Construction process digital twinning online deduction method and device

Abstract

The invention relates to the technical field of digital twinning, in particular to a construction process digital twinning online deduction method and device.The method comprises the steps that a structured knowledge base covering safety risks, potential quality hazards and a process dynamic interaction mechanism is constructed, and a core knowledge basis is provided for generation of scheduling responses; aiming at the uniqueness of construction process scheduling, a reinforcement learning network special for dynamic scheduling of the construction process is designed, and intelligent and near-real-time construction scheduling decision is realized; aiming at the problem that a reinforcement learning model is difficult to effectively learn and respond to sudden construction disturbance, an adaptive model training method is researched, so that the model can learn how to dynamically adjust a process network when disturbance occurs, and the influence caused by process adjustment is reduced as much as possible while an engineering construction target is ensured. Therefore, the problems that most of dynamic change elements of the construction site considered in the prior art are resource change, weather change and the like, and response to events such as safety problems or potential quality hazards of the construction site is not achieved are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of digital twinning, in particular to a construction process digital twinning online deduction method and device. BACKGROUND

[0002] Traditional construction scheduling methods usually construct optimization models based on static constraint conditions and generate initial scheduling schemes through offline calculation. Although these methods can optimize resource allocation and schedule management to a certain extent, they often need to recalculate the scheduling scheme when encountering unexpected problems, causing decision delay and affecting the safety and efficiency of the construction process. In addition, scheduling methods based on heuristic algorithms and mathematical programming still face problems such as premature convergence and local optimal solution in optimization problems, making these methods limited in the face of complex construction scenarios. Therefore, there is an urgent need for a new scheduling method that can adapt to the dynamically changing construction environment during the construction process.

[0003] Digital twinning technology has received widespread attention in recent years and is considered an effective means to address the challenges of dynamic scheduling on construction sites. Digital twinning realizes the virtual-real mapping of physical systems through high-fidelity virtual models, can synchronize the state of physical systems in real time, and reflects the operating conditions of related physical systems, which enables the construction site to monitor and optimize the construction process comprehensively by obtaining real-time data. Digital twinning technology greatly improves the real-time state data acquisition capability of human, machine, material, and environment elements on the construction site, and can more accurately reflect the dynamic changes on the site, providing strong support for construction scheduling. Although digital twinning can provide high-precision virtual simulation capabilities, it still faces a series of challenges such as large data size and complex elements in actual application.

[0004] Reinforcement learning (RL) provides an effective decision-making framework for solving the challenges of dynamic scheduling on construction sites. As an adaptive dynamic decision-making method, reinforcement learning continuously adjusts its decision-making strategy through continuous interaction with the environment, and can adapt to rapidly changing scenarios. However, traditional RL relies on manually designed state features and is difficult to handle high-dimensional action spaces (such as process parallel adjustment and resource multi-objective allocation), while the spatio-temporal coupling between processes in the construction scheduling scenario can lead to an explosion of action dimensions. Deep reinforcement learning (DRL) technology automatically extracts high-dimensional state features through deep neural networks and uses Q-learning, policy gradient, and other algorithms to realize the dimension reduction mapping of high-dimensional action spaces. Its adaptability advantage has been verified in dynamic decision-making problems. Deep reinforcement learning can continuously interact with the environment and optimize its strategy through trial and error learning, which is very effective for dynamic scheduling. However, the application of deep reinforcement learning in construction is still in the development stage.

[0005] Currently, existing technologies can be used to implement adaptive scheduling methods for construction cash flow based on deep reinforcement learning. The main goal is to optimize the balance between construction period, resources and cost, and to introduce a cash flow perspective to assist in the adaptive optimization of scheduling strategies. Alternatively, a hybrid method of reinforcement learning and agent-based modeling (ABM) can be used, combined with graph embedding networks to optimize the scheduling of process sequences and resource constraints in construction projects, and to simulate the dynamic evolution process in complex construction environments. In addition, existing technologies can also combine the valid action sampling mechanism (VAS) with the DRL scheduling method of graph convolutional networks, accelerating training convergence through reward shaping, and demonstrating good scheduling and rescheduling capabilities in large-scale construction projects.

[0006] Therefore, the existing construction process scheduling technology has the following shortcomings: 1. Incomplete analysis of the impact of multi-factor coupling: The existing technology does not provide a detailed discussion on how factors such as safety, quality, and process influence and restrict each other during the construction process.

[0007] 2. The potential of digital twin technology in process scheduling has not yet been realized: The existing scheduling process is not integrated with the actual construction process. For example, when an emergency occurs on site, it is impossible to respond quickly.

[0008] 3. The dynamic scheduling method for process is not yet mature: There is no method for dynamically scheduling work processes to address quality risks and safety issues during the construction process.

[0009] In summary, the dynamic change factors of the construction site considered by existing technologies are mostly resource changes and weather changes, and they do not respond to events such as safety issues or quality hazards at the construction site, which urgently needs to be solved. Summary of the Invention

[0010] The present application provides a method and device for online simulation of digital twins of the construction process to solve the problem that the existing technology considers that the dynamic changing factors of the construction site are mostly resource changes and weather changes, and does not respond to events such as safety issues or quality hazards at the construction site.

[0011] The first aspect embodiment of the application provides a construction process digital twin online deduction method, applied to an offline training stage, including the following steps: based on an original construction progress plan, a preset safety-process coupling knowledge sub-library and a quality-process coupling knowledge sub-library, a safety-quality-process coupling evolution knowledge base is constructed, and a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base is established; the coupling relationship knowledge graph is modeled to determine a corresponding state space, an action space, a state transition mechanism and a disturbance mechanism, and a multi-objective reward function corresponding to the original construction progress plan is constructed, so as to construct a process scheduling optimization model based on the state space, the action space, the state transition mechanism, the multi-objective reward function and the disturbance mechanism; through the process scheduling optimization model, a hierarchical deep reinforcement learning network model is determined, and a construction simulation environment is constructed by using the hierarchical deep reinforcement learning network model, and based on the construction simulation environment, the hierarchical deep reinforcement learning network model is trained, so as to perform digital twin online deduction on the original construction progress plan based on the trained hierarchical deep reinforcement learning network model in an online deduction stage.

[0012] Optionally, in one embodiment of the present application, the safety-quality-process coupling evolution knowledge base is constructed based on the original construction schedule, the preset safety-process coupling knowledge subbase and the quality-process coupling knowledge subbase, and a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base is established, including: obtaining multiple safety risks and multiple safety accidents in the target construction process, and determining the risk level and risk category of each safety risk among the multiple safety risks, and determining the corresponding process scheduling impact type based on the risk level and the risk category; based on the multiple safety accidents, establishing an accident type mapping relationship between the risk category and the safety accident type, a level mapping relationship between each safety risk and the risk level, a risk category mapping relationship between each safety risk and the risk category, and a process mapping relationship between each safety risk and the process scheduling impact type, so as to determine the accident type mapping relationship, the level mapping relationship, and the risk category mapping relationship. The safety-process coupling knowledge sub-base is constructed based on the mapping relationship between the system and the process; multiple quality hazards in the target construction process are obtained, and the severity of each quality hazard is quantified to obtain the corresponding severity level, and based on the severity level, multiple process interaction patterns are established, and the status indicators corresponding to each quality hazard are determined; mapping relationships between the status indicators and the quality hazards, the severity levels and process interaction patterns are respectively established to construct the quality-process coupling knowledge sub-base according to the mapping relationships; based on the safety-process coupling knowledge sub-base and the quality-process coupling knowledge sub-base, the safety-quality-process coupling evolution knowledge base is constructed, and through a preset graph database strategy, the safety-process knowledge graph and the quality-process knowledge graph corresponding to the safety-process coupling knowledge sub-base and the quality-process coupling knowledge sub-base are respectively established to construct a corresponding visual coupling relationship knowledge graph according to the safety-process knowledge graph and the quality-process knowledge graph.

[0013] Optionally, in an embodiment of the present application, the coupling relationship knowledge graph is modeled, a corresponding state space, an action space, a state transition mechanism and a disturbance mechanism are determined, and a multi-objective reward function corresponding to the original construction progress plan is constructed, so as to construct a process scheduling optimization model based on the state space, the action space, the state transition mechanism, the multi-objective reward function and the disturbance mechanism, including: calculating a global environment state and a process feature state in a construction process corresponding to the original construction progress plan, and determining the state space based on a preset double-layer state representation strategy in combination with the global environment state and the process feature state; constructing the action space based on a preset binary action representation strategy, and determining whether all predecessor processes of a current process are completed according to the construction progress of the original construction progress plan, wherein, in the case that the all predecessor processes are completed, the state of the current process is updated to determine the state transition mechanism; obtaining disturbance data in the construction process corresponding to the original construction progress plan, and generating corresponding processing measures according to the disturbance data and a preset response strategy to determine the disturbance mechanism.

[0014] Optionally, in an embodiment of the present application, the layered deep reinforcement learning network model is determined through the process scheduling optimization model, and a construction simulation environment is constructed using the layered deep reinforcement learning network model, and the layered deep reinforcement learning network model is trained based on the construction simulation environment, including: constructing the layered deep reinforcement learning network model based on a pre-constructed input layer, an encoding layer, a process attention layer, a feature fusion layer, a double-flow architecture layer and an output layer; extracting component geometric information and spatial relationships from a preset building information model, and obtaining construction activity data corresponding to the component geometric information and the spatial relationships based on a preset four-dimensional work breakdown structure; determining a plurality of process constraint rules based on the construction activity data, so as to construct a disturbance simulation generation algorithm according to the plurality of process constraint rules and the coupling relationship knowledge graph, and construct the construction simulation environment through the disturbance simulation generation algorithm, so as to train the layered deep reinforcement learning network model through a preset conventional construction path and a disturbance processing path in the construction simulation environment.

[0015] Optionally, in an embodiment of the present application, the mathematical expression of the multi-objective reward function is:

[0016] wherein, represents a state improvement reward; represents a parallel process reward; represents an idle time penalty; represents a resource overrun penalty; represents a time delay penalty.

[0017] The second aspect embodiment of the present application provides a construction process digital twin online deduction method, applied to an online deduction stage, which comprises the following steps: monitoring a current construction site state corresponding to a target construction site, and obtaining corresponding disturbance information according to the current construction site state; inputting the disturbance information into a pre-constructed coupling relationship knowledge graph to output a corresponding process scheduling scheme and process adjustment suggestion; performing format conversion on the process scheduling scheme and the process adjustment suggestion to obtain corresponding target format data, and inputting the target format data into a pre-trained hierarchical deep reinforcement learning network model to output a final process scheduling scheme.

[0018] The third aspect embodiment of the present application provides a construction process digital twin online deduction method, applied to an offline training stage, which comprises the following steps: a graph construction module, configured to construct a safety-quality-process coupling evolution knowledge base based on an original construction progress scheme, a preset safety-process coupling knowledge sub-library and a quality-process coupling knowledge sub-library, and establish a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base; a modeling module, configured to model the coupling relationship knowledge graph, determine a corresponding state space, action space, state transition mechanism and disturbance mechanism, and construct a multi-objective reward function corresponding to the original construction progress scheme, so as to construct a process scheduling optimization model based on the state space, the action space, the state transition mechanism, the multi-objective reward function and the disturbance mechanism; a training module, configured to determine a hierarchical deep reinforcement learning network model through the process scheduling optimization model, and construct a construction simulation environment using the hierarchical deep reinforcement learning network model, and train the hierarchical deep reinforcement learning network model based on the construction simulation environment, so as to perform digital twin online deduction on the original construction progress scheme based on the trained hierarchical deep reinforcement learning network model in the online deduction stage.

[0019] Optionally, in an embodiment of the present application, the graph construction module comprises: an acquisition unit configured to acquire a plurality of safety risks and a plurality of safety accidents in a target construction process, determine a risk level and a risk category of each safety risk in the plurality of safety risks, and determine a corresponding process scheduling influence type based on the risk level and the risk category; a first mapping unit configured to establish an accident type mapping relationship between the risk category and a safety accident type, a level mapping relationship between each safety risk and the risk level, a risk category mapping relationship between each safety risk and the risk category, and a process mapping relationship between each safety risk and the process scheduling influence type based on the plurality of safety accidents, so as to construct the safety-process coupling knowledge sub-library according to the accident type mapping relationship, the level mapping relationship, the risk category mapping relationship, and the process mapping relationship; a determination unit configured to acquire a plurality of quality hidden troubles in the target construction process, and quantify a severity of each quality hidden trouble to obtain a corresponding severity level, and establish a plurality of process interaction influence modes based on the severity level, and determine a state index corresponding to each quality hidden trouble; a second mapping unit configured to respectively establish mapping relationships between the state index and the quality hidden trouble, the severity level, and the process interaction influence mode, so as to construct the quality-process coupling knowledge sub-library according to the mapping relationships; and an establishment unit configured to construct the safety-quality-process coupling evolution knowledge base based on the safety-process coupling knowledge sub-library and the quality-process coupling knowledge sub-library, and respectively establish a safety-process knowledge graph and a quality-process knowledge graph corresponding to the safety-process coupling knowledge sub-library and the quality-process coupling knowledge sub-library through a preset graph database strategy, so as to construct a corresponding visual coupling relationship knowledge graph according to the safety-process knowledge graph and the quality-process knowledge graph.

[0020] Optionally, in an embodiment of the present application, the modeling module comprises: a calculation unit configured to calculate a global environment state and a process feature state in a construction process corresponding to the original construction progress scheme, and determine the state space based on a preset double-layer state representation strategy in combination with the global environment state and the process feature state; a judgment unit configured to construct the action space based on a preset binary action representation strategy, and determine whether all predecessor processes of a current process are completed according to a construction progress of the original construction progress scheme, wherein, in the case that the all predecessor processes are completed, the state of the current process is updated to determine the state transition mechanism; and a generation unit configured to acquire disturbance data in the construction process corresponding to the original construction progress scheme, and generate corresponding processing measures according to the disturbance data and a preset response strategy to determine the disturbance mechanism.

[0021] Optionally, in one embodiment of the present application, the training module includes: a first construction unit, used to construct the hierarchical deep reinforcement learning network model based on a pre-constructed input layer, encoding layer, process attention layer, feature fusion layer, dual-stream architecture layer and output layer; an extraction unit, used to extract component geometric information and spatial relationships from a preset building information model, and obtain construction activity data corresponding to the component geometric information and the spatial relationship based on a preset four-dimensional work breakdown structure; a second construction unit, used to determine a plurality of process constraint rules based on the construction activity data, so as to construct a disturbance simulation generation algorithm according to the plurality of process constraint rules and the coupling relationship knowledge graph, and construct the construction simulation environment through the disturbance simulation generation algorithm, so as to train the hierarchical deep reinforcement learning network model through the preset conventional construction path and disturbance processing path under the construction simulation environment.

[0022] Optionally, in one embodiment of the present application, the mathematical expression of the multi-objective reward function is:

[0023] in, represents the state improvement reward; Indicates the parallel process reward; represents the idle time penalty; Indicates resource overlimit penalty; represents the time delay penalty.

[0024] The fourth aspect of the present application provides a method for online deduction of a digital twin of a construction process, which is applied to the online deduction stage, and includes the following steps: a monitoring module for monitoring the current construction site status corresponding to the target construction site, and obtaining corresponding disturbance information based on the current construction site status; an input module for inputting the disturbance information into a pre-built coupling relationship knowledge graph to output corresponding process scheduling plans and process adjustment suggestions; a deduction module for converting the format of the process scheduling plan and the process adjustment suggestions to obtain corresponding target format data, and inputting the target format data into a pre-trained hierarchical deep reinforcement learning network model to output the final process scheduling plan.

[0025] The fifth aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the online simulation method of the digital twin of the construction process as described in the above embodiment.

[0026] The sixth aspect of the embodiment of the present application provides a computer readable storage medium storing a computer program, which is executed by a processor to implement the construction process digital twin online deduction method.

[0027] Therefore, the embodiments of the present application have the following beneficial effects: The embodiments of the present application can construct a safety-quality-process coupling evolution knowledge base based on an original construction progress plan, a preset safety-process coupling knowledge base and a quality-process coupling knowledge base, and establish a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base; model the coupling relationship knowledge graph to determine a corresponding state space, action space, state transition mechanism and disturbance mechanism, and construct a multi-objective reward function corresponding to the original construction progress plan, to construct a process scheduling optimization model based on the state space, action space, state transition mechanism, multi-objective reward function and disturbance mechanism; determine a hierarchical deep reinforcement learning network model through the process scheduling optimization model, and construct a construction simulation environment using the hierarchical deep reinforcement learning network model, and train the hierarchical deep reinforcement learning network model based on the construction simulation environment, to perform digital twin online deduction on the original construction progress plan based on the trained hierarchical deep reinforcement learning network model in the online deduction stage. The present application can minimize the modification and adjustment of the process network when a sudden external influence or disturbance occurs, to ensure that the cost, duration and other targets remain unchanged as much as possible from the original plan, while minimizing the impact of process adjustment on personnel organization, material procurement and other work. In addition, the present application has the advantages of fast algorithm distance calculation speed in online deduction, and low process adjustment range and impact. Therefore, the problem that the existing technology considers that the dynamic change elements of the construction site are mainly resource changes and weather changes, and does not realize the response to construction site safety problems or quality hidden dangers and other events is solved.

[0028] Additional aspects and advantages of the present application will be in part apparent and in part pointed out hereinafter. BRIEF DESCRIPTION OF DRAWINGS

[0029] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which: Figure 1 A flowchart of a construction process digital twin online deduction method applied to an offline training stage according to an embodiment of the present application is provided. Figure 2 A process state transition mechanism schematic diagram provided by an embodiment of the present application is provided. Figure 3 A disturbance modeling mechanism schematic diagram provided by an embodiment of the present application is provided. Figure 4 An overall execution framework diagram of construction process digital twin online deduction provided for an embodiment of the present application; Figure 5 A building information model component information processing diagram provided for an embodiment of the present application; Figure 6 A model training flow diagram provided for an embodiment of the present application; Figure 7 A flowchart of a construction process digital twin online deduction method applied to an online deduction stage according to an embodiment of the present application; Figure 8 An online deduction flow diagram provided for an embodiment of the present application; Figure 9 An online deduction flow diagram provided for an embodiment of the present application; Figure 9 (a) in the above is a three-dimensional view of a case project provided for an embodiment of the present application; Figure 9 (b) in the above is a case project XY axis network division diagram provided for an embodiment of the present application; Figure 10 A case (1) schedule change Gantt chart provided for an embodiment of the present application; Figure 11 A case (1) cumulative progress curve comparison diagram provided for an embodiment of the present application; Figure 12 A case (1) various types of worker demand curve comparison diagram provided for an embodiment of the present application; Figure 13 A case (1) various types of material demand curve comparison diagram provided for an embodiment of the present application; Figure 14 A case (2) schedule change Gantt chart provided for an embodiment of the present application; Figure 15 A case (2) cumulative progress curve comparison diagram provided for an embodiment of the present application; Figure 16 A case (2) various types of worker demand curve comparison diagram provided for an embodiment of the present application; Figure 17 A case (2) various types of material demand curve comparison diagram provided for an embodiment of the present application; Figure 18 A case (3) schedule change Gantt chart provided for an embodiment of the present application; Figure 19 A case (3) cumulative progress curve comparison schematic diagram provided for an embodiment of the present application; Figure 20 A case (3) each type of worker demand curve comparison schematic diagram provided for an embodiment of the present application; Figure 21 A case (3) each type of material demand curve comparison schematic diagram provided for an embodiment of the present application; Figure 22 An example diagram of a construction process digital twin online deduction device applied in an offline training phase according to an embodiment of the present application; Figure 23 An example diagram of a construction process digital twin online deduction device applied in an online deduction phase according to an embodiment of the present application; Figure 24 A structural schematic diagram of an electronic device provided by an embodiment of the present application.

[0030] Among them, 10 - construction process digital twin online deduction device applied in offline training phase, 20 - construction process digital twin online deduction device applied in online deduction phase; 101 - atlas construction module, 102 - modeling module, 103 - training module; 201 - monitoring module, 202 - input module, 203 - deduction module; 2401 - memory, 2402 - processor, 2403 - communication interface. DETAILED DESCRIPTION

[0031] The embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0032] The construction process digital twin online deduction method and device of the embodiments of the present application are described below with reference to the accompanying drawings. In view of the problems mentioned in the background art, the present application provides a construction process digital twin online deduction method. In the method, a safety-quality-process coupling evolution knowledge base is constructed based on an original construction schedule, a preset safety-process coupling knowledge sub-base, and a quality-process coupling knowledge sub-base, and a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base is established. The coupling relationship knowledge graph is modeled to determine the corresponding state space, action space, state transition mechanism, and disturbance mechanism, and a multi-objective reward function corresponding to the original construction schedule is constructed, so as to construct a process scheduling optimization model based on the state space, action space, state transition mechanism, multi-objective reward function, and disturbance mechanism. The process scheduling optimization model is used to determine a hierarchical deep reinforcement learning network model, and the hierarchical deep reinforcement learning network model is used to construct a construction simulation environment. Based on the construction simulation environment, the hierarchical deep reinforcement learning network model is trained, so as to perform digital twin online deduction on the original construction schedule based on the trained hierarchical deep reinforcement learning network model in the online deduction stage. The present application can realize multi-element coupling evolution mechanism modeling of construction safety, quality, and process, construct a process scheduling optimization model based on Markov chain, and finally realize digital twin online deduction, thereby assisting construction decision-making. Thus, the problem that the prior art considers that the dynamic change elements of the construction site are mainly resource changes and weather changes, and does not realize response to construction site safety problems or quality hidden dangers and the like is solved.

[0033] Specifically, Figure 1 A flowchart of a construction process digital twin online deduction method applied to an offline training stage provided by the embodiments of the present application.

[0034] As Figure 1 shown, the construction process digital twin online deduction method includes the following steps: In step S101, a safety-quality-process coupling evolution knowledge base is constructed based on an original construction schedule, a preset safety-process coupling knowledge sub-base, and a quality-process coupling knowledge sub-base, and a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base is established.

[0035] The embodiments of the present application first establish a structured safety-quality-process coupling evolution knowledge base based on the safety-process coupling knowledge sub-base and the quality-process coupling knowledge sub-base, thereby providing a core knowledge base for generating scheduling responses by constructing a structured knowledge base covering safety risks, quality hidden dangers, and process dynamic interaction mechanisms.

[0036] Optionally, in an embodiment of the present application, based on the original construction progress plan, the preset safety-process coupling knowledge base and the quality-process coupling knowledge base, a safety-quality-process coupling evolution knowledge base is constructed, and a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base is established, including: obtaining multiple safety risks and multiple safety accidents in the target construction process, and determining the risk level and risk category of each safety risk in the multiple safety risks, and based on the risk level and risk category, determining the corresponding process scheduling influence type; based on the multiple safety accidents, establishing an accident type mapping relationship between the risk category and the safety accident type, a level mapping relationship between each safety risk and the risk level, a risk category mapping relationship between each safety risk and the risk category, and a process mapping relationship between each safety risk and the process scheduling influence type, to construct a safety-process coupling knowledge base according to the accident type mapping relationship, the level mapping relationship, the risk category mapping relationship and the process mapping relationship; obtaining multiple quality hidden dangers in the target construction process, and quantifying the severity of each quality hidden danger to obtain a corresponding severity level, and based on the severity level, establishing multiple process interaction influence modes and determining the state index corresponding to each quality hidden danger; respectively establishing mapping relationships between the state index and the quality hidden danger, the severity level and the process interaction influence mode, to construct a quality-process coupling knowledge base according to the mapping relationships; based on the safety-process coupling knowledge base and the quality-process coupling knowledge base, constructing a safety-quality-process coupling evolution knowledge base, and respectively establishing a safety-process knowledge graph and a quality-process knowledge graph corresponding to the safety-process coupling knowledge base and the quality-process coupling knowledge base through a preset graph database strategy, to construct a corresponding visual coupling relationship knowledge graph according to the safety-process knowledge graph and the quality-process knowledge graph.

[0037] In actual execution, the embodiment of the present application can deeply investigate domestic and foreign construction field specifications, related literature and construction accident case data, systematically sort out safety risks, safety accidents and quality hidden dangers in the construction process, and establish a structured safety-quality-process coupling evolution knowledge base, aiming to systematically store dynamic scheduling rules between safety problems, quality hidden dangers and process scheduling in the construction process, and provide core knowledge support for realizing digital twin driven construction process online deduction. The knowledge base is composed of two core sub-bases, a safety-process coupling knowledge base and a quality-process coupling knowledge base.

[0038] Among them, the core entities of the safety-process coupling knowledge base are 138 safety risks and 12 safety accidents, the safety risks cover five categories of personnel factors, object factors, environmental factors, management factors and structural factors, and the safety accidents include vehicle injury, object strike, subsidence, lifting injury, electric shock, poisoning / suffocation, falling from height, fire, mechanical injury, collapse and overturning, etc.

[0039] Based on a large number of case analyses, the embodiments of the present application can determine the risk levels of different safety problems, and summarize four typical interaction modes of process parameter adjustment type, process increment type, process replacement type and process lag type, and the interaction modes are described as follows: 1) The process parameter adjustment type is mainly for slight safety risks, and only adjusts the time or resource parameters of the process without changing the logical structure of the process; 2) The process increment type is suitable for larger safety hazards, and needs to add new safety prevention steps based on the original process; 3) The process replacement type deals with serious safety hazards, and needs to reorganize or completely modify the original process; 4) The process lag type is used to deal with serious safety accidents that have occurred, resulting in the overall suspension and rectification of construction.

[0040] Thus, the safety-process coupling knowledge sub-library clearly establishes four types of mapping relationships of safety risk type-possible accident type (i.e. accident type mapping relationship), safety risk-risk level (i.e. level mapping relationship), safety risk-risk category (i.e. risk category mapping relationship), and safety risk-process scheduling influence type (i.e. process mapping relationship).

[0041] The core entities of the quality-process coupling knowledge sub-library are nine quality hazards that occur in reinforced concrete structures, including honeycomb, cracks, peeling, cavities, exposed reinforcement, dimensional deviation, structural displacement, insufficient formwork stability and steel bar binding quality defects.

[0042] Based on a large number of case analyses, the severity of each type of quality hazard is quantified, forming three typical interaction modes of process parameter adjustment type, stage repair embedding type and emergency defect repair type, and the interaction modes are described as follows: 1) The process parameter adjustment type is suitable for slight quality problems, which can be solved by adjusting the process time or resource parameters; 2) The stage repair embedding type is for moderate quality hazards, and inserts a repair process after the completion of the current construction stage; 3) The emergency defect repair type deals with serious quality problems, and needs to immediately interrupt other processes for repair.

[0043] Thus, the quality-process coupling knowledge sub-library of the embodiments of the present application clearly establishes three types of mapping relationships of quality hazard-state indicator, state indicator-severity level, and state indicator-process scheduling influence type.

[0044] To realize the structured storage of the coupling mechanism, the embodiment of the application can use Neo4j graph database technology to construct a safety-quality-process coupling relationship knowledge graph. In the safety-process knowledge graph, six types of core nodes are designed, including safety risks, sub-part engineering, accident types, risk levels, risk categories, and process scheduling types, as well as five types of relationships between nodes; in the quality-process knowledge graph, five types of core nodes and four types of node relationships are designed. The finally formed knowledge graph contains 138 safety risks, 12 accident types, 13 quality hidden dangers, and 7 corresponding process scheduling strategies, thereby realizing the intuitive visualization and computer-readable structured expression of the multi-element coupling relationship of construction, and providing a knowledge base for subsequent digital twin driven process deduction.

[0045] In step S102, the coupling relationship knowledge graph is modeled, the corresponding state space, action space, state transition mechanism and disturbance mechanism are determined, and a multi-objective reward function corresponding to the original construction schedule scheme is constructed, so as to construct a process scheduling optimization model based on the state space, action space, state transition mechanism, multi-objective reward function and disturbance mechanism.

[0046] Further, the embodiment of the application can design a reinforcement learning network architecture specially used for construction process dynamic scheduling according to the uniqueness of construction process scheduling, so as to construct a reinforcement learning network and model for process scheduling, thereby realizing intelligent scheduling decision.

[0047] Optionally, in an embodiment of the application, the coupling relationship knowledge graph is modeled, the corresponding state space, action space, state transition mechanism and disturbance mechanism are determined, and a multi-objective reward function corresponding to the original construction schedule scheme is constructed, so as to construct a process scheduling optimization model based on the state space, action space, state transition mechanism, multi-objective reward function and disturbance mechanism, including: calculating the global environmental state and process characteristic state in the construction process corresponding to the original construction schedule scheme, and determining the state space based on a preset double-layer state representation strategy combined with the global environmental state and process characteristic state; constructing the action space based on a preset binary action representation strategy, and judging whether all predecessor processes of the current process are completed according to the construction schedule of the original construction schedule scheme, wherein, in the case that all predecessor processes are completed, the state of the current process is updated to determine the state transition mechanism; obtaining disturbance data in the construction process corresponding to the original construction schedule scheme, and generating corresponding processing measures according to the disturbance data and a preset response strategy to determine the disturbance mechanism.

[0048] It should be noted that the embodiment of the application can construct a process scheduling optimization model based on Markov decision process theory, so that the process scheduling problem can be abstracted into a mathematical form for representation and solution, and the specific process is as follows: (1) State space modeling: In the embodiments of the present application, the state space modeling can adopt a double-layer state representation method to comprehensively capture the construction environment state and process feature information.

[0049] The first layer is a global environment state , which is used to describe the macro characteristics of the construction simulation environment, as shown in formula (1), including five key dimensions of average resource utilization rate, resource utilization rate standard deviation, process completion rate, completion rate standard deviation and estimated delay; the average resource utilization rate reflects the overall efficiency of resource use; the resource utilization rate standard deviation measures the balance of resource utilization of different types of workers; the process completion rate reflects the overall progress of the project; the completion rate standard deviation is calculated based on the Bernoulli distribution, which helps the model to identify different stages of the project; and the estimated delay is used to predict the possible delay of the project.

[0050] (1) wherein, represents the average resource utilization rate; represents the resource utilization rate standard deviation; represents the process completion rate; represents the completion rate standard deviation; represents the estimated delay.

[0051] The second layer is a process feature state, as shown in formula (2), which is used to describe the micro characteristics of each specific process, including seven dimensions: planned duration, number of workers required, number of materials required, floor, X direction partition information, Y direction partition information and number of successor processes. Each process can be in one of the four discrete states at any time: not ready (all predecessor processes have not been completed), ready (all predecessor processes have been completed), in progress (has been selected for execution) and completed (execution is complete).

[0052] (2) wherein, represents the planned duration; represents the number of workers required; represents the number of materials required; represents the floor; represents the X direction partition information; represents the Y direction partition information; represents the number of successor processes.

[0053] (2) Action space modeling Action space modeling defines how the system makes process scheduling decisions. The embodiments of the present application use a binary action representation method to independently decide whether to start each process that has been prepared. If there are multiple ready processes at present, the action space can be represented as a multi-dimensional binary vector, and each element in the vector takes the value of 0 or 1, 1 indicating starting the process, and 0 indicating not starting. The advantage of this representation method is that it can flexibly handle the case of parallel execution of multiple processes, which conforms to the parallel operation characteristics in actual construction; at the same time, the action selection of the process is limited by multiple constraint conditions, mainly including logical dependency constraints and working time constraints. The logical dependency constraint ensures that the process can only be started after all its predecessor processes are completed, which is a basic requirement to ensure the correctness of the construction process; the working time constraint considers the working time setting in the present construction (such as 9 a.m.-17 p.m.), and if the process duration crosses the non-working time, the system will appropriately extend the completion time. The binary action space design has higher flexibility, and can ensure the engineering rationality of the scheduling scheme in combination with the logical dependency and working time constraints.

[0054] (3) Process state transition mechanism The process state transition mechanism (i.e., the state transition mechanism) is the core logic of the entire scheduling system, which controls how the process changes from one state to another. As shown in Figure 2 , at system initialization, all processes are in the "not ready" state by default, and the processes without predecessor process dependency will automatically change to the "ready" state, indicating that they can immediately start execution. As the construction process advances, when all predecessor processes of a process are completed, its state changes from "not ready" to "ready", and enters the schedulable queue. The system selects one or more processes from the schedulable queue for execution, and the selected process state is updated to "in progress" and its planned completion time is recorded. When the process reaches the planned completion time, its state is automatically updated to "completed", and the state update of its successor process is triggered.

[0055] (4) Disturbance factor modeling Disturbance factor modeling unifies the uncertainty in the construction process as system disturbance, including external variables such as safety problems, quality hidden dangers, resource supply fluctuations, and weather changes. The specific modeling mechanism is shown in Figure 3 . Specifically, when a disturbance in the construction environment is detected, the system first records the type, severity, location information, and influence range of the disturbance, then generates processing measures based on the preset response strategy, dynamically modifies the dependency network of the process, and finally re-plans the optimal process execution plan.

[0056] For different types of disturbances, the embodiments of the present application design different processing mechanisms, as shown in Table 1.

[0057] Table 1

[0058] For quality problems, according to its severity is divided into three levels of slight, medium and serious, the embodiment of the application can adopt different processing strategies: slight hidden danger only to strengthen supervision, medium hidden danger after all processes are completed in the same floor are repaired, serious hidden danger is repaired immediately and all the processes that have not started need to wait for the repair to be completed. For safety risks, also according to the severity is divided into three levels, using process parameter adjustment, increasing preparation process or replacing the original process. For safety accidents, according to the casualty to determine the downtime all the processes that have not started need to wait for the end of the downtime. The mechanism can systematically combine various unexpected situations in the construction site with the process optimization decision framework, and provide key environmental interaction capabilities for the intelligent agent to learn to adaptively adjust the scheduling scheme under disturbance.

[0059] Optionally, in an embodiment of the application, the mathematical expression of the multi-objective reward function is:

[0060] wherein, state improvement reward; parallel process reward; represents idle time penalty; represents resource overrun penalty; represents time delay penalty.

[0061] In addition, the embodiment of the application also needs to design a multi-objective reward mechanism in the process of constructing the process scheduling optimization model. The multi-objective reward mechanism design comprehensively considers various optimization objectives in the actual construction process scheduling, forming a multi-level and multi-objective comprehensive evaluation system, as shown in equation (3): (3) wherein, state improvement reward; parallel process reward; represents idle time penalty; represents resource overrun penalty; represents time delay penalty.

[0062] It should be noted that the multi-objective reward mechanism includes five core rewards or penalties, namely, state improvement reward, parallel process reward, idle time penalty, resource over-limit penalty, and time delay penalty. Among them, the state improvement reward evaluates the change of the global environment state after each decision, and positive rewards are obtained when the average resource utilization is improved, the resource utilization standard deviation is reduced, the completion rate is improved, and the delay rate is reduced; the parallel process reward is used to evaluate the number of processes performed simultaneously in each state, and to encourage decisions to fully utilize resource availability; the idle time penalty gives a certain penalty when there is a ready process but no process is started, avoiding downtime or resource waste; the resource over-limit penalty gives a penalty when resource usage exceeds the resource limit, ensuring that the scheduling will not allocate too much resource; the time delay penalty punishes the project delay, and early completion will obtain additional rewards.

[0063] Therefore, the embodiment of the present application can effectively guide the agent to learn a scheduling strategy that balances the conflict of multiple objectives by constructing a multi-objective reward function, which is the core driving force for realizing process optimization scheduling.

[0064] In step S103, a hierarchical deep reinforcement learning network model is determined through the process scheduling optimization model, and a construction simulation environment is constructed using the hierarchical deep reinforcement learning network model. Based on the construction simulation environment, the hierarchical deep reinforcement learning network model is trained to perform digital twin online reasoning on the original construction progress plan based on the trained hierarchical deep reinforcement learning network model in the online reasoning phase.

[0065] After that, the embodiment of the present application can also determine an adaptive model training method considering disturbance events to solve the problem that the reinforcement learning model is difficult to effectively learn and respond to sudden construction disturbances, so that the model can learn how to dynamically adjust the process network when the disturbance occurs.

[0066] Therefore, the embodiment of the present application can realize the multi-element coupling evolution mechanism modeling of construction safety, quality, and process by the original construction progress plan (i.e. the original construction progress plan) and the real-time construction quality / safety problem, construct a process scheduling optimization model based on Markov chain, and finally realize digital twin online reasoning in the subsequent online reasoning phase, thereby assisting construction decision-making, as shown in the overall execution framework. Figure 4 The process scheduling optimization model based on Markov chain and digital twin online reasoning are core steps and core requirements.

[0067] Optionally, in an embodiment of the present application, a hierarchical deep reinforcement learning network model is determined through a process scheduling optimization model, a construction simulation environment is constructed using the hierarchical deep reinforcement learning network model, and the hierarchical deep reinforcement learning network model is trained based on the construction simulation environment, including: based on a pre-constructed input layer, an encoding layer, a process attention layer, a feature fusion layer, a double-flow architecture layer, and an output layer, a hierarchical deep reinforcement learning network model is constructed; component geometric information and spatial relationships are extracted from a pre-set building information model, and construction activity data corresponding to the component geometric information and spatial relationships is obtained based on a pre-set four-dimensional work breakdown structure; based on the construction activity data, a plurality of process constraint rules are determined, a disturbance simulation generation algorithm is constructed according to the plurality of process constraint rules and the coupling relationship knowledge graph, and the construction simulation environment is constructed through the disturbance simulation generation algorithm, so that the hierarchical deep reinforcement learning network model is trained through a pre-set conventional construction path and a disturbance processing path in the construction simulation environment.

[0068] As an implementable way, the embodiments of the present application can use a hierarchical deep reinforcement learning algorithm (Hierarchical Deep Q-Network, H-DQN) as a basic algorithm framework for process scheduling to realize construction process scheduling decision-making under disturbance. The algorithm uses hierarchical strategy, on the one hand, determines the target or sub-task (such as selecting the target of optimizing the construction period, safety priority or quality priority) through meta-strategy (high layer), and on the other hand, executes specific scheduling actions (such as selecting specific processes or resource allocation) through low layer strategy. The specific process is as follows: (1) H-DQN network architecture design The embodiments of the present application design a hierarchical deep reinforcement learning network architecture specially used for high-dimensional and multi-constrained construction process dynamic scheduling problems. The architecture effectively integrates the domain knowledge of construction scheduling, solves the problems that traditional methods are difficult to handle large-scale processes, high-dimensional action space (process parallel selection), and sudden disturbance through multi-level feature extraction, attention mechanism capturing complex dependencies, high layer strategy guidance, double-flow value evaluation, etc., and contains six main components: input layer, encoding layer, process attention layer, feature fusion layer, double-flow architecture layer, and output layer. The specific introduction of each component is as follows: 1) The input layer adopts a double-channel parallel design, which receives global environment state and process feature state information respectively, avoids information interference, and lays a foundation for subsequent hierarchical processing; 2) The encoding layer uses nonlinear transformation to map the original input to a high-dimensional representation space; 3) The process attention layer captures the complex dependency relationship between processes through four attention heads (resource dependency head, spatial proximity head, time sequence dependency head, and duration head) specially designed for construction process scheduling, which can improve the model's understanding ability of complex process networks; 4) The feature fusion layer combines global information with process features to form a comprehensive representation, allowing each process decision to consider both its own characteristics and the global environmental state; 5) The double-flow architecture layer decomposes the Q value into state value and action advantage, improving the model evaluation capability; 6) The output layer generates process selection decisions, deeply integrating high-level strategy guidance and low-level Q value evaluation to achieve intelligent and adaptive process selection.

[0069] (2) Construction simulation environment construction To establish a virtual environment that can accurately map the actual construction process, the embodiment of the application designs a process scheduling simulation environment based on BIM (Building Information Modeling). It mainly includes four steps: BIM data extraction and processing, work breakdown structure (WBS) data extraction, process constraint rule definition, and disturbance simulation generation. The specific introduction of each step is as follows: 1) Extract component geometric information and spatial relationships from BIM models through Revit-Dynamo plugins, including coordinate extraction, vertical partition judgment, horizontal partition judgment, and attribute marking, to convert building entities into digital representations, as shown in Figure 5 ; 2) Build a four-dimensional work breakdown structure to decompose construction activities into four dimensions: component type, process type, material type, and worker type; 3) Define six process constraint rules, including Finish to Start rules, component construction sequence rules, vertical priority rules, intra-regional construction sequence rules, sequential construction rules, and resource constraint rules; 4) Based on the safety-quality-process coupling relationship knowledge graph, develop a disturbance simulation generation algorithm to realize the whole process simulation from disturbance type determination to structured data generation.

[0070] (3) Simulation training The model training process of the embodiment of the application adopts a parallel architecture of conventional construction path and disturbance handling path, as shown in Figure 6 . The conventional construction path handles normal process execution under non-disturbance conditions, including four steps of disturbance detection, original plan execution, process start, and time advancement; the disturbance handling path is activated when a sudden situation is detected, including six steps of disturbance information recording, disturbance impact handling, dispatching system activation, Q value calculation, process selection, and model updating. The key optimization parameters in the model training process are shown in Table 2.

[0071] Table 2

[0072] Therefore, the embodiment of the application realizes modeling analysis of the coupling relationship between construction safety, quality and process, sorts out and presents the correlation between 138 kinds of safety risks, 12 types of accidents, 13 types of quality hidden dangers and 7 kinds of process influence modes, can more accurately reflect the mutual influence relationship between various elements in the construction process, and can more completely describe and analyze the complexity of the construction process; in addition, the embodiment of the application can be well applied to specific scenes such as real-time risk early warning, multi-objective scheduling optimization and three-dimensional visual decision of the construction process of the building engineering, for the multi-element coupling problem of safety problems, quality hidden dangers and process logic in the construction process, through the digital twin dynamic deduction method based on deep reinforcement learning and building information model.

[0073] According to the construction process digital twin online deduction method applied to the offline training stage provided by the embodiment of the application, a safety-quality-process coupling evolution knowledge base is constructed based on the original construction progress scheme, the preset safety-process coupling knowledge sub-base and the quality-process coupling knowledge sub-base, and a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base is established; the coupling relationship knowledge graph is modeled to determine the corresponding state space, action space, state transition mechanism and disturbance mechanism, and a multi-objective reward function corresponding to the original construction progress scheme is constructed, to construct a process scheduling optimization model based on the state space, action space, state transition mechanism, multi-objective reward function and disturbance mechanism; the process scheduling optimization model is used to determine a hierarchical deep reinforcement learning network model, and the hierarchical deep reinforcement learning network model is used to construct a construction simulation environment, and the hierarchical deep reinforcement learning network model is trained based on the construction simulation environment, to perform digital twin online deduction on the original construction progress scheme based on the trained hierarchical deep reinforcement learning network model in the online deduction stage. The application can realize modeling of the multi-element coupling evolution mechanism of construction safety, quality and process, construct a process scheduling optimization model based on Markov chain, and finally realize digital twin online deduction, thereby assisting construction decision-making.

[0074] Figure 7 The flowchart of the construction process digital twin online deduction method applied to the online deduction stage provided by the embodiment of the application.

[0075] As Figure 7 shown, the construction process digital twin online deduction method includes the following steps: In step S701, the current construction site state corresponding to the target construction site is monitored, and the corresponding disturbance information is obtained according to the current construction site state.

[0076] In step S702, the disturbance information is input into the pre-constructed coupling relationship knowledge graph to output the corresponding process scheduling scheme and process adjustment suggestion.

[0077] In step S703, the process scheduling scheme and the process adjustment suggestion are format-converted to obtain corresponding target format data, and the target format data are input into the pre-trained hierarchical deep reinforcement learning network model to output a final process scheduling scheme.

[0078] It should be noted that when the trained model is deployed to an actual construction environment, the embodiment of the present application realizes an online deduction process based on digital twinning, as shown in Figure 8 The specific process is as described below: (1) Disturbance information acquisition Embodiments of the present application can first monitor the construction site state in real time through digital twinning technology, and when detecting safety risks, quality hazards or other disturbance events, the system automatically records the detailed information of the disturbance, including problem description, occurrence location, occurrence time, severity, and other key parameters, such as personnel casualty in the case of safety accidents; (2) Knowledge graph processing Secondly, embodiments of the present application can input the acquired disturbance information (such as sudden safety problems, quality problems or material supply shortage, weather mutation and other influencing factors on normal construction) into the pre-constructed construction safety-quality-process coupling correlation knowledge graph for processing, obtain the corresponding process scheduling scheme through the matching of the problem and the knowledge graph, and generate targeted process adjustment suggestions; (3) Model input Thirdly, embodiments of the present application can convert the scheduling suggestions, safety risk assessment, quality control measures and other information output by the knowledge graph into a data format suitable for input of the H-DQN model, including process state, executed process, to-be-executed process and related constraint conditions and other state representations; (4) Model scheduling decision Finally, embodiments of the present application can call the trained H-DQN model to make intelligent scheduling decisions for the processes that have not been executed, and the model calculates the Q value of each process according to the current construction state and disturbance characteristics, and outputs an optimal process scheduling scheme (i.e., the final process scheduling scheme) based on multi-objective optimization.

[0079] Further, embodiments of the present application can send the generated optimal process scheduling scheme to the site management personnel through the construction management platform, provide specific process execution suggestions and resource allocation guidance, and realize full-process automated response from disturbance detection to scheduling execution.

[0080] Thus, embodiments of the present application construct a process online deduction system based on digital twinning data with real-time response capability, so that when safety problems or quality hazards and other events occur, the dynamic changes, working state, risk level and other characteristics of the construction site can be reflected and appropriate process optimization scheduling can be performed.

[0081] The construction process digital twin online deduction method execution logic and effect of the present application are described below through a specific embodiment of a certain reinforced concrete structure residential building group as an example, combined with the accompanying drawings.

[0082] Figure 9 The online deduction schematic diagram for the certain reinforced concrete structure residential building group engineering case in the specific embodiment is shown in FIG. 1. Figure 9 As shown in the specific embodiment of the present application, the certain reinforced concrete structure residential building group engineering contains 5 buildings, each with 4 floors, as shown in FIG. 1(a). Figure 9 The engineering adopts a traditional reinforced concrete frame structure, and the main components include four categories of structural columns, structural walls, beams, and floors. Through BIM model establishment and axis network division, the entire construction area is divided into a grid system of 6 regions in the X direction and 4 regions in the Y direction, forming 24 basic construction units, as shown in FIG. 1(b). Figure 9

[0083] The certain reinforced concrete structure residential building group engineering is divided into 660 processes, including 165 processes each for steel bar binding, formwork, concrete pouring, and concrete pouring, with a total concrete volume of 8491.3 cubic meters. According to the construction constraint rules, 1575 tight front and tight back process relationships are established, with the number of predecessor processes for a single process ranging from 0 to 3. The specific embodiment of the present application uses the Minizinc constraint programming method to generate an original static scheduling plan, with the shortest construction period as the optimization objective, and obtains an optimal plan construction period of 789 days.

[0084] After that, the embodiment of the present application can select three representative quality hidden trouble disturbance cases for detailed analysis and verification: Case (1): On the 154th day of construction, a wall crack with a moderate severity was found in the 1st floor, 4th region in the X direction, and 1st region in the Y direction; at this time, 353 processes had been completed, and the completion progress was 53.6%.

[0085] Case (2): On the 397th day of construction, a wall crack with a moderate severity was found in the 1st floor, 4th region in the X direction, and 1st region in the Y direction; at this time, 572 processes had been completed, and the completion progress was 86.7%.

[0086] Case (3): On the 154th day of construction, a wall crack with a severe severity was found in the 1st floor, 4th region in the X direction, and 1st region in the Y direction; at this time, 353 processes had been completed, and the completion progress was 53.6%.

[0087] ​Through real-time monitoring by digital twinning technology, the embodiments of the present application can detect quality problems and automatically record detailed information, such as for case (1), the recorded information includes the problem description "cracks appear in the concrete wall", the location of occurrence "1st floor X4Y1 area", the severity "moderate", the discovery time "2024-06-01 10:30:00" and other key parameters.

[0088] The embodiments of the present application can input the disturbance information into the pre-constructed quality-process coupling relationship knowledge graph, and match the related nodes and relationships through the Cypher query language. According to the mapping relationship "concrete wall cracks-moderate severity-stage repair embedded type" in the knowledge graph, it is determined that a repair process needs to be inserted after all processes on this floor are completed.

[0089] Further, the embodiments of the present application can convert the disturbance information into the input format of the H-DQN model. The model solving times are (1) 13.02 seconds, (2) 2.95 seconds and (3) 7.99 seconds, respectively. In case (1), it is determined to perform the repair process on the 271st day, and the total construction period is increased by 2 days; in case (2), it is determined to perform the repair process on the 407th day, and the total construction period is increased by 1 day; in case (3), it is determined to perform the repair process on the 155th day, and the total construction period is increased by 3 days.

[0090] The scheduling results are displayed through a visual interface. The Gantt chart shows that the model strictly follows the original plan and maintains the established construction sequence before the disturbance occurs. As shown in Figure 10 , after the disturbance in case (1), the embodiments of the present application add one repair process and adjust the time arrangement of subsequent processes. As shown in Figure 11 , the cumulative progress curve shows that although the mid-term progress is slightly delayed, through reasonable resource allocation, the progress loss is successfully made up in the later period. Figure 12 Figure 13 As shown in

[0091] Similarly, the scheduling change Gantt chart, cumulative progress curve comparison, type worker demand curve comparison and type material demand curve comparison of case (2) are shown in Figure 14 , Figure 15 , Figure 16 and Figure 17 , respectively. The scheduling change Gantt chart, cumulative progress curve comparison, type worker demand curve comparison and type material demand curve comparison of case (3) are shown in Figure 18 , Figure 19 , Figure 20 , Figure 21 ​As shown; where. Case (2) has fewer subsequent incomplete process quantities, and the adjustment is relatively small, so the red line and the blue line in the cumulative progress curve are highly coincident; in case (3), since the risk level of the disturbance is serious, the repair process needs to be executed immediately, so the repair process is immediately executed in the scheduling scheme, which conforms to the scheduling logic.

[0092] In summary, the hierarchical deep reinforcement learning network model has fast response speed when dealing with disturbance cases, can complete scheduling decisions for more than 600 processes within 15 seconds, has small impact on the construction period, balanced resource utilization, and no resource over-limit situation; in addition, the hierarchical deep reinforcement learning network model scheduling strategy is reasonable, the repair process is arranged at the appropriate time point, and excessive interference to the critical path is avoided.

[0093] According to the construction process digital twin online deduction method applied to the online deduction stage, the current construction site state corresponding to the target construction site is monitored, and the corresponding disturbance information is obtained according to the current construction site state; the disturbance information is input into the pre-constructed coupling relationship knowledge graph to output the corresponding process scheduling scheme and process adjustment suggestion; the process scheduling scheme and the process adjustment suggestion are format-converted to obtain corresponding target format data, and the target format data are input into the pre-trained hierarchical deep reinforcement learning network model to output the final process scheduling scheme. The application can realize the modeling of the multi-element coupling evolution mechanism of construction safety, quality and process, construct a process scheduling optimization model based on Markov chain, and finally realize digital twin online deduction, thereby assisting construction decision-making.

[0094] Secondly, the construction process digital twin online deduction device according to the embodiment of the application is described with reference to the accompanying drawings.

[0095] Figure 22 is a block schematic diagram of the construction process digital twin online deduction device applied to the offline training stage of the embodiment of the application.

[0096] As Figure 22 shown, the construction process digital twin online deduction device applied to the offline training stage 10 includes a graph construction module 101, a modeling module 102, and a training module 103.

[0097] The graph construction module 101 is configured to construct a safety-quality-process coupling evolution knowledge base based on an original construction progress scheme, a preset safety-process coupling knowledge sub-library, and a quality-process coupling knowledge sub-library, and establish a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base.

[0098] The modeling module 102 is configured to model the coupling relationship knowledge graph, determine a corresponding state space, an action space, a state transition mechanism and a disturbance mechanism, and construct a multi-objective reward function corresponding to the original construction progress plan, so as to construct a process scheduling optimization model based on the state space, the action space, the state transition mechanism, the multi-objective reward function and the disturbance mechanism.

[0099] The training module 103 is configured to determine a hierarchical deep reinforcement learning network model through the process scheduling optimization model, construct a construction simulation environment by using the hierarchical deep reinforcement learning network model, and train the hierarchical deep reinforcement learning network model based on the construction simulation environment, so as to perform digital twin online deduction on the original construction progress plan based on the trained hierarchical deep reinforcement learning network model in the online deduction stage.

[0100] Optionally, in an embodiment of the present application, the graph construction module 101 comprises an acquisition unit, a first mapping unit, a determination unit, a second mapping unit and a building unit.

[0101] The acquisition unit is configured to acquire a plurality of safety risks and a plurality of safety accidents in a target construction process, determine a risk level and a risk category of each safety risk in the plurality of safety risks, and determine a corresponding process scheduling influence type based on the risk level and the risk category.

[0102] The first mapping unit is configured to establish an accident type mapping relationship between the risk categories and the safety accident types, a level mapping relationship between each safety risk and the risk level, a risk category mapping relationship between each safety risk and the risk category, and a process mapping relationship between each safety risk and the process scheduling influence type based on the plurality of safety accidents, so as to construct a safety-process coupling knowledge sub-library according to the accident type mapping relationship, the level mapping relationship, the risk category mapping relationship and the process mapping relationship.

[0103] The determination unit is configured to acquire a plurality of quality hidden troubles in the target construction process, quantify a severity of each quality hidden trouble to obtain a corresponding severity level, establish a plurality of process interaction influence modes based on the severity level, and determine a state index corresponding to each quality hidden trouble.

[0104] The second mapping unit is configured to respectively establish mapping relationships between the state index and the quality hidden trouble, the severity level and the process interaction influence mode, so as to construct a quality-process coupling knowledge sub-library according to the mapping relationships.

[0105] The establishment unit is configured to construct a safety-quality-process coupling evolution knowledge base based on the safety-process coupling knowledge base and the quality-process coupling knowledge base, and establish a safety-process knowledge graph and a quality-process knowledge graph corresponding to the safety-process coupling knowledge base and the quality-process coupling knowledge base respectively through a preset graph database strategy, so as to construct a corresponding visual coupling relationship knowledge graph according to the safety-process knowledge graph and the quality-process knowledge graph.

[0106] Optionally, in an embodiment of the present application, the modeling module 102 comprises a calculation unit, a judgment unit and a generation unit.

[0107] The calculation unit is configured to calculate a global environment state and a process feature state in a construction process corresponding to the original construction schedule scheme, and determine a state space based on a preset double-layer state representation strategy in combination with the global environment state and the process feature state.

[0108] The judgment unit is configured to construct an action space based on a preset binary action representation strategy, and determine whether all predecessor processes of a current process are completed according to the construction schedule of the original construction schedule scheme, wherein in the case that all the predecessor processes are completed, the state of the current process is updated to determine a state transition mechanism.

[0109] The generation unit is configured to obtain disturbance data in the construction process corresponding to the original construction schedule scheme, and generate corresponding processing measures according to the disturbance data and a preset response strategy to determine a disturbance mechanism.

[0110] Optionally, in an embodiment of the present application, the training module 103 comprises a first construction unit, an extraction unit and a second construction unit.

[0111] The first construction unit is configured to construct a hierarchical deep reinforcement learning network model based on a previously constructed input layer, an encoding layer, a process attention layer, a feature fusion layer, a double-flow architecture layer and an output layer.

[0112] The extraction unit is configured to extract component geometric information and spatial relationships from a preset building information model, and obtain construction activity data corresponding to the component geometric information and the spatial relationships based on a preset four-dimensional work breakdown structure.

[0113] The second construction unit is configured to determine a plurality of process constraint rules based on the construction activity data, construct a disturbance simulation generation algorithm according to the plurality of process constraint rules and the coupling relationship knowledge graph, and construct a construction simulation environment through the disturbance simulation generation algorithm, so as to train the hierarchical deep reinforcement learning network model through a preset conventional construction path and a disturbance processing path in the construction simulation environment.

[0114] Optionally, in an embodiment of the present application, the mathematical expression of the multi-objective reward function is:

[0115] wherein, represents a state improvement reward; represents a parallel process reward; represents an idle time penalty; represents a resource overrun penalty; represents a time delay penalty.

[0116] It should be noted that the foregoing explanation and description of the construction process digital twin online deduction method embodiment applied to the offline training stage also applies to the construction process digital twin online deduction device 10 of the embodiment applied to the offline training stage, which will not be described here.

[0117] According to the construction process digital twin online deduction device 10 applied to the offline training stage, the atlas construction module 101 is used to construct a safety-quality-process coupling evolution knowledge base based on the original construction progress plan, the preset safety-process coupling knowledge sub-library and the quality-process coupling knowledge sub-library, and establish a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base; the modeling module 102 is used to model the coupling relationship knowledge graph, determine the corresponding state space, action space, state transition mechanism and disturbance mechanism, and construct a multi-objective reward function corresponding to the original construction progress plan, so as to construct a process scheduling optimization model based on the state space, the action space, the state transition mechanism, the multi-objective reward function and the disturbance mechanism; the training module 103 is used to determine a hierarchical deep reinforcement learning network model through the process scheduling optimization model, and construct a construction simulation environment using the hierarchical deep reinforcement learning network model, and train the hierarchical deep reinforcement learning network model based on the construction simulation environment, so as to perform digital twin online deduction on the original construction progress plan based on the trained hierarchical deep reinforcement learning network model in the online deduction stage. The present application can realize construction safety, quality and process multi-element coupling evolution mechanism modeling, construct a process scheduling optimization model based on Markov chain, and finally realize digital twin online deduction, thereby assisting construction decision-making.

[0118] Figure 23 FIG. 1 is a block schematic diagram of a construction process digital twin online deduction device applied to an online deduction stage according to an embodiment of the present application.

[0119] As Figure 23 shown, the construction process digital twin online deduction device 20 applied to the online deduction stage includes a monitoring module 201, an input module 202 and a deduction module 203.

[0120] The monitoring module 201 is configured to monitor a current construction site state corresponding to the target construction site, and obtain corresponding disturbance information according to the current construction site state.

[0121] The input module 202 is configured to input the disturbance information into a pre-constructed coupling relationship knowledge graph, to output a corresponding process scheduling scheme and a process adjustment suggestion.

[0122] The deduction module 203 is configured to perform format conversion on the process scheduling scheme and the process adjustment suggestion, to obtain corresponding target format data, and input the target format data into a pre-trained hierarchical deep reinforcement learning network model, to output a final process scheduling scheme.

[0123] It should be noted that the foregoing explanation and description of the construction process digital twin online deduction method applied to the online deduction phase also applies to the construction process digital twin online deduction device 20 applied to the online deduction phase of the embodiment, which will not be described here.

[0124] The construction process digital twin online deduction device 20 applied to the online deduction phase according to the embodiment of the present application comprises a monitoring module 201 configured to monitor a current construction site state corresponding to the target construction site, and obtain corresponding disturbance information according to the current construction site state; an input module 202 configured to input the disturbance information into a pre-constructed coupling relationship knowledge graph, to output a corresponding process scheduling scheme and a process adjustment suggestion; and a deduction module 203 configured to perform format conversion on the process scheduling scheme and the process adjustment suggestion, to obtain corresponding target format data, and input the target format data into a pre-trained hierarchical deep reinforcement learning network model, to output a final process scheduling scheme. The present application can realize construction safety, quality and process multi-element coupling evolution mechanism modeling, construct a process scheduling optimization model based on Markov chain, and finally realize digital twin online deduction, thereby assisting construction decision-making.

[0125] Figure 24 The electronic device provided by the embodiment of the present application is shown in the structural schematic diagram. The electronic device can comprise: The memory 2401, the processor 2402, and the computer program stored in the memory 2401 and executable on the processor 2402.

[0126] The processor 2402 executes the program to realize the construction process digital twin online deduction method provided in the above embodiments.

[0127] Further, the electronic device further comprises: The communication interface 2403 is configured to communicate between the memory 2401 and the processor 2402.

[0128] The memory 2401 is configured to store a computer program capable of being executed on the processor 2402.

[0129] The memory 2401 can include a high-speed RAM memory, and can further include a non-volatile memory, for example, at least one disk memory.

[0130] If the memory 2401, the processor 2402 and the communication interface 2403 are independently implemented, the communication interface 2403, the memory 2401 and the processor 2402 can be connected with each other through a bus and complete communication between each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, Figure 24 Only one thick line is used in the figure to represent the bus, but it does not mean that there is only one bus or only one type of bus.

[0131] Optionally, in a specific implementation, if the memory 2401, the processor 2402 and the communication interface 2403 are integrated on a chip, the memory 2401, the processor 2402 and the communication interface 2403 can complete communication between each other through an internal interface.

[0132] The processor 2402 can be a Central Processing Unit (CPU), or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement one or more embodiments of the present application.

[0133] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the program is executed by a processor to implement the construction process digital twin online deduction method.

[0134] In the description of the application, reference to "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that a particular feature, structure, material, or characteristic being described is included in at least one embodiment or example of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment or example. Furthermore, the described specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples. In addition, the usage of "N" means at least two, for example, two, three or the like, unless explicitly stated otherwise.

[0135] Furthermore, the terms "first", "second", or the like, are used merely as a designation of certain elements or features of the application, and do not imply or connote relative importance or a specific order of precedence. Thus, features defined with "first", "second", etc. can include at least one of the features, either explicitly or implicitly.

[0136] Any process or method descriptions or blocks in flow charts or otherwise described herein represent embodiments of modules, segments, or portions of code which include one or more executable instructions for implementing specific logic functions or steps, and alternate implementations are possible. In some embodiments, the processes or methods described in flow charts or otherwise described herein are not necessarily performed in the order shown or discussed, including, for example, performing or depending from other operations or stages, in parallel, in reverse order, or in other orders.

[0137] The logic and / or steps represented in the flowcharts and / or described herein, for example, can be considered as a sequence of executable instructions stored in a computer readable medium, which can be executed by an instruction execution system, apparatus or device, such as a computer-based system, a processor-based system, or other system that can fetch the instructions from the instruction execution system, apparatus or device and execute the instructions, or a combination of the above. For the purposes of this specification, a "computer readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus or device. The computer readable medium can be a computer readable storage medium or a computer readable signal medium. The computer readable storage medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a propagation medium. The computer readable signal medium can include, but is not limited to, a computer readable medium that facilitates transfer of the program from one place to another. A specific example of a computer readable medium is a non-transitory computer-readable storage medium. A specific example of a computer readable signal medium is a source or destination of the computer readable medium. Another specific example of a computer readable signal medium is a computer readable signal travelling through space. Thus, a computer readable medium can take many forms of hardware to carry out the program for use by or in connection with the instruction execution system, apparatus or device.

[0138] It should be understood that aspects of the application can be implemented in hardware, software, firmware or a combination thereof. In the above embodiments, the N steps or methods can be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware and in another embodiment, the hardware can be implemented using any or a combination of the following technologies, which are all well known in the art: a discrete logic circuit having logic gates for implementing logic functions upon data signals, an application specific integrated circuit having appropriate combinational logic gates, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0139] Those of skill in the art would understand that the steps carried out by the above-mentioned embodiments can be implemented by a program instructing the relevant hardware to complete all or part of the steps, and the program can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiments or a combination thereof.

[0140] In addition, each of the functional units in the various embodiments of the present application can be integrated in one processing module, or each of the units can be physically present separately, or two or more units can be integrated in one module. The integrated module can be implemented in the form of hardware or in the form of a software functional module. When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0141] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A construction process digital twin online deduction method, applied in the offline training stage, characterized by: The following steps are involved: Based on the original construction schedule, the preset safety-process coupling knowledge sub-base and the quality-process coupling knowledge sub-base, a safety-quality-process coupling evolution knowledge base is constructed, and a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base is established; Modeling the coupling relationship knowledge graph, determining the corresponding state space, action space, state transition mechanism, and disturbance mechanism, and constructing a multi-objective reward function corresponding to the original construction schedule plan, so as to construct a process scheduling optimization model based on the state space, the action space, the state transition mechanism, the multi-objective reward function, and the disturbance mechanism; A hierarchical deep reinforcement learning network model is determined through the process scheduling optimization model, and the hierarchical deep reinforcement learning network model is used to construct a construction simulation environment. Based on the construction simulation environment, the hierarchical deep reinforcement learning network model is trained, so that in the online deduction stage, the digital twin online deduction of the original construction schedule plan is performed based on the trained hierarchical deep reinforcement learning network model.

2. The construction process digital twin online deduction method according to claim 1 is characterized in that: The safety-quality-process coupling evolution knowledge base is constructed based on the original construction schedule, the preset safety-process coupling knowledge sub-base, and the quality-process coupling knowledge sub-base, and the coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base is established, including: Acquire multiple safety risks and multiple safety accidents in the target construction process, determine the risk level and risk category of each of the multiple safety risks, and determine the corresponding process scheduling impact type based on the risk level and risk category; Based on the multiple safety incidents, establishing an accident type mapping relationship between the risk category and the safety accident type, a level mapping relationship between each safety risk and the risk level, a risk category mapping relationship between each safety risk and the risk category, and a process mapping relationship between each safety risk and the process scheduling impact type, so as to construct the safety-process coupling knowledge sub-base according to the accident type mapping relationship, the level mapping relationship, the risk category mapping relationship, and the process mapping relationship; Acquire multiple quality hazards in the target construction process, quantify the severity of each quality hazard to obtain a corresponding severity level, establish a multiple process interaction impact model based on the severity level, and determine a status indicator corresponding to each quality hazard; Establishing mapping relationships between the status indicators and the quality hazards, the severity levels, and the process interaction impact patterns, respectively, to construct the quality-process coupling knowledge sub-base according to the mapping relationships; Based on the safety-process coupling knowledge subbase and the quality-process coupling knowledge subbase, the safety-quality-process coupling evolution knowledge base is constructed, and through a preset graph database strategy, the safety-process knowledge graph and quality-process knowledge graph corresponding to the safety-process coupling knowledge subbase and the quality-process coupling knowledge subbase are respectively established to construct a corresponding visual coupling relationship knowledge graph based on the safety-process knowledge graph and the quality-process knowledge graph.

3. The construction process digital twin online deduction method according to claim 2 is characterized in that: The coupling relationship knowledge graph is modeled to determine the corresponding state space, action space, state transition mechanism, and disturbance mechanism, and a multi-objective reward function corresponding to the original construction schedule is constructed, so as to construct a process scheduling optimization model based on the state space, the action space, the state transition mechanism, the multi-objective reward function, and the disturbance mechanism, including: Calculating the global environment state and process characteristic state during the construction process corresponding to the original construction schedule, and combining the global environment state and the process characteristic state based on a preset two-layer state representation strategy to determine the state space; Based on a preset binary action representation strategy, the action space is constructed, and according to the construction progress of the original construction schedule, it is determined whether all predecessor processes of the current process are completed. If all the predecessor processes are completed, the state of the current process is updated to determine the state transition mechanism; Acquire disturbance data during the construction process corresponding to the original construction schedule, and generate corresponding processing measures based on the disturbance data and a preset response strategy to determine the disturbance mechanism.

4. The construction process digital twin online deduction method according to claim 3 is characterized in that: The step of determining a hierarchical deep reinforcement learning network model through the process scheduling optimization model, constructing a construction simulation environment using the hierarchical deep reinforcement learning network model, and training the hierarchical deep reinforcement learning network model based on the construction simulation environment includes: Constructing the hierarchical deep reinforcement learning network model based on the pre-built input layer, encoding layer, process attention layer, feature fusion layer, two-stream architecture layer and output layer; Extracting component geometry information and spatial relationships from a preset building information model, and obtaining construction activity data corresponding to the component geometry information and the spatial relationships based on a preset four-dimensional work breakdown structure; Based on the construction activity data, multiple process constraint rules are determined, and a disturbance simulation generation algorithm is constructed according to the multiple process constraint rules and the coupling relationship knowledge graph. The construction simulation environment is constructed through the disturbance simulation generation algorithm, so that the hierarchical deep reinforcement learning network model is trained in the construction simulation environment through preset conventional construction paths and disturbance processing paths.

5. The construction process digital twin online deduction method according to claim 1 is characterized in that: The mathematical expression of the multi-objective reward function is: in, represents the state improvement reward; Indicates the parallel process reward; represents the idle time penalty; Indicates resource overlimit penalty; represents the time delay penalty.

6. A construction process digital twin online deduction method, applied to the online deduction stage, characterized by: The method for online deduction of a digital twin of a construction process applied to an offline training phase according to any one of claims 1 to 5 is adopted, wherein the method comprises the following steps: Monitoring a current construction site state corresponding to a target construction site, and obtaining corresponding disturbance information based on the current construction site state; Input the disturbance information into a pre-built coupling relationship knowledge graph to output corresponding process scheduling solutions and process adjustment suggestions; The process scheduling plan and the process adjustment suggestion are format converted to obtain corresponding target format data, and the target format data is input into a pre-trained hierarchical deep reinforcement learning network model to output the final process scheduling plan.

7. A construction process digital twin online deduction device, used in the offline training stage, characterized by: include: A graph construction module is used to construct a safety-quality-process coupling evolution knowledge base based on the original construction schedule, the preset safety-process coupling knowledge sub-base, and the quality-process coupling knowledge sub-base, and to establish a coupling relationship knowledge graph corresponding to the safety-quality-process coupling evolution knowledge base; a modeling module, configured to model the coupling relationship knowledge graph, determine the corresponding state space, action space, state transition mechanism, and disturbance mechanism, and construct a multi-objective reward function corresponding to the original construction schedule, so as to construct a process scheduling optimization model based on the state space, the action space, the state transition mechanism, the multi-objective reward function, and the disturbance mechanism; The training module is used to determine the hierarchical deep reinforcement learning network model through the process scheduling optimization model, and use the hierarchical deep reinforcement learning network model to build a construction simulation environment, and train the hierarchical deep reinforcement learning network model based on the construction simulation environment, so as to perform digital twin online deduction of the original construction schedule plan based on the trained hierarchical deep reinforcement learning network model in the online deduction stage.

8. A construction process digital twin online deduction device, used in the online deduction stage, characterized by: The construction process digital twin online deduction device for offline training according to claim 7 is used, wherein the device comprises: A monitoring module, configured to monitor the current construction site status corresponding to the target construction site and obtain corresponding disturbance information based on the current construction site status; An input module, configured to input the disturbance information into a pre-built coupling relationship knowledge graph to output corresponding process scheduling solutions and process adjustment suggestions; The deduction module is used to convert the format of the process scheduling plan and the process adjustment suggestion to obtain corresponding target format data, and input the target format data into a pre-trained hierarchical deep reinforcement learning network model to output the final process scheduling plan.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the online deduction method for digital twins of a construction process as described in any one of claims 1 to 5 or claim 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the online deduction method of digital twin of the construction process as described in any one of claims 1 to 5 or claim 6.

Images