A BIM-based construction project construction optimization management method and management platform

Abstract

The application provides a kind of BIM-based construction engineering construction optimization management method and management platform, belongs to the technical field of construction engineering management, the application provides dynamic digital base for collaborative simulation optimization by constructing the construction digital twin model updated synchronously with physical site, multiple decision unit simulation is carried out in digital twin environment using unified training, independent execution collaborative strategy learning framework, combined with knowledge graph compliance filtering and risk feature integration, automatically solve the construction process scheme located in the non-dominated optimal solution set, finally through scheme automatic analysis push and feedback back writing, drive BIM model and decision model continue adaptive evolution. Through the deep integration of construction digital twin model and multi-objective collaborative simulation optimization, the automatic generation and continuous evolution of construction scheme from experience dependence to data-driven are realized, and the construction period is shortened and the cost is reduced under the premise of ensuring construction safety compliance.

Description

Technical Field

[0001] This invention relates to the field of building engineering management technology, specifically to a BIM-based method and platform for optimizing the management of building engineering construction. Background Technology

[0002] With the increasing scale and complexity of construction projects, traditional construction management methods are no longer sufficient to meet the needs of modern construction engineering. Current BIM construction management primarily relies on manual experience to decompose tasks and arrange processes using the model; the BIM model itself serves only as a static display tool and cannot reflect the real-time construction status on site. Meanwhile, the lack of a dynamic mapping mechanism between the heterogeneous data from multiple sources on the construction site, such as progress, quality, cost, safety, and carbon emissions, and the BIM model leads to construction decisions lagging behind on-site changes. Furthermore, the independent operations of various construction teams and equipment in work processes and resource allocation, coupled with a lack of collaborative optimization methods, make it difficult to make a holistic balance among multiple objectives such as schedule, quality, cost, and carbon emissions. This results in crude construction plans, resource waste, and frequent safety hazards. The main problem lies in the lack of real-time mapping and collaborative optimization decision-making capabilities between existing BIM models and dynamic data from the construction site. It cannot automatically generate optimal construction plans that take into account schedule, cost, quality, and carbon emissions. Therefore, a BIM-based construction optimization management method and platform for building engineering is proposed. Summary of the Invention

[0003] To address the aforementioned technical problems, this invention provides a BIM-based construction engineering optimization management method and platform. Through the deep integration of construction digital twin models and multi-objective collaborative simulation optimization, it enables the automatic generation and continuous evolution of construction plans from experience-dependent to data-driven, thereby shortening the construction period and reducing costs while ensuring construction safety and compliance.

[0004] The technical solution adopted in this invention is as follows: A BIM-based construction optimization management method for building projects includes the following steps: Step S1: Obtain the BIM model for the design phase, parse the IFC attributes of each component in the BIM model, extract geometric dimensions, spatial coordinates and material information, bind the corresponding construction task, required resources, process logic constraints and standard time consumption to each component instance; at the same time, analyze the spatial topological relationship between components, generate spatial conflict constraints, merge with process logic to form a task network diagram with spatial-logical joint constraints, and attach the task network diagram to the BIM model; Step S2: The raw data collected by the heterogeneous sensing devices on site is uploaded after being denoised by edge nodes and time aligned. The computer uses a component ID matching algorithm to map the processed component status, location and environmental data to the corresponding components in the BIM model, and dynamically updates the component's progress attributes, installation deviation and real-time status labels, thereby generating a construction digital twin model in the computer that is updated synchronously with the physical construction site. Step S3: Based on the completion rate and resource consumption data of each component in the construction digital twin model, construct the schedule calculation function, cost calculation function, quality defect risk equivalent calculation function, and construction carbon emission calculation function respectively; use the membership function to map the four heterogeneous index values ​​to dimensionless satisfaction values, and sum them according to the preset construction stage weight coefficients to form a comprehensive optimization objective; at the same time, extract the resource upper limit and process logic from the task network diagram as hard constraints to form a constrained multi-objective optimization function. Step S4: Taking each construction team or large equipment as an independent decision-making unit, and the construction digital twin model as the digital environment for collaborative simulation of each decision-making unit, the constraints defined by the task network graph are encoded as state update rules of the digital environment, and the multi-objective optimization function is used as the comprehensive evaluation value; using a unified training and independent execution collaborative strategy learning method, the construction process is simulated in the digital environment. Each decision-making unit outputs candidate procedures, resource allocation, and process selection actions, and before the action output, a pre-constructed construction knowledge graph is called for compliance filtering. At the same time, the potential risk features obtained by graph reasoning are incorporated into the input state parameter set of each decision-making unit; through simulation iteration, a set of construction process schemes located in the non-dominated optimal solution set is obtained from the digital environment. The construction process scheme includes optimized procedure sequences, resource allocation, and process parameters. Step S5: Automatically parse the selected construction process scheme into component-level task instructions and push them to the execution terminal. At the same time, write back the process, resource and process data in the construction process scheme to the corresponding component attributes of the BIM model. Receive execution feedback data and update the construction digital twin model and the action decision model of each decision unit again to form a continuous adaptive optimization closed loop of the BIM model.

[0005] Further, in step S1, the analysis of the spatial topological relationship between components and the generation of spatial conflict constraints specifically involves: by calculating the spatial positional relationship of the bounding boxes of each component, identifying possible hoisting overlaps, path collisions, and temporary support dependencies, generating hard constraints that prohibit simultaneous construction, and merging them with the process logic constraints into the task network diagram.

[0006] Further, in step S2, the dynamic updating of the component's progress attributes, installation deviation, and real-time status label specifically involves: the cloud-based BIM engine receiving the edge-preprocessed status event stream, mapping the hoisting and positioning signal to the corresponding component's progress percentage based on the component ID, calculating the positioning trajectory data as an installation deviation vector, and updating the component status label, thereby obtaining the construction digital twin model.

[0007] Further, in step S3, the step of mapping the four heterogeneous index values ​​to dimensionless satisfaction values ​​using membership functions and then summing them according to preset construction stage weight coefficients is as follows: linear or nonlinear membership functions are constructed for construction period, cost, quality defect risk equivalent, and carbon emissions respectively, and each index value is normalized to the [0, 1] interval. Then, the weighted summation is performed according to the weight vectors set for each construction stage—foundation, main structure, and decoration—to obtain the comprehensive optimization target.

[0008] Further, in step S4, the compliance filtering by calling the pre-built construction knowledge graph specifically involves: converting the current state snapshot of the construction digital twin model and the candidate actions of each decision unit into SPARQL query statements, performing pattern matching in a graph database storing construction accident cases, standard clauses, and component failure modes; if a violation pattern is matched, the candidate action is removed from the candidate action set or a penalty value is applied; the potential risk features include the component failure links and risk propagation intensity obtained by graph reasoning, which are encoded into fixed-dimensional real number vectors and then concatenated into the input state parameter set of each decision unit.

[0009] Furthermore, in step S4, the collaborative strategy learning method using unified training and independent execution includes: in the offline stage, using the construction digital twin model to generate a simulation scenario containing random disturbances and abnormal events at high speed, and training all decision units centrally to obtain an initial action decision model; in the online stage, each decision unit loads the initial action decision model, independently executes action selection according to the state distribution of the real-time construction digital twin model, and combines knowledge graph filtering to output the construction process scheme.

[0010] This invention also proposes a BIM-based construction optimization management platform, comprising: The BIM model management module is used to acquire the design BIM model, bind construction tasks, resources and process logic to components, analyze spatial topology and generate a task network diagram with spatial-logical joint constraints, and attach the task network diagram to the BIM model. The IoT data platform is used to access and preprocess raw data from heterogeneous sensing devices in the field, and output standardized component status events. The digital twin engine receives the component status events, dynamically updates the component progress and attributes of the internal BIM model through a component ID matching algorithm, generates a construction digital twin model, and provides model snapshot and status rollback functions for simulation. The multi-objective optimization configuration module, based on the construction digital twin model and task network diagram, constructs membership functions for construction period, cost, quality defect risk equivalent, and carbon emissions, and dynamically weights them to form a constrained multi-objective optimization function; The construction process simulation optimization module takes each construction team or equipment as an independent decision-making unit, uses the construction digital twin model as the simulation digital environment, loads the task network graph as the state update rule, uses the multi-objective optimization function as the comprehensive evaluation value, executes a collaborative strategy learning with unified training and independent execution, and calls the knowledge graph module to perform compliance filtering on candidate actions during the simulation process, and finally solves the non-dominated optimal construction process scheme. The knowledge graph module uses a graph database to store construction accident cases, standard clauses, and component failure modes. It has a built-in inference engine to receive simulation states and candidate actions, and output violation filtering signals and potential risk feature vectors. The scheme write-back and push module is used to parse the construction process scheme into component-level task instructions and push them to the execution terminal. At the same time, it writes back the procedures, resources and process parameters in the scheme to the corresponding component attributes of the BIM model and collects execution feedback to drive the digital twin model to update. The platform is deployed in a cloud-edge collaborative architecture, with edge nodes responsible for preprocessing perception data and issuing instructions, and cloud servers responsible for BIM model updates, digital twin simulation, and training and inference of simulation optimization modules.

[0011] Furthermore, the digital twin engine includes a cloud-based BIM engine and an edge computing unit; the edge computing unit performs noise reduction, alignment, and event extraction on the perceived data, while the cloud-based BIM engine maintains the construction digital twin model, uses component IDs to map events to incremental updates of component attributes, and supports snapshot storage and version rollback of the model.

[0012] Furthermore, the construction process simulation optimization module includes a simulation environment unit, an offline training unit, and an online inference unit; the simulation environment unit loads a snapshot of the construction digital twin model and controls state updates based on the task network graph; the offline training unit uses the simulation environment unit to generate a construction scenario with random disturbances and uniformly trains the action decision models of all decision units; the online inference unit calls the trained action decision model based on the real-time construction digital twin model state, and after filtering by the knowledge graph module, generates the construction process scheme.

[0013] Furthermore, the knowledge graph module uses a graph database and performs safety and compliance queries on the candidate actions passed in by the construction process simulation optimization module through the SPARQL inference engine, returning violation markers and potential risk feature vectors containing component failure links and risk propagation intensity. The potential risk feature vectors are merged into the input state parameter set of the corresponding decision unit.

[0014] The beneficial effects of this invention are: This invention utilizes edge-sensing devices and component ID matching algorithms to map real-time progress, deviation, and environmental data from the construction site onto a BIM model, constructing a construction digital twin model that is synchronously updated with the physical construction site. This provides a dynamic digital foundation for collaborative simulation optimization. Based on this, it constructs and dynamically weights membership functions for four objectives: construction period, cost, quality defect risk equivalent, and carbon emissions. A unified training and independently executed collaborative strategy learning framework is used to perform multi-decision unit simulations within the digital twin environment. Combined with knowledge graph compliance filtering and risk feature integration, it automatically solves for construction process schemes located in the non-dominated optimal solution set. Finally, through automatic scheme parsing and feedback, it drives the continuous adaptive evolution of the BIM model and decision-making model. Through the deep integration of the construction digital twin model and multi-objective collaborative simulation optimization, it achieves the automatic generation and continuous evolution of construction schemes from experience-dependent to data-driven, shortening the construction period and reducing costs while ensuring construction safety and compliance. Attached Figure Description

[0015] Figure 1 This is a flowchart of the construction optimization management method for building engineering according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating the simulation decision-making process in one embodiment of the present invention; Figure 3 This is an adaptive closed-loop flowchart of an embodiment of the present invention; Figure 4 This is a schematic diagram of the cloud-edge collaborative architecture of the construction engineering optimization management platform of the present invention. Figure 5 This is a schematic diagram of the linkage block of the core module in an embodiment of the present invention. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1 like Figures 1-3As shown in the figure, a BIM-based construction optimization management method for building projects according to an embodiment of the present invention includes the following steps: Step S1: Obtain the BIM model in the design phase, parse the IFC attributes of each component in the BIM model, extract geometric dimensions, spatial coordinates and material information, bind the corresponding construction task, required resources, process logic constraints and standard time consumption to each component instance; at the same time, analyze the spatial topology relationship between components, generate spatial conflict constraints, merge with process logic to form a task network diagram with spatial-logical joint constraints, and attach the task network diagram to the BIM model; Step S2: The raw data collected by the heterogeneous sensing devices on site is uploaded after noise reduction and time alignment at the edge nodes. The computer uses a component ID matching algorithm to map the processed component status, location and environmental data to the corresponding components in the BIM model, dynamically updating the component's progress attributes, installation deviation and real-time status labels, thereby generating a construction digital twin model in the computer that is updated synchronously with the physical construction site. Step S3: Based on the completion rate and resource consumption data of each component in the construction digital twin model, construct the schedule calculation function, cost calculation function, quality defect risk equivalent calculation function, and construction carbon emission calculation function respectively; use the membership function to map the four heterogeneous index values ​​to dimensionless satisfaction values, and sum them according to the preset construction stage weight coefficients to form a comprehensive optimization objective; at the same time, extract the resource upper limit and process logic from the task network diagram as hard constraints to form a constrained multi-objective optimization function. Step S4: Taking each construction team or large equipment as an independent decision-making unit, and the construction digital twin model as the digital environment for collaborative simulation of each decision-making unit, the constraints defined by the task network graph are encoded as state update rules of the digital environment, and the multi-objective optimization function is used as the comprehensive evaluation value; using a unified training and independent execution collaborative strategy learning method, the construction process is simulated in the digital environment. Each decision-making unit outputs candidate procedures, resource allocation, and process selection actions, and before the action output, a pre-built construction knowledge graph is called for compliance filtering. At the same time, the potential risk features obtained by graph reasoning are incorporated into the input state parameter set of each decision-making unit; through simulation iteration, a set of construction process schemes located in the non-dominated optimal solution set is obtained from the digital environment. The construction process scheme includes optimized procedure sequences, resource allocation, and process parameters. The encoding of state update rules employs an action masking mechanism. Specifically, at each decision step, the simulation environment reads a snapshot of the current state of the construction digital twin model and extracts the currently effective process logic constraints and spatial conflict constraints from the task network diagram. Based on these constraints, the system generates a masking flag that corresponds one-to-one with the candidate action list.

[0018] If a candidate action triggers any constraint, the flag corresponding to that action is marked as prohibited. Before outputting the final action probability, the policy network of the decision-making unit forces the output probability of prohibited actions to zero based on the masking flag, thus completely excluding them from action selection. Through this mechanism, the agent can only choose from allowed actions that meet both spatial and logical constraints from the beginning.

[0019] When spatial conflict constraints conflict with process logic constraints, the system handles the situation according to the following rules: First, spatial safety has the highest priority, and spatial collisions and hoisting interference must be prevented. Second, if a task cannot start on time as required by the process due to spatial constraints, the system automatically postpones the start time of the task and simultaneously postpones all subsequent tasks that depend on it. Finally, if the above postponement causes the total project duration to exceed the preset warning value, the system outputs a conflict alarm to the management personnel, prompting manual intervention to adjust the spatial construction plan or process logic.

[0020] Step S5: Automatically parse the selected construction process scheme into component-level task instructions and push them to the execution terminal. At the same time, write back the process, resource and process data in the construction process scheme to the corresponding component attributes in the BIM model. Receive execution feedback data and update the construction digital twin model and the action decision model of each decision unit again to form a continuous adaptive optimization closed loop of the BIM model.

[0021] The computer first acquires the BIM model delivered by the design, parses the IFC data to extract the geometric dimensions, spatial coordinates and material properties of each component, and binds the construction task, required machinery and team resources, process logic constraints and standard time consumption to each component instance.

[0022] Simultaneously, the computer analyzes the spatial topological relationships between components, identifies potential hoisting overlaps, path collisions, and temporary support dependencies during construction by calculating the spatial positional relationships of bounding boxes, and automatically generates spatial conflict constraints. These constraints are then merged with process logic constraints to form a task network diagram with combined spatial and logical constraints. The task network diagram is essentially a structured set of constraints that specifies which tasks must be completed before starting and which tasks cannot be executed simultaneously. This set of constraints is added to the BIM model, becoming the foundational constraint set for all subsequent simulation calculations.

[0023] The construction site deploys heterogeneous sensing devices such as UWB positioning base stations, tower crane monitoring sensors, and environmental sensors. Edge computing nodes perform timestamp alignment, noise filtering, and event slicing on the raw data stream to form standardized component status events. The cloud-based BIM engine maps the status data in the events to the corresponding components in the BIM model through a component ID matching algorithm.

[0024] For any component Let its design coordinates be... The actual positioning coordinates returned by the sensing system are The computer calculates the installation deviation vector. This deviation vector reflects both the geometric accuracy of component installation and serves as input for subsequent quality risk calculations. Simultaneously, progress attributes and status labels are dynamically updated based on placement signals and continuous operation time, thereby generating a construction digital twin model in computer memory that is synchronized with the physical construction site.

[0025] Based on the completion status and resource consumption data of each component in the digital twin model, this invention constructs a unified multi-objective optimization function to evaluate the construction status and decision-making at any given time. Specifically, it establishes four independent objective functions: construction period, cost, equivalent risk of quality defects, and carbon emissions from construction. The equivalent risk of quality defects is expressed as the modulus of component installation deviation. With allowable threshold The calculation is based on parameters such as the ratio of carbon emissions to mechanical usage time and material loss.

[0026] To integrate four indicators with different dimensions into a single comprehensive optimization objective, a membership function is used for dimensionless transformation: for each indicator... Set the minimum acceptable value and the highest value Construct linear membership functions when , when when This maps all indicators to the [0, 1] interval, with larger values ​​indicating better performance. A dynamic weight vector is then introduced. The sum of the weights is 1, and they are adjustable according to the construction stage. Therefore, the overall optimization objective is:

[0027] In the formula, These represent the membership values ​​for construction period, cost, quality defect risk equivalent, and carbon emissions, respectively. This function... This will serve as the basis for evaluating each decision in subsequent simulations. Meanwhile, the resource ceiling, process precedence relationships, and prohibited parallel sets extracted from the task network graph will be used as hard constraints to form a complete constrained multi-objective optimization function.

[0028] Next, each construction team or large piece of equipment is treated as an independent decision-making unit, and the construction digital twin model serves as a simulated digital environment for their interaction. The state of the environment is defined by a snapshot of the digital twin model, and the constraints of the task network graph are encoded as state update rules to prevent any actions that violate process logic or spatial conflicts.

[0029] Each time the decision-making unit outputs a legal action and executes it, the environment is updated to the new state, and an immediate evaluation value is generated. This evaluation value is defined as the increment of the comprehensive optimization objective.

[0030] and These are snapshots of the state of the digital twin model before and after the action is executed. This is the comprehensive optimization objective function constructed in the previous step. This approach directly links the effect of each decision step to the overall improvement in schedule, cost, quality, and carbon emissions.

[0031] Under this evaluation system, a collaborative strategy learning framework with unified training and independent execution is adopted: In the offline stage, a large number of scenarios with random disturbances are generated by high-speed simulation using digital twin models, allowing each decision-making unit to learn how to collaboratively select actions to maximize long-term cumulative rewards under centralized guidance; In the online stage, each decision-making unit loads the trained action decision model and independently generates candidate actions based on local observations from the real-time construction digital twin model.

[0032] Incremental reward value As a global reward signal, it is suitable for mainstream multi-agent reinforcement learning algorithms under the centralized training-distributed execution (CTDE) framework. During the training phase, all decision-making units share this global reward signal to carry out joint policy optimization; during the execution phase, each decision-making unit makes decisions based on its own local observations, without relying on global information.

[0033] Before the final output of the action, the system calls a pre-built construction knowledge graph for compliance filtering, converting the current state and candidate actions into graph queries. When a violation pattern is matched, the action is directly eliminated. At the same time, potential risk features such as component failure chains and risk propagation intensity obtained from graph inference are incorporated into the observations of the decision-making unit, further improving decision security. After iterative solving, the computer obtains a set of construction process schemes located in the non-dominated optimal solution set. Each scheme includes optimized process sequences, resource allocation, and process parameters.

[0034] In addition to the linear membership function mentioned above, when the sensitivity of the indicator value is high near the acceptable boundary, the method also allows the use of an S-shaped nonlinear membership function as an alternative. The steepness coefficient of this nonlinear function can be preset by the administrator in the platform interface according to the sensitivity of the indicator boundary value in the actual project. The larger the steepness coefficient, the faster the membership degree decreases or increases near the boundary, thus more sensitively reflecting the risk of the indicator value approaching the unacceptable range.

[0035] After a scheme is selected, the computer parses it into component-level task instructions and pushes them to the execution terminal. At the same time, the planned time of the process, resource allocation and process parameters in the scheme are written back to the attributes of the corresponding components in the BIM model, so that the BIM model is upgraded into a dynamic model with embedded optimized construction information.

[0036] Example 2 like Figure 4 and Figure 5 As shown, this invention also provides a BIM-based construction engineering optimization management platform. The platform is deployed using a cloud-edge collaborative architecture and is used to execute all the steps in the aforementioned method embodiments. The platform as a whole consists of a BIM model management module, an IoT data platform, a digital twin engine, a multi-objective optimization configuration module, a construction process simulation optimization module, a knowledge graph module, and a scheme write-back and push module.

[0037] The BIM model management module is deployed on a cloud server, embedding an IFC parser and a spatial topology analysis engine. This module receives BIM model files delivered during the design phase, extracts the geometric dimensions, spatial coordinates, and material properties of each component, and automatically binds construction tasks, required machinery and team resources, process logic constraints, and standard time consumption to each component instance through manual assistance or a rule base. Subsequently, the spatial topology analysis engine traverses all component bounding boxes, calculates the spatial relationships between components, identifies potential conflicts such as intersecting hoisting paths and support dependencies, generates spatial conflict constraints, and merges them with process logic to form a task network diagram with combined spatial and logical constraints. This task network diagram is stored in a graph data structure and attached to the BIM model for use by the digital twin engine and simulation optimization module.

[0038] The IoT data platform, acting as a data bridge between the edge and the cloud, is deployed at edge computing nodes and the cloud access layer on the construction site. The platform connects to heterogeneous sensing devices such as UWB positioning base stations, tower crane slewing / amplitude sensors, anemometers, and temperature and humidity sensors. It performs unified timestamp alignment, noise filtering, and event slicing on the raw data stream, transforming fragmented sensor signals into standardized component status events. Examples include the component's hoisting status, actual coordinate information, and the tower crane's current slewing angle and amplitude. The pre-processed event stream is then pushed in real-time to the cloud-based digital twin engine via a message queue.

[0039] The digital twin engine comprises two parts: a cloud-based BIM engine and an edge computing unit. The edge computing unit handles the preprocessing of perception data, while the cloud-based BIM engine maintains the digital twin model centered on the BIM model. Upon receiving component status events, the cloud-based BIM engine locates the corresponding component instance in the BIM model using a component ID matching algorithm. It then dynamically updates the component's progress attributes, installation deviation vector, and real-time status label using the status data carried in the event. The installation deviation vector, calculated from the difference between the component's actual positioning coordinates and its design coordinates, is persistently recorded for quality risk assessment. To support the high-speed simulation requirements of the simulation optimization module, the digital twin engine features model snapshot and state rollback capabilities: it quickly saves a copy of the current state of all components before simulation begins, and allows users to revert to the snapshot state or submit the state changes resulting from the simulation after simulation. Through these mechanisms, the digital twin engine maintains a construction digital twin model that is synchronized with the physical construction site and can be simulated.

[0040] The multi-objective optimization configuration module provides a user interface that allows project managers to set acceptable minimum and maximum values ​​for four objectives: schedule, cost, quality defect risk equivalent, and construction carbon emissions, as well as weighting coefficients for each construction stage. The configuration module obtains real-time completion and resource consumption data for each component from the digital twin engine, automatically establishes membership functions, renders the four heterogeneous indicators dimensionless and performs a weighted summation to generate a comprehensive optimization objective F. Simultaneously, it reads resource limits and process logic from the task network diagram as hard constraints, forming a constrained multi-objective optimization function. This function is provided as a service interface to the construction process simulation optimization module for invocation.

[0041] The construction process simulation and optimization module is the core computing module of the platform, deployed on a cloud-based server with GPU acceleration capabilities. Internally, it is divided into a simulation environment unit, an offline training unit, and an online inference unit. The simulation environment unit directly loads model snapshots provided by the digital twin engine and task network diagrams generated by the BIM model management module. It encodes the predecessor-successor relationships and spatial conflict constraints in the task network diagram into state update rules, thereby constructing a multi-party interactive simulation environment with each construction team and equipment as independent decision-making units. The offline training unit generates a large number of virtual construction scenarios with random disturbances (such as equipment failure, material delays, and sudden weather changes) through high-speed simulation within this environment. It performs centralized offline pre-training of the action decision models for all decision units to obtain initial decision models. The online inference unit is deployed on an inference server close to the production environment. It pulls real-time model states from the digital twin engine at fixed time steps or through an event-triggered mechanism. Each decision unit loads its trained action decision model and independently outputs candidate actions based on its local observations. Before outputting the final action, the reasoning unit initiates a compliance query to the knowledge graph module. Based on the returned filtering results, it prunes or penalizes the candidate actions. After multiple iterations, it generates a set of construction process schemes located in the non-dominated optimal solution set. Each scheme includes optimized process scheduling, resource allocation, and process method selection.

[0042] The knowledge graph module uses a graph database to store construction accident cases, code provisions, and component failure modes, and incorporates a built-in SPARQL inference engine. When it receives the current digital twin state snapshot and candidate action list from the simulation optimization module, the module automatically converts the state and action into RDF triples. It then uses SPARQL queries to match violation patterns in the knowledge graph. If violations such as "work at height without protection" or "lifting in winds exceeding level six" are matched, a filter flag for that action is returned. Simultaneously, the inference engine, starting from the current state, performs multi-step reasoning along the mechanical dependencies and technological relationships between components, identifying possible failure chains and risk propagation strengths. This potential risk information is encoded into a fixed-length potential risk feature vector, which is returned to the simulation optimization module. The simulation optimization module then concatenates this vector into the input state parameter set of the corresponding decision-making unit to enhance decision-making safety.

[0043] The encoding is a fixed-dimensional real-number vector, implemented using graph embedding. Specifically, taking the component nodes and their directed connections in the component failure chain output by the knowledge graph inference as input, the type attribute, current stress state label, and spatial coordinates of each component are extracted as initial features. Through multi-layer neighborhood aggregation operations of the graph embedding network, the features of each node are fused along the propagation direction of the failure chain. Then, mean pooling is performed on all the fused node features to output a fixed-dimensional real-number vector. When the knowledge graph inference does not output a failure chain, this risk feature vector is set to a zero vector, with the dimension remaining consistent with when there is output.

[0044] The fixed-dimensional real vector obtained after the above encoding is directly appended to the end of the original input state parameters of the decision unit, forming a new input state with expanded dimensions. The input dimension of the first fully connected layer of the policy network of the decision unit is expanded accordingly, while the content and structure of the original state parameters remain unchanged.

[0045] In this invention, a component failure link refers to a set of directed paths for risk transmission, which are derived from a component in the current construction state, based on the mechanical transmission relationship, process dependency relationship and spatial adjacency relationship between components stored in the knowledge graph. The risk propagation intensity refers to the cumulative measure of risk propagation from the source component to the end component along the failure link. The higher the value, the greater the impact range after the source component has a problem.

[0046] During compliance filtering, the system converts the current state snapshot and each candidate action output by the decision unit into a SPARQL query. If the query result matches a violation pattern, the system directly removes the candidate action from the action space; this action is not included in the calculation of the probability distribution output by the policy network, nor in gradient updates. If no violation pattern is matched, the candidate action is retained. This hard-removal mechanism ensures that the agent cannot select any non-compliant actions during the learning process.

[0047] The scheme write-back and push module is responsible for converting the construction process schemes generated by the simulation optimization module into executable on-site instructions and model update data. The module parses the planned start and end times, allocated resources, and process parameters in the scheme into task instructions for team leaders or equipment controllers, and pushes them to the corresponding mobile terminals or equipment controllers via wireless network. Simultaneously, the module writes the above data into the BIM model maintained by the BIM model management module, updating the construction attribute fields of the corresponding components, thus upgrading the BIM model from a simple design model to a dynamic construction information model containing optimized construction process information. Actual progress, deviations, and abnormal events generated during on-site execution are collected and uploaded again via the IoT data platform. The digital twin engine updates the model state accordingly, and the simulation optimization module stores the newly generated state transition data in the experience playback pool for online incremental training of the action decision model, achieving adaptive evolution of the strategy.

[0048] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0049] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0050] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0051] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0052] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0053] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A BIM-based method for optimizing construction management, characterized in that, Includes the following steps: Step S1: Obtain the BIM model for the design phase, parse the IFC attributes of each component in the BIM model, extract geometric dimensions, spatial coordinates and material information, bind the corresponding construction task, required resources, process logic constraints and standard time consumption to each component instance; at the same time, analyze the spatial topological relationship between components, generate spatial conflict constraints, merge with process logic to form a task network diagram with spatial-logical joint constraints, and attach the task network diagram to the BIM model; Step S2: The raw data collected by the heterogeneous sensing devices on site is uploaded after being denoised by edge nodes and time aligned. The computer uses a component ID matching algorithm to map the processed component status, location and environmental data to the corresponding components in the BIM model, and dynamically updates the component's progress attributes, installation deviation and real-time status labels, thereby generating a construction digital twin model in the computer that is updated synchronously with the physical construction site. Step S3: Based on the completion rate and resource consumption data of each component in the construction digital twin model, construct the schedule calculation function, cost calculation function, quality defect risk equivalent calculation function, and construction carbon emission calculation function respectively; use the membership function to map the four heterogeneous index values ​​to dimensionless satisfaction values, and sum them according to the preset construction stage weight coefficients to form a comprehensive optimization objective; at the same time, extract the resource upper limit and process logic from the task network diagram as hard constraints to form a constrained multi-objective optimization function. Step S4: Taking each construction team or large equipment as an independent decision-making unit, and the construction digital twin model as the digital environment for collaborative simulation of each decision-making unit, the constraints defined by the task network graph are encoded as state update rules of the digital environment, and the multi-objective optimization function is used as the comprehensive evaluation value; using a unified training and independent execution collaborative strategy learning method, the construction process is simulated in the digital environment. Each decision-making unit outputs candidate procedures, resource allocation, and process selection actions, and before the action output, a pre-constructed construction knowledge graph is called for compliance filtering. At the same time, the potential risk features obtained by graph reasoning are incorporated into the input state parameter set of each decision-making unit; through simulation iteration, a set of construction process schemes located in the non-dominated optimal solution set is obtained from the digital environment. The construction process scheme includes optimized procedure sequences, resource allocation, and process parameters. Step S5: Automatically parse the selected construction process scheme into component-level task instructions and push them to the execution terminal. At the same time, write back the process, resource and process data in the construction process scheme to the corresponding component attributes of the BIM model. Receive execution feedback data and update the construction digital twin model and the action decision model of each decision unit again to form a continuous adaptive optimization closed loop of the BIM model.

2. The BIM-based construction optimization management method for building projects according to claim 1, characterized in that, In step S1, the analysis of the spatial topological relationship between components and the generation of spatial conflict constraints are specifically as follows: by calculating the spatial positional relationship of the bounding boxes of each component, identifying possible hoisting overlaps, path collisions and temporary support dependencies, generating hard constraints that prohibit simultaneous construction, and merging them with the process logic constraints into the task network diagram.

3. The BIM-based construction optimization management method for building projects according to claim 1, characterized in that, In step S2, the dynamic updating of the component's progress attributes, installation deviation, and real-time status label specifically involves: the cloud-based BIM engine receiving the edge-preprocessed status event stream, mapping the hoisting and positioning signal to the corresponding component's progress percentage based on the component ID, calculating the positioning trajectory data as an installation deviation vector, and updating the component status label to obtain the construction digital twin model.

4. The BIM-based construction optimization management method for building projects according to claim 1, characterized in that, In step S3, the four heterogeneous index values ​​are mapped to dimensionless satisfaction values ​​using membership functions, and then weighted and summed according to preset construction stage weight coefficients. Specifically, linear or nonlinear membership functions are constructed for construction period, cost, quality defect risk equivalent, and carbon emissions, and each index value is normalized to the [0, 1] interval. Then, the weighted summation is performed according to the weight vectors set for each construction stage—foundation, main structure, and decoration—to obtain the comprehensive optimization target.

5. The BIM-based construction optimization management method for building projects according to claim 1, characterized in that, In step S4, the compliance filtering by calling the pre-built construction knowledge graph specifically involves: converting the current state snapshot of the construction digital twin model and the candidate actions of each decision unit into SPARQL query statements, performing pattern matching in a graph database storing construction accident cases, standard clauses, and component failure modes; if a violation pattern is matched, the candidate action is removed from the candidate action set or a penalty value is applied; the potential risk features include the component failure links and risk propagation intensity obtained by graph reasoning, which are encoded into fixed-dimensional real number vectors and then concatenated into the input state parameter set of each decision unit.

6. The BIM-based construction optimization management method for building projects according to claim 1, characterized in that, In step S4, the collaborative strategy learning method using unified training and independent execution includes: in the offline stage, using the construction digital twin model to generate a simulation scenario containing random disturbances and abnormal events at high speed, and training all decision units centrally to obtain an initial action decision model; in the online stage, each decision unit loads the initial action decision model, independently executes action selection according to the state distribution of the real-time construction digital twin model, and combines knowledge graph filtering to output the construction process scheme.

7. A BIM-based construction optimization management platform for building projects, used to implement the method as described in any one of claims 1 to 6, characterized in that, The management platform includes: The BIM model management module is used to acquire the design BIM model, bind construction tasks, resources and process logic to components, analyze spatial topology and generate a task network diagram with spatial-logical joint constraints, and attach the task network diagram to the BIM model. The IoT data platform is used to access and preprocess raw data from heterogeneous sensing devices in the field, and output standardized component status events. The digital twin engine receives the component status events, dynamically updates the component progress and attributes of the internal BIM model through a component ID matching algorithm, generates a construction digital twin model, and provides model snapshot and status rollback functions for simulation. The multi-objective optimization configuration module, based on the construction digital twin model and task network diagram, constructs membership functions for construction period, cost, quality defect risk equivalent, and carbon emissions, and dynamically weights them to form a constrained multi-objective optimization function; The construction process simulation optimization module takes each construction team or equipment as an independent decision-making unit, uses the construction digital twin model as the simulation digital environment, loads the task network graph as the state update rule, uses the multi-objective optimization function as the comprehensive evaluation value, executes a collaborative strategy learning with unified training and independent execution, and calls the knowledge graph module to perform compliance filtering on candidate actions during the simulation process, and finally solves the non-dominated optimal construction process scheme. The knowledge graph module uses a graph database to store construction accident cases, standard clauses, and component failure modes. It has a built-in inference engine to receive simulation states and candidate actions, and output violation filtering signals and potential risk feature vectors. The scheme write-back and push module is used to parse the construction process scheme into component-level task instructions and push them to the execution terminal. At the same time, it writes back the procedures, resources and process parameters in the scheme to the corresponding component attributes of the BIM model and collects execution feedback to drive the digital twin model to update. The platform is deployed in a cloud-edge collaborative architecture, with edge nodes responsible for preprocessing perception data and issuing instructions, and cloud servers responsible for BIM model updates, digital twin simulation, and training and inference of simulation optimization modules.

8. The BIM-based construction optimization management platform for building projects according to claim 7, characterized in that, The digital twin engine includes a cloud-based BIM engine and an edge computing unit. The edge computing unit performs noise reduction, alignment, and event extraction on the perceived data. The cloud-based BIM engine maintains the construction digital twin model, uses component IDs to map events to incremental updates of component attributes, and supports snapshot storage and version rollback of the model.

9. The BIM-based construction optimization management platform for building projects according to claim 7, characterized in that, The construction process simulation optimization module includes a simulation environment unit, an offline training unit, and an online inference unit. The simulation environment unit loads a snapshot of the construction digital twin model and controls the state update according to the task network graph. The offline training unit uses the simulation environment unit to generate a construction scenario with random disturbances and uniformly trains the action decision models of all decision units. The online inference unit calls the trained action decision model according to the real-time construction digital twin model state, and after filtering by the knowledge graph module, generates the construction process scheme.

10. The BIM-based construction optimization management platform for building projects according to claim 7, characterized in that, The knowledge graph module uses a graph database and performs safety and compliance queries on the candidate actions passed from the construction process simulation and optimization module through the SPARQL inference engine. It returns violation markers and potential risk feature vectors containing component failure links and risk propagation intensity. The potential risk feature vectors are merged into the input state parameter set of the corresponding decision unit.

Images