are kept highly consistent is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of engineering management, and in particular to a system for dynamically controlling and allocating construction progress of steel structures at a wharf. Background Art

[0002] The construction of large-scale steel structures for docks is a crucial component of infrastructure development. The process involves the hoisting of massive components, complex welding techniques, and precise installation, placing extremely high demands on construction progress, resource allocation, and operational safety. Against this backdrop, achieving precise control over the entire construction process to ensure on-time, high-quality project completion while effectively mitigating potential risks has become a critical issue urgently needed within the industry.

[0003] Currently, the industry has developed a variety of technologies and methods for managing steel structure construction at docks. For example, the use of project management software enables digital management of construction plans, task allocation, and basic progress tracking. Building Information Modeling (BIM) technology is widely used for pre-construction design optimization and collision detection, improving the feasibility of solutions. Furthermore, some construction sites have deployed sensor equipment to collect equipment operating data or environmental parameters, assisting management personnel in monitoring specific information. These technologies have, to a certain extent, optimized the organization and presentation of construction information, improved the efficiency of planning, and provided data support for the construction process.

[0004] However, existing technologies still exhibit significant shortcomings in addressing the dynamic and complex nature of wharf steel structure construction. First, current construction plans are mostly static. When unexpected conditions occur at the construction site, such as logistics delays for key components or sudden severe weather, adjustments to the original plan are often delayed and rely on manual experience and judgment, resulting in the inability to efficiently and immediately reconfigure resources, which in turn affects the overall construction progress. Second, although some information collection methods have been adopted, the ability to integrate and synchronize multi-source heterogeneous data in real time is still insufficient, making it difficult to form a comprehensive, real-time mapping of the physical construction site status. This makes it difficult for managers to obtain unified, high-quality, and timely information when making decisions, which restricts the accuracy and response speed of decisions. Finally, existing risk management focuses on post-analysis or static prevention based on experience, lacking the ability to quantitatively predict and proactively avoid dynamic and coupled risks during the construction process. When multiple risk factors exist simultaneously or interact, traditional methods cannot effectively identify their potential combined impacts, making it difficult for the construction safety assurance system to achieve forward-looking defense. To this end, those skilled in the art have proposed a dynamic control and resource allocation system for wharf steel structure construction progress to address the above problems. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a dynamic management and control system for the construction progress of dock steel structures and resource allocation, which solves the common problems of static planning, information lag and lack of risk foresight in existing dock steel structure construction management methods.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A system for dynamic control and resource allocation of dock steel structure construction progress is provided, which includes: A real-time data perception and fusion module collects real-time status data of the physical construction site and the initial construction plan containing 3D geometric information and task logic relationships, and processes the collected data and the acquired plan to generate a unified fused data stream; A dynamic construction digital twin construction and evolution module, configured to construct and update in real time a dynamic construction digital twin synchronized with the physical construction site based on the fused data stream, wherein the dynamic construction digital twin includes a dynamic empirical risk layer that evolves based on the agent's experience; A multi-agent adaptive decision engine, configured to conduct simulation training and real-time deduction based on the dynamic construction digital twin to generate an adaptive construction plan including resource scheduling instructions, and to feed back risk signals for updating the dynamic empirical risk layer to the dynamic construction digital twin construction and evolution module during the training process; The human-machine collaboration and closed-loop execution module is used to visualize the adaptive construction plan for manual review and decision-making, and generate executable work instructions based on the decided plan and send them to the physical construction site.

[0007] Preferably, the specific working steps of the real-time data perception and fusion module include: Through the IoT sensor network deployed at the physical construction site, dynamic physical information of construction equipment, materials and environment is collected; Accessing a project management system to obtain the initial construction plan; Performing data cleaning and standardization preprocessing on the collected dynamic physical information and the obtained initial construction log; The pre-processed data are aligned and associated to generate the unified fusion data stream.

[0008] Preferably, the specific working steps of the dynamic construction digital twin construction and evolution module include: Initializing a static skeleton of the dynamic construction digital twin based on the initial construction plan; Continuously receiving the fused data stream to drive the state of the digital model in the dynamic construction digital twin to be updated synchronously with the physical construction site; Constructing and maintaining the dynamic empirical risk layer, wherein the dynamic empirical risk layer is a four-dimensional tensor used to quantify the potential risk of performing a construction task at a specific time and space coordinate; In response to receiving the risk signal fed back by the multi-agent adaptive decision engine, the dynamic empirical risk layer is evolved and updated.

[0009] Preferably, the evolution and update of the dynamic empirical risk layer is achieved by the following formula: ; in, For the After the updated risk layer, For the The risk layer at the time of the update, is the space-time coordinate of the event, is the default risk layer learning rate, For the agent at time Observation information, For the agent at time The action performed, Evaluates a function for an event.

[0010] The specific working mode of the event evaluation function is: When the agent performs the action under the observation information and causes a preset safety penalty or conflict penalty, the event evaluation function returns a preset positive risk value; When the agent performs the action under the observation information and does not result in a preset safety penalty or conflict penalty, the event evaluation function returns zero.

[0011] Preferably, the specific working steps of the multi-agent adaptive decision engine include: Abstract the key resources in the construction process into a group of independent agents, and define the observation space, action space and a composite reward function for each agent; Centrally training the group of independent agents in the dynamic construction digital twin environment to learn and generate an optimal collaborative decision-making strategy; In actual operation, distributed decision-making is performed based on the learned optimal collaborative decision-making strategy and the real-time status of the dynamic construction digital twin to generate the adaptive construction plan.

[0012] The composite reward function is constructed as follows: Determined by taking the weighted sum of global progress rewards, resource cost penalties, safety penalties, and coordination rewards; Among them, the global progress reward is related to the completion status of the critical path task; the resource cost penalty is related to the idle time of resources or the energy consumption of unnecessary movement; the safety penalty is related to whether the action of the intelligent agent enters a high-risk area or violates safety rules; the collaborative reward is related to the coordination efficiency between the actions of multiple intelligent agents.

[0013] Preferably, the multi-agent adaptive decision engine further includes an adaptive task generation step when generating the adaptive construction plan, which step includes: Monitoring the deviation metric between the actual state and the planned state of the dynamic construction digital twin; triggering a task generation mechanism when the deviation metric exceeds a preset threshold; The task generation mechanism explores and evaluates the expected returns of the new task list formed by dynamically decomposing or merging the original construction tasks through forward deduction; And the new task list with the maximum expected reward is selected as part of the adaptive construction plan.

[0014] The deviation metric is determined by calculating the weighted norm distance between the actual state vector and the planned state vector of the dynamic construction digital twin.

[0015] Preferably, the specific manner in which the multi-agent adaptive decision engine feeds back risk signals for updating the dynamic empirical risk layer during the training process includes: In each simulation training step, monitoring the safety penalty component and the collaborative reward component preset in the composite reward function for representing negative results; When it is determined that the value of the safety penalty component triggers a safety penalty condition, or the value of the collaboration reward component triggers a conflict penalty condition, it is determined that a negative risk event has occurred; Based on the negative risk event, the risk signal including the time and space coordinate information of the event is generated.

[0016] Preferably, the specific working steps of the human-machine collaboration and closed-loop execution module include: receiving the adaptive construction plan and converting it into a visual form including a four-dimensional construction animation and a resource scheduling path diagram; Providing an interactive interface for project managers to review, fine-tune or approve the visual plan; Parse approved plans into clear work instructions for specific construction resources; And issue the clear work instructions through mobile terminals or on-site command systems.

[0017] The present invention provides a system for dynamic control of dock steel structure construction progress and resource allocation. It has the following beneficial effects: 1. The present invention integrates real-time perceived physical site data with digital models to construct a dynamic construction digital twin that can be updated synchronously with the physical construction site status, achieving the technical effect of maintaining a high degree of consistency between digital information and the physical entity status. Compared with the existing technology that relies on static plans or preset scripts for simulation, the present invention solves the inherent defects of its plans being seriously disconnected from reality and the significant lag in information transmission.

[0018] 2. The present invention proposes a unique risk management mechanism, the core of which lies in a dynamic empirical risk layer generated by multiple intelligent agents through autonomous evolution of their own experience in massive simulation training, which realizes the quantification and forward-looking identification of potential construction risks. It effectively makes up for the shortcomings of existing methods that rely on manual experience for post-event summaries or static safety inspections, and solves the problem of being unable to predict sudden and complex risks caused by the coupling of multiple dynamic factors.

[0019] 3. The present invention provides a dynamic programming method based on a multi-agent adaptive decision-making engine. This method has the ability to adaptively generate tasks. When major construction deviations are detected, it can actively decompose and reorganize the original tasks to generate new and better solutions. This technical solution breaks through the framework of the existing technology in which the optimization algorithm can only sort the fixed task list, and solves the bottleneck of its rigid strategy when facing major emergencies and must rely on human intervention for creative problem solving. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of the system function modules of the present invention; Figure 2 This is a schematic diagram of the workflow of the real-time data perception and fusion module of the present invention; Figure 3 The data structure and interaction diagram of the dynamic construction digital twin construction and evolution module of the present invention; Figure 4 This is a schematic diagram of the workflow of the multi-agent adaptive decision-making engine of the present invention; Figure 5 This is a schematic diagram of the workflow of the human-machine collaboration and closed-loop execution module of the present invention.

[0021] Among them, 10. Real-time data perception and fusion module; 20. Dynamic construction digital twin construction and evolution module; 30. Multi-agent adaptive decision engine; 40. Human-machine collaboration and closed-loop execution module. DETAILED DESCRIPTION

[0022] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0023] Refer to the attached Figure 1 , Figure 1 Figure 1 is a schematic diagram of the functional modules of a system for dynamic management and control of dock steel structure construction progress and resource allocation, according to one embodiment of the present invention. The system provided by the present invention may include: a real-time data perception and fusion module 10, a dynamic construction digital twin construction and evolution module 20, a multi-agent adaptive decision engine 30, and a human-machine collaboration and closed-loop execution module 40.

[0024] The real-time data perception and fusion module 10 collects real-time status data from the physical construction site and the initial construction plan from the project management system. It preprocesses and standardizes the collected and acquired heterogeneous data, ultimately generating a unified, structured, fused data stream that is transmitted to the dynamic construction digital twin construction and evolution module 20.

[0025] The dynamic construction digital twin construction and evolution module 20 has the function of receiving the fused data stream from the real-time data perception and fusion module 10, and constructing and maintaining a dynamic construction digital twin based on the data stream. The dynamic construction digital twin is a high-fidelity digital replica of the physical construction site, and the state of its internal model is updated in real time as the state of the physical site changes. Another function of the dynamic construction digital twin construction and evolution module 20 is to construct and maintain a dynamic empirical risk layer, which is used to quantify the potential risks in the construction process, and its value will evolve and update based on the risk signal from the multi-agent adaptive decision engine 30. The dynamic construction digital twin construction and evolution module 20 will provide the dynamic construction digital twin carrying the complete site status to the multi-agent adaptive decision engine 30 as the environmental basis for its calculation and deduction.

[0026] The multi-agent adaptive decision-making engine 30 performs large-scale simulation training and real-time forward deduction based on the dynamic construction digital twin provided by the dynamic construction digital twin construction and evolution module 20. Through these operations, the engine generates an adaptive construction plan that includes specific resource scheduling instructions. During the simulation training operation, when a preset negative event is determined to have occurred, the engine generates a risk signal and feeds this signal back to the dynamic construction digital twin construction and evolution module 20. The adaptive construction plan ultimately generated by the engine is transmitted to the human-machine collaboration and closed-loop execution module 40.

[0027] The human-machine collaboration and closed-loop execution module 40 receives the adaptive construction plan from the multi-agent adaptive decision engine 30. It converts the plan data into a visual representation for project management to review and make final decisions. Once the plan is approved, the human-machine collaboration and closed-loop execution module 40 interprets it into clear, executable work instructions and distributes these instructions to the corresponding execution units at the physical construction site.

[0028] The system provided by the present invention has an overall workflow that constitutes a closed-loop control process. The process begins with the continuous collection of the physical construction site status by the real-time data perception and fusion module 10. The generated fusion data stream drives the dynamic construction digital twin construction and evolution module 20 to synchronize the status. The multi-agent adaptive decision engine 30 makes decisions based on the updated dynamic construction digital twin and generates an adaptive construction plan. The plan is manually reviewed and instructions are issued by the human-machine collaboration and closed-loop execution module 40 to guide the operations at the physical construction site. The state changes generated by the physical construction site after executing the new instructions will be captured again by the real-time data perception and fusion module 10, thereby starting a new round of perception, modeling, decision-making and execution cycles, and realizing rolling optimization and control of the entire construction process.

[0029] The following, in conjunction with the accompanying drawings, will describe in detail the internal structure and specific implementation methods of the real-time data perception and fusion module 10, the dynamic construction digital twin construction and evolution module 20, the multi-agent adaptive decision engine 30, and the human-machine collaboration and closed-loop execution module 40 in an embodiment of the present invention.

[0030] Refer to the attached Figure 2 , Figure 2 1 is a schematic diagram of the workflow of the real-time data perception and fusion module 10 according to one embodiment of the present invention. The specific implementation of the real-time data perception and fusion module 10 is as follows: First, the real-time data perception and fusion module 10 collects multi-source heterogeneous data. This step is carried out from both physical and digital dimensions. In the physical dimension, dynamic physical information is collected by deploying an IoT sensor network on key entities at the construction site.

[0031] Specifically, large-scale construction equipment (such as crawler cranes and transport vehicles) is equipped with Global Positioning System (GPS) modules and Inertial Measurement Units (IMUs) to obtain their three-dimensional coordinates and attitude data (pitch, roll, and yaw), respectively. Prefabricated steel structures awaiting installation are affixed with Radio Frequency Identification (RFID) tags or spray-painted with QR codes. Readers and writers deployed at key nodes along the path (such as factory gates, construction site entrances, and temporary storage yards) collect information on their logistics status. Meteorological sensors are installed in key operating areas (such as high-altitude lifting areas) to collect environmental parameters such as wind speed, direction, temperature, and humidity.

[0032] In the digital dimension, the real-time data perception and fusion module 10 accesses the project management system through a standard data interface to obtain the building information model (BIM) file containing the three-dimensional geometric information, attributes and spatial relationships of all components, as well as the initial construction plan file that defines the task list, logical dependencies and estimated construction period.

[0033] Next, the real-time data perception and fusion module 10 performs data preprocessing on the collected multi-source heterogeneous data. This step aims to eliminate data noise and standardize the data format. For continuous signal data collected from sensors (such as GPS coordinates), a Kalman filter algorithm is used to smooth the data and eliminate abnormal transition points. Data from all sources is standardized.

[0034] This conversion includes: converting all spatial location data (derived from GPS and BIM models) into a preset local coordinate system for the construction site; synchronizing the time of all devices through the Network Time Protocol (NTP) and adding a timestamp in a unified format to each data record; and converting all physical quantity measurements (such as wind speed and temperature) into International Standard Units (SI).

[0035] Finally, the real-time data perception and fusion module 10 performs data fusion and distribution. This step aligns the pre-processed multiple independent data streams according to their unified timestamps and integrates them into a structured fused data stream. , the fused data stream Can be represented as a data set: ; in, The state information of each device in the set includes its three-dimensional coordinates, posture parameters, and a status flag that represents its working state (such as idle, moving, or working); The status of all key materials at that moment. The status information of each material in the set includes its current location and a status flag that represents its logistics status (such as in transit, arrived on site, installed); It is a set of environmental parameters at that moment, including wind speed, temperature and other values.

[0036] The real-time data perception and fusion module 10 generates the structured fusion data stream It is transmitted to the dynamic construction digital twin construction and evolution module 20 in real time and continuously through a message queue transmission protocol (such as MQTT) as the data source for updating its internal state.

[0037] Refer to the attached Figure 3 , Figure 3 : is a data structure and interaction diagram of a dynamic construction digital twin building and evolution module 20 according to one embodiment of the present invention. The specific implementation of the dynamic construction digital twin building and evolution module 20 is as follows: The Dynamic Construction Digital Twin Construction and Evolution Module 20 first initializes the Dynamic Construction Digital Twin (CDT). Based on the initial construction plan and BIM model files obtained from the Real-Time Data Perception and Fusion Module 10, this step constructs a static geometric model in three-dimensional space, encompassing all components, equipment, and site. It then loads the task list, resource library, and logical constraints between tasks to form the static skeleton of the CDT.

[0038] After initialization, the dynamic construction digital twin building and evolution module 20 enters the continuous state synchronization update phase. It continuously receives the fused data stream from the real-time data perception and fusion module 10. The dynamic construction digital twin construction and evolution module 20 traverses each piece of status information in the data stream, retrieves the corresponding digital model in the CDT using the entity's unique identifier (e.g., equipment number, component ID), and updates its state parameters (e.g., 3D coordinates, posture, and logistics status flags) to the latest values ​​in the data stream. This ensures that the state of the CDT remains synchronized with the state of the physical construction site in the time dimension.

[0039] One of the core functions of the dynamic construction digital twin construction and evolution module 20 is to build and maintain a dynamic empirical risk layer. This layer is constructed as a four-dimensional tensor data structure, whose three dimensions correspond to the local coordinate system of the construction site. Spatial coordinates, the fourth dimension corresponds to time . This tensor is at any spacetime coordinate point The value of is used to quantify the potential risk of performing construction tasks near this point. When the system is initialized, all values ​​of this risk layer are set to zero.

[0040] This dynamic empirical risk layer is not static; instead, it evolves and updates based on risk signals from the multi-agent adaptive decision engine 30. During system operation, when the multi-agent adaptive decision engine 30 determines a negative risk event has occurred during its internal simulation training, it generates and sends a risk signal to this module. Upon receiving this signal, this module triggers an update to the risk layer. This evolutionary update is implemented using the following formula: ; in, For the After the updated risk layer, For the The risk layer at the time of the update, is the space-time coordinate of the event, is the default risk layer learning rate, For the agent at time Observation information, For the agent at time The action performed, Evaluates a function for an event.

[0041] The specific working method is: when the agent is observing information Next action When the result triggers a preset safety penalty (such as entering a restricted area) or a conflict penalty (such as physical interference with other resources), the function returns a preset positive risk value; in all other normal cases where no penalty is triggered, the function returns zero.

[0042] Ultimately, the dynamic construction digital twin construction and evolution module 20 will provide the complete CDT status data of the updated equipment, materials, environmental status and evolved dynamic experience risk layer to the multi-agent adaptive decision engine 30 as the environmental basis for its subsequent simulation training and real-time decision-making.

[0043] Refer to the attached Figure 4 , Figure 4 FIG. 1 is a schematic diagram of the workflow of a multi-agent adaptive decision engine 30 according to an embodiment of the present invention. The specific implementation of the multi-agent adaptive decision engine 30 is as follows: The engine first performs multi-agent modeling. This step abstracts the key dispatchable resources in the dock steel structure construction process, such as crawler cranes, flatbed trucks, and welding teams, into a group of independent agents. Let the agent set be For each agent , defining its three elements: Observation Space: Agent In time Obtained local observation information , which includes the state of the agent itself (three-dimensional coordinates, posture, load), its local environment information (the state of other agents within a certain range, information on components to be processed), the risk value of its location and planned path on the dynamic experience risk layer, and information on its associated tasks in the initial construction plan.

[0044] Action Space: Agent A set of discrete actions that can be performed For example, for a crane modeled as an intelligent agent, its action space may include: standby, move to the target point, grab the component, rotate to the target angle, and release the component.

[0045] Composite reward function: used to evaluate the agent during training In time Execute an action Rewards received This function is composed of a weighted sum of multiple components to guide the agents to learn the desired collaborative behavior.

[0046] In a specific embodiment, the compound reward function can be expressed as: ; in, , , , It is a preset weight coefficient used to balance different optimization objectives; It is a global progress reward. When any agent completes a task on the critical path, all agents receive a positive reward. For the intelligent agent The associated cost penalty is negative and proportional to the idle time and energy consumption of unnecessary movement; For safety penalties, when the agent If the action causes the user to enter a high-risk area defined by the dynamic experience risk layer, or violates the preset safety rules, the user will receive a negative reward. For collaborative rewards, when the agent When its behavior forms an efficient cooperation with other agents, it receives a positive reward; when its behavior conflicts or interferes with other agents, it receives a negative reward.

[0047] The engine then conducts offline centralized policy training on the group of agents in a dynamic construction digital twin environment. The goal of the training is to find a set of optimal policy network parameters. , so that the cumulative expected return of all agents Maximize. The objective function is expressed as: ; in, Represented by parameters The joint strategy of all agents determined by is the expectation operator, is the maximum time step of a simulation training, is a discount factor between 0 and 1, used to balance short-term and long-term rewards. The compound reward function defined above is For intelligent agents The calculated reward value.

[0048] During the training process, when the or When the negative value part of is triggered, that is, when a negative risk event occurs, this engine generates a risk signal containing the time and space coordinate information of the event, and feeds it back to the dynamic construction digital twin construction and evolution module 20.

[0049] In actual operation, the engine uses the trained strategy to perform online real-time decision making. The engine continuously monitors the deviation measurement between the actual state and the planned state of the dynamic construction digital twin. When this deviation metric exceeds a preset threshold, indicating a significant construction deviation, an adaptive task generation mechanism is triggered. This mechanism uses high-speed forward deduction within the dynamic construction digital twin to evaluate the expected returns of adhering to the original plan versus executing a set of alternative new task lists. The new task list that maximizes the expected returns is then selected as part of an adaptive construction plan. This adaptive construction plan contains specific scheduling instructions for all relevant resources for a period of time in the future.

[0050] Finally, the multi-agent adaptive decision engine 30 transmits the generated adaptive construction plan including resource scheduling instructions to the human-machine collaboration and closed-loop execution module 40.

[0051] Refer to the attached Figure 5 , Figure 51 is a schematic diagram of the workflow of the human-machine collaboration and closed-loop execution module 40 according to one embodiment of the present invention. The specific implementation of the human-machine collaboration and closed-loop execution module 40 is as follows: The human-machine collaboration and closed-loop execution module 40 first receives the adaptive construction plan transmitted from the multi-agent adaptive decision engine 30. The plan is a data structure that contains an updated task list, resource allocations for a period of time in the future, and a specific action sequence. After receiving the plan, this module performs visualization processing of the plan and converts it into one or more visualization forms that are easy for humans to understand. For example, based on the action sequence and time information in the plan, a four-dimensional construction animation containing a time dimension is generated to dynamically display the subsequent construction process; on the three-dimensional model of the dynamic construction digital twin, the scheduling paths and work points of each resource in the future period of time are superimposed and displayed in the form of highlighted paths or icons; at the same time, an updated construction Gantt chart is generated to clearly display the start and end times and logical relationships of the tasks.

[0052] The human-machine collaboration and closed-loop execution module 40 then provides a human-machine interaction interface. This interface can be deployed on a display device in the project command center or on a mobile terminal held by a project manager. Through this interactive interface, project managers can browse the aforementioned four-dimensional construction animation, scheduling path diagram, and Gantt chart, and conduct a comprehensive review of the plan generated by the multi-agent adaptive decision engine 30. This interface provides interactive functions, allowing managers to make final decisions on the plan, including direct approval or fine-tuning before approval. Fine-tuning operations may include: making small-scale drag-and-drop adjustments to the movement path of a single resource, or changing the execution order of non-critical tasks.

[0053] After the adaptive construction plan is approved (either directly or after fine-tuning), the human-machine collaboration and closed-loop execution module 40 performs the instruction parsing and generation step. This step parses the abstract action instructions for the agent in the plan (e.g., "move to target," "grab component") into clear and executable work instructions for specific construction resources and operators. For example, a "move to target" instruction is parsed into a navigation instruction containing the specific target's 3D coordinates and a recommended path; a "grab component" instruction is parsed into a lifting task instruction containing the component's unique identifier, lifting weight, and target installation location.

[0054] Finally, the human-machine collaboration and closed-loop execution module 40 issues clear work instructions via a pre-set communication link. These instructions are sent via a wireless network to mobile terminal applications used by on-site operators, or directly to the control units of semi-automated construction equipment capable of receiving and executing instructions. The results of executing these instructions at the physical construction site, as well as the resulting changes in the physical site status, are then collected by the real-time data perception and fusion module 10, forming a complete, continuously operating closed-loop control process for the system.

[0055] In order to further illustrate the collaborative working process of the technical solution of the present invention, a specific working scenario example will be used below to illustrate.

[0056] During a dock steel structure construction project, a key prefabricated component, A, originally scheduled for hoisting that afternoon, would arrive at the construction site delayed due to a failure in an upstream transport link. Simultaneously, real-time data from meteorological sensors installed in the work area indicated that afternoon wind speeds were expected to exceed the safe threshold for hoisting operations. The system provided by this invention will sense these emergencies, make decisions, and develop a response plan.

[0057] First, the real-time data perception and fusion module 10 continuously collects dynamic information about the construction site. In this scenario, its sensor network captures the following information: The RFID reader does not detect the entry signal of component A at the scheduled arrival time of the component and receives a delay notification sent by the logistics system.

[0058] Meteorological sensors measure and predict in real time that during the planned lifting period of component A, the wind speed in the operation area will continue to exceed the safety upper limit of 12 meters per second.

[0059] The real-time data perception and fusion module 10 preprocesses, standardizes, and fuses this information, generating a fused data stream that includes the logistics status anomaly (delay) of component A, the environmental risk (high winds) in the lifting area, and the current status of all on-site resources. This fused data stream is transmitted to the dynamic construction digital twin construction and evolution module 20.

[0060] Secondly, the dynamic construction digital twin construction and evolution module 20 receives the fused data stream from the real-time data perception and fusion module 10. The dynamic construction digital twin construction and evolution module 20 updates the internal state of the dynamic construction digital twin based on the data stream.

[0061] Specifically, the status of component A in the digital twin is updated to "Transportation Delayed," while its position in the on-site model remains "Not Entered." Meanwhile, due to the excessive winds predicted by the meteorological forecast, the multi-agent adaptive decision engine 30 did not generate a safety penalty signal. However, it will assess potential risks during the simulation phase, combining current environmental data.

[0062] The multi-agent adaptive decision-making engine 30 then continuously monitors the deviation between the real-time status of the dynamic construction digital twin and the initial construction plan. When the system identifies a significant deviation between the current actual progress and the plan due to delays in component A and predicted wind speed exceeding the standard, the multi-agent adaptive decision-making engine 30's adaptive task generation mechanism is triggered.

[0063] The multi-agent adaptive decision engine 30 then performs high-speed forward deduction in the dynamic construction digital twin.

[0064] First, it evaluated the expected return of carrying out the operation according to the original plan (i.e., lifting component A that afternoon). The simulation results showed that this option would result in high risk and low efficiency due to the strong winds.

[0065] The engine then explores and evaluates a new set of candidate tasks. For example, it might deduce the following alternatives: The task of hoisting component A originally scheduled for the afternoon will be postponed to the next day when the wind speed is expected to be safe.

[0066] In the afternoon of the same day, idle lifting resources and installation teams were re-dispatched to carry out preparatory tasks that were originally planned to have a lower priority but were not affected by wind power, such as transporting steel to other areas, equipment maintenance, or ground welding operations.

[0067] Plan a temporary storage area and transfer task for component A after its arrival to avoid it occupying space on the critical path.

[0068] The multi-agent adaptive decision engine 30 calculates the expected returns for all candidate task lists and ultimately selects the dynamic task list that maximizes the expected return as the optimal adaptive construction plan. This plan will include a new task sequence, resource allocation, and specific execution instructions. For example, it might include the instruction to "deploy crane No. 2 and welder team No. 3 to area B for ground pre-welding work." This adaptive construction plan is transmitted to the human-machine collaboration and closed-loop execution module 40.

[0069] Finally, the human-machine collaboration and closed-loop execution module 40 receives the adaptive construction plan from the multi-agent adaptive decision engine 30. This module converts the plan into a visual form, for example, dynamically demonstrating the adjusted construction progress and resource scheduling path in a four-dimensional construction animation and updating the Gantt chart to display the new task start and end times.

[0070] The project manager reviews these visualization plans through the interactive interface and approves them after confirming their rationality.

[0071] Once approved, the module parses the plan into specific, executable work instructions. For example, a command might be issued to the mobile terminal of crane operator No. 2: "Immediately proceed to Area B to transfer pre-welded components"; or to welding team No. 3: "Perform pre-welding work on ground components in Area B." These instructions are then distributed to the corresponding execution units at the physical construction site.

[0072] The physical construction site executes the work according to the new instructions. Real-time data generated during the work, such as equipment position, component status, and environmental parameters, is again collected by the real-time data perception and fusion module 10, forming a continuous, rolling closed loop to ensure the system's continuous and dynamic control of the construction site.

[0073] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A dynamic control and resource allocation system for the construction progress of a wharf steel structure, characterized in that: include: A real-time data perception and fusion module, which collects real-time status data of the physical construction site and the initial construction plan, and generates a unified fused data stream; A dynamic construction digital twin construction and evolution module, configured to construct and update in real time a dynamic construction digital twin synchronized with the physical construction site based on the fused data stream, wherein the dynamic construction digital twin includes a dynamic empirical risk layer that evolves based on the agent's experience; A multi-agent adaptive decision-making engine for performing simulation training and real-time deduction based on the dynamic construction digital twin, generating an adaptive construction plan including resource scheduling instructions, and providing feedback during the training process on risk signals used to update the dynamic empirical risk layer; The human-machine collaboration and closed-loop execution module is used to visualize the adaptive construction plan for manual review and decision-making, and generate executable work instructions based on the decided plan and send them to the physical construction site.

2. A dynamic control and resource allocation system for the construction progress of a wharf steel structure according to claim 1, characterized in that: The specific working steps of the real-time data perception and fusion module include: Collecting dynamic physical information of construction equipment, materials, and environment at the physical construction site through an Internet of Things sensor network; Access the project management system to obtain the initial construction plan including 3D geometry information and task logic relationships; Performing data cleaning and standardization preprocessing on the collected dynamic physical information and the obtained initial construction plan; The pre-processed data are aligned and associated to generate the unified fusion data stream.

3. A system for dynamic control and resource allocation of dock steel structure construction progress according to claim 1, characterized in that: The specific working steps of the dynamic construction digital twin construction and evolution module include: Initializing a static skeleton of the dynamic construction digital twin based on the initial construction plan; Continuously receiving the fused data stream to drive the state of the digital model in the dynamic construction digital twin to be updated synchronously with the physical construction site; Constructing and maintaining the dynamic empirical risk layer, wherein the dynamic empirical risk layer is a four-dimensional tensor used to quantify the potential risk of performing a construction task at a specific time and space coordinate; In response to receiving the risk signal fed back by the multi-agent adaptive decision engine, the dynamic empirical risk layer is evolutionarily updated.

4. A system for dynamic control and resource allocation of dock steel structure construction progress according to claim 3, characterized in that: The evolution and update of the dynamic empirical risk layer is achieved through the following formula: ； in, For the After the updated risk layer, For the The risk layer at the time of the update, is the space-time coordinate of the event, is the default risk layer learning rate, For the agent at time Observation information, For the agent at time The action performed, Evaluates a function for an event.

5. A system for dynamic control and resource allocation of dock steel structure construction progress according to claim 4, characterized in that: The event evaluation function The specific working method is: When the agent observes the information Execute the above action When a preset safety penalty or conflict penalty is caused, the event evaluation function returns a preset positive risk value; When the agent observes the information Execute the above action When no preset safety penalty or conflict penalty is incurred, the event evaluation function returns zero.

6. A system for dynamic control and resource allocation of dock steel structure construction progress according to claim 1, characterized in that: The specific working steps of the multi-agent adaptive decision engine include: Abstract the key resources in the construction process into a group of independent agents, and define the observation space, action space and compound reward function for each agent; In the dynamic construction digital twin environment, the group of independent intelligent agents are centrally trained to learn and generate an optimal collaborative decision-making strategy; In actual operation, based on the learned optimal collaborative decision-making strategy and the real-time status of the dynamic construction digital twin, distributed decision-making is performed to generate the adaptive construction plan.

7. A system for dynamic control and resource allocation of dock steel structure construction progress according to claim 6, characterized in that: The composite reward function is constructed as follows: Determined by taking the weighted sum of global progress rewards, resource cost penalties, safety penalties, and coordination rewards; Among them, the global progress reward is related to the completion status of the critical path task; the resource cost penalty is related to the idle time of resources or the energy consumption of unnecessary movement; the safety penalty is related to whether the action of the intelligent agent enters a high-risk area or violates safety rules; the collaborative reward is related to the coordination efficiency between the actions of multiple intelligent agents.

8. A system for dynamic control and resource allocation of dock steel structure construction progress according to claim 6, characterized in that: The multi-agent adaptive decision engine further includes an adaptive task generation step when generating the adaptive construction plan, which step includes: monitoring a deviation metric between an actual state and a planned state of the dynamic construction digital twin; When the deviation metric exceeds a preset threshold, a task generation mechanism is triggered; The task generation mechanism explores and evaluates the expected returns of the new task list formed by dynamically decomposing or merging the original construction tasks through forward deduction; The new task list with the greatest expected reward is selected as part of the adaptive construction solution.

9. A system for dynamic control and resource allocation of dock steel structure construction progress according to claim 6, characterized in that: The specific manner in which the multi-agent adaptive decision engine feeds back risk signals for updating the dynamic empirical risk layer during the training process includes: In each simulation training step, monitoring the safety penalty component and the collaborative reward component preset in the composite reward function for representing negative results; When it is determined that the value of the safety penalty component triggers a safety penalty condition, or the value of the collaboration reward component triggers a conflict penalty condition, it is determined that a negative risk event has occurred; Based on the negative risk event, the risk signal including the spatiotemporal coordinate information of the event is generated.

10. A system for dynamic control and resource allocation of dock steel structure construction progress according to claim 1, characterized in that: The specific working steps of the human-machine collaboration and closed-loop execution module include: receiving the adaptive construction plan and converting it into a visual form including a four-dimensional construction animation and a resource scheduling path diagram; Providing an interactive interface for project managers to review, fine-tune or approve the visual plan; Parse approved plans into clear work instructions for specific construction resources; The clear work instructions are issued through a mobile terminal or an on-site command system.

Images