Multi-agent construction platform building method

Abstract

The invention provides a multi-agent construction platform building method, and belongs to the technical field of construction platforms, and the method comprises the steps (1)-(11): (1) obtaining a construction platform parameterized model through a BIM system, and building a dynamically updated three-dimensional coordinate system; according to the method, efficient cooperation of multiple agents in a complex construction scene can be realized, the dynamic environment adaptability and the task execution coordination capability are improved, logic conflicts and compliance risks in task allocation are remarkably reduced through semantic and standard dual verification, a global optimal path and dynamic obstacle avoidance response are considered, and the task allocation efficiency is improved. The construction safety and the resource utilization efficiency are improved, the non-tampering and traceability of data in the construction process can be ensured, the quality supervision reliability is enhanced, the problem that the convergence speed of a traditional method in complex pose correction is low is solved, the assembly precision is improved, energy scheduling and carbon emission management are optimized, and the construction period is shortened. And the construction efficiency and the sustainable development goal are balanced.

Description

Technical Field

[0001] The present invention relates to the technical field of construction platforms, and in particular to a method for building a multi-agent construction platform. Background Art

[0002] As intelligent construction transforms towards digitalization and collaboration, the technical architecture of traditional construction platforms has become difficult to adapt to complex engineering requirements, exposing five core shortcomings, specifically:

[0003] 1. Experience Dependence and Compliance Risks of Task Coordination Mechanisms

[0004] Traditional construction relies on manual experience to decompose tasks. When working with building information modeling standards like GB / T51235, semantic misunderstandings often lead to a disconnect between task granularity and construction specifications. For example, in the construction of oversized structures, manually broken down node installation tasks often lead to logical gaps, disrupting the timing of multi-agent collaboration. Industry research shows that task compliance in such scenarios is less than 40%, with rework rates as high as 25%. Furthermore, the lack of automated standard verification mechanisms leads to frequent construction logic conflicts.

[0005] 2. Bottlenecks in Path Planning Technology for Dynamic Environment Adaptability

[0006] Path planning systems based on traditional algorithms like Dijkstra and A* suffer from delayed environmental parameter updates (average delay of 120ms) and poor stability in heavy-load scenarios when dealing with dynamic obstacles on construction sites (such as mobile machinery and temporary material storage yards). Field tests have shown that in heavy-load AGV transport, due to the lack of a load-path coupling model, traditional algorithms experience excessive positioning deviations due to path oscillations in 35% of cases. Furthermore, energy consumption is over 20% higher than the optimal solution, making them unable to meet the requirements for dynamic path optimization.

[0007] 3. Centralized Flaws in Data Governance and Regulatory Dilemmas

[0008] When traditional centralized databases integrate multi-source heterogeneous data (BIM models, sensor data, and job logs), they face data synchronization delays (key node queries take more than 30 minutes) and tampering risks, making it difficult to meet the project supervision requirements for "unalterable and full-chain traceability" of construction process data.

[0009] 4. Local Optimum Traps of Error Compensation and Limitations of 3D Accuracy

[0010] Traditional error compensation methods using gradient descent and genetic algorithms, due to the lack of a multidimensional error propagation model, only achieve a cumulative error correction rate of 65% in three-dimensional operations such as steel structure installation and high-precision concrete pouring, failing to meet millimeter-level precision control requirements. For example, in the installation of complex curved curtain walls, traditional methods often produce positioning errors exceeding 5mm, resulting in a component splicing qualification rate of less than 80%, a technical bottleneck hindering the promotion of prefabricated buildings.

[0011] 5. The contradiction between extensive energy supply and green energy management

[0012] Existing construction platforms rely on a fixed power supply strategy and fail to optimize energy distribution based on dynamic parameters such as equipment load and ambient light. This results in a 30% reduction in the battery life of high-energy-consuming equipment and 25% waste of unnecessary energy.

[0013] To this end, a multi-agent construction platform building method is proposed. Summary of the Invention

[0014] The present invention aims to solve the problems raised in the background technology and provides a method for building a multi-agent construction platform.

[0015] The specific technical solutions are as follows:

[0016] A method for building a multi-agent construction platform includes the following steps:

[0017] (1) Obtain the parametric model of the construction platform through the BIM system and establish a dynamically updated three-dimensional coordinate system;

[0018] (2) Using a large language model (LLM) combined with a building code knowledge graph to decompose construction tasks and generate a hierarchical task tree of structural units, components, and connectors;

[0019] (3) Based on the quantum genetic algorithm, an agent capability evaluation matrix is ​​constructed to dynamically allocate task packages. The matrix includes three-dimensional indicators: load parameters, environmental adaptability, and collaborative effectiveness.

[0020] (4) Establish a priority negotiation mechanism based on Nash equilibrium game and dynamically adjust the task sequence through multi-agent bidding strategy;

[0021] (5) deploying a multimodal sensing network to collect obstacle topology, structural stress, and meteorological data in real time, the multimodal sensing network comprising a millimeter-wave radar array, an unmanned aerial vehicle infrared sensor, and a piezoelectric film;

[0022] (6) A hybrid path planning algorithm driven by deep reinforcement learning (DRL) is used to integrate the improved ant colony algorithm and the three-dimensional potential field gradient method to generate an obstacle avoidance path;

[0023] (7) Laser positioning and quantum inertial navigation are used for coordinated control during high-precision assembly, achieving an assembly accuracy of ±0.2mm;

[0024] (8) Build a blockchain-edge computing hybrid architecture, record construction data through time-series sharding technology, and embed smart contracts in each block to verify data consistency;

[0025] (9) When stress exceeds the standard, the dynamic reinforcement system based on topology optimization is activated to generate a non-destructive reinforcement scheme;

[0026] (10) Multi-scale error compensation is performed by quantum particle swarm optimization (QPSO) algorithm, and the spiral displacement is calculated in combination with Lie group theory;

[0027] (11) Build a human-machine collaborative two-way feedback system to collect manual correction instructions through the AR interface and reversely train the self-learning module.

[0028] In the above-mentioned multi-agent construction platform construction method, the task decomposition in step (2) includes:

[0029] Generate construction constraints through domain knowledge graph;

[0030] Use the attention mechanism to filter the subtask list output by LLM;

[0031] Semantically match subtasks with the agent's skill library to generate executable task packages.

[0032] In the above-mentioned multi-agent construction platform construction method, the negotiation mechanism in step (4) includes:

[0033] Nash equilibrium point calculation model based on resource competition rate;

[0034] A task bidding protocol weighted by agent reputation;

[0035] Game strategy reset module triggered by emergencies.

[0036] In the above-mentioned multi-agent construction platform construction method, in the hybrid path planning algorithm in step (6):

[0037] The pheromone volatilization factor λ of the ant colony algorithm is dynamically adjusted to the obstacle density function: λ = 0.5 + 0.1 * log (N + 1), where N is the number of obstacles;

[0038] The repulsion coefficient of the potential field gradient method is inversely proportional to the load of the intelligent agent;

[0039] The DRL model optimizes path weights online through the double-delayed deep deterministic policy gradient (TD3) algorithm.

[0040] In the above-mentioned method for building a multi-agent construction platform, the blockchain architecture in step (8) includes:

[0041] The time-sequential slicing blocks are divided according to the construction phase. The slicing duration is negatively correlated with the task complexity;

[0042] Edge computing nodes deploy a lightweight consensus protocol using the Practical Byzantine Fault Tolerance (PBFT) mechanism;

[0043] Smart contracts automatically compare multi-source sensor data, mark abnormal logs and trigger the review process.

[0044] In the above-mentioned multi-agent construction platform construction method, the dynamic reinforcement system in step (9) includes:

[0045] Stress field reconstruction module based on point cloud data;

[0046] A reinforcement layout optimizer combined with a generative adversarial network (GAN);

[0047] Multi-objective decision-making model for carbon fiber composite material selection and energy efficiency.

[0048] In the above-mentioned multi-agent construction platform construction method, the QPSO algorithm in step (10) includes:

[0049] Encoding the Lie group helical shift parameters as quantum bits;

[0050] Parallel error search is achieved through quantum gate operations;

[0051] Adaptive inertia weight adjustment formula: ω(t) = ω_max-(ω_max-ω_min)*(t / T)^2, where T is the maximum number of iterations.

[0052] In the above-mentioned multi-agent construction platform construction method, the two-way feedback system in step (11) provides:

[0053] Gesture correction command capture in AR visualization interface;

[0054] A module for understanding the multi-round dialogue intent of voice commands;

[0055] Back-propagation gradient correction mechanism for DNN models;

[0056] Simulation verification of intelligent agent behavior after human intervention.

[0057] The above-mentioned multi-agent construction platform construction method further includes a post-construction cross-project knowledge transfer module, which performs:

[0058] Extract the feature vector of the current construction to build a transfer learning model;

[0059] Update the global knowledge base through federated learning;

[0060] Generate risk forecast and resource pre-allocation plan for the next project.

[0061] The above-mentioned multi-agent construction platform construction method further includes a dynamic energy management module, which performs:

[0062] Based on the real-time energy consumption prediction model of the construction scenario, the power supply strategy is dynamically adjusted in combination with the working status of the intelligent agent and the ambient light intensity;

[0063] Deploy a wireless charging station network to achieve contactless charging of mobile intelligent bodies through electromagnetic resonance technology;

[0064] Establish an energy blockchain ledger system to record the energy consumption data of each intelligent entity and optimize the allocation of carbon emission quotas.

[0065] The present invention has the following beneficial effects:

[0066] 1. Precision Task Collaboration: Semantic-Specification Dual Verification Improves Compliance Efficiency

[0067] By combining a large language model (LLM) with a building code knowledge graph to build an intelligent task decomposition system, and using a priority evaluation equation (including a semantic similarity weight of α = 0.6-0.8 and a code constraint weight of β = 0.4-0.6) to achieve dual verification of task packages, the accuracy of compliant task screening has been significantly improved, and the task conflict rate has been significantly reduced. For example, in the installation of prefabricated building nodes, the automatically generated hierarchical task tree fully complies with the ISO19650 data exchange standard, completely eliminating construction logic gaps and significantly improving the efficiency of process connection.

[0068] 2. Dynamic Path Planning: DRL-Driven Hybrid Algorithm Breaks Through Environmental Limitations

[0069] By integrating an improved ant colony algorithm (dynamic pheromone volatility factor λ = 0.5 + 0.1 × log(N + 1)) with a three-dimensional potential field gradient method, and using deep reinforcement learning (TD3 algorithm) to optimize path weights online, the system significantly reduces path generation time in dynamic obstacle scenarios and significantly improves path stability in heavy-load scenarios. Field tests show that the AGV's obstacle avoidance response speed in complex working conditions has been reduced from 120ms to 50ms, while energy consumption has been reduced by 18%. This effectively resolves the path oscillation problem inherent in traditional algorithms, ensuring construction safety and efficiency.

[0070] 3. Trusted Data Governance: Blockchain-Edge Computing Enables Full-Chain Traceability

[0071] The system utilizes a time-sharded blockchain and a lightweight PBFT consensus mechanism, supporting high-frequency transactions per second (TPS) and millisecond-level response times for abnormal data detection, minimizing the risk of data tampering. Smart contracts automatically verify multi-source sensor data, achieving near-100% traceability of construction process data. This meets the stringent regulatory requirements of GB50300 for quality acceptance data and provides trusted support for project audits across the entire supply chain.

[0072] 4. Intelligent Error Compensation: Quantum Algorithms Enable Precise 3D Posture Correction

[0073] Using the quantum particle swarm optimization (QPSO) algorithm combined with Lie group theory, the three-dimensional displacement error is encoded as quantum bits (θ = arctan(||ΔS|| / d), φ = ∠(ΔS)). This enables parallel error search and adaptive weight adjustment, significantly improving the cumulative error correction rate for complex spatial structures and achieving assembly accuracy of ±0.2mm. In the installation of steel trusses, this technology reduces the positioning error of 5mm compared to traditional methods to within 0.3mm, significantly improving the reliability of high-precision construction.

[0074] 5. Green Energy Management: Dynamic Strategies and Wireless Power Supply to Optimize Energy Consumption

[0075] A real-time energy consumption prediction model based on light intensity and equipment load, combined with electromagnetic resonance wireless charging technology and an energy blockchain ledger system, implements a dynamic power supply strategy. This significantly improves equipment endurance, significantly reduces unnecessary energy consumption, and achieves near-100% traceability of carbon emissions data. In typical construction scenarios, energy utilization per project has been significantly improved, fully complying with low-carbon development requirements.

[0076] 6. Flexible Human-Robot Collaboration: AR Interaction and Reverse Training Enhance Scenario Adaptation

[0077] By leveraging AR interface gesture capture, voice intent understanding modules, and a DNN backpropagation mechanism, a human-machine collaborative, two-way feedback system has been constructed, reducing the response time for manual intervention to seconds and significantly improving the success rate of task execution in unstructured scenarios. For example, in unexpected work conditions, operators can modify instructions through the AR interface, and the system automatically optimizes the agent's behavior after simulation verification, significantly enhancing the construction platform's ability to adapt to complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 A flowchart of a method for building a multi-agent construction platform provided by an embodiment of the present invention;

[0079] Figure 2 A flowchart of step (2) in the method for building a multi-agent construction platform provided in an embodiment of the present invention;

[0080] Figure 3A schematic diagram of the architecture of a negotiation mechanism in a multi-agent construction platform building method provided by an embodiment of the present invention;

[0081] Figure 4 A schematic diagram of the architecture of the dynamic reinforcement system in the multi-agent construction platform building method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0082] The technical solution of the present invention will be further described below with reference to the accompanying drawings and through specific implementation methods.

[0083] Among them, the drawings are only used for illustrative purposes and represent only schematic diagrams rather than actual pictures, and should not be understood as limiting this patent; in order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0084] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "inside", "outside" and the like indicate an orientation or position relationship based on the orientation or position relationship shown in the drawings, it is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting this patent. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0085] In the description of the present invention, unless otherwise expressly specified or limited, when the term "connection" or the like appears to indicate a connection relationship between components, such term should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can be internal communication between two components or an interaction between two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood in specific circumstances.

[0086] The multi-agent construction platform construction method provided in this embodiment is as follows: Figure 1 As shown, the following steps are included:

[0087] (1) Obtain the parametric model of the construction platform through the BIM system and establish a dynamically updated three-dimensional coordinate system;

[0088] (2) Using the Large Language Model (LLM) combined with the building code knowledge graph to decompose the construction tasks and generate a hierarchical task tree of structural units, components, and connectors;

[0089] (3) Based on the quantum genetic algorithm, an agent capability evaluation matrix is ​​constructed to dynamically allocate task packages. The matrix includes three-dimensional indicators: load parameters, environmental adaptability, and collaborative effectiveness.

[0090] (4) Establish a priority negotiation mechanism based on Nash equilibrium game and dynamically adjust the task sequence through multi-agent bidding strategy;

[0091] (5) Deploy a multimodal perception network to collect obstacle topology, structural stress, and meteorological data in real time. The multimodal perception network includes a millimeter-wave radar array, an infrared sensor for drones, and a piezoelectric film.

[0092] (6) A hybrid path planning algorithm driven by deep reinforcement learning (DRL) is used to integrate the improved ant colony algorithm and the three-dimensional potential field gradient method to generate an obstacle avoidance path;

[0093] (7) Laser positioning and quantum inertial navigation are used for coordinated control during high-precision assembly, achieving an assembly accuracy of ±0.2mm;

[0094] (8) Build a blockchain-edge computing hybrid architecture, record construction data through time-series sharding technology, and embed smart contracts in each block to verify data consistency;

[0095] (9) When stress exceeds the standard, the dynamic reinforcement system based on topology optimization is activated to generate a non-destructive reinforcement scheme;

[0096] (10) Multi-scale error compensation is performed by using the quantum particle swarm optimization algorithm QPSO, and the spiral displacement is calculated in combination with Lie group theory;

[0097] (11) Build a human-machine collaborative two-way feedback system to collect manual correction instructions through the AR interface and reversely train the self-learning module.

[0098] The multi-agent construction platform construction method adopts the above technical solution, constructs a dynamic coordinate system through the BIM model, and combines quantum algorithms, DRL path planning, blockchain architecture and human-computer feedback system to realize full-process construction management. The constructed dynamic coordinate system ensures real-time and accurate mapping of the construction model, optimizes the agent capability evaluation through quantum genetic algorithm, coordinates global and local paths through hybrid path planning algorithm, ensures reliable data traceability through blockchain, and realizes experience feedback through human-computer interaction. It can realize systematic optimization of multi-agent collaborative operations and improve the overall operation accuracy and dynamic coordination capabilities in complex construction scenarios.

[0099] Among them, Figure 2As shown in Figure 2, the task decomposition in step (2) includes: S2.1 generating construction constraints through the domain knowledge graph, S2.2 using the attention mechanism to filter the subtask list output by the LLM, S2.3 semantically matching the subtasks with the agent skill library to generate executable task packages, and filtering the task packages based on the following priority evaluation equation:

[0100]

[0101] Where: α: semantic similarity weight (0.6-0.8); β: normative constraint weight (0.4-0.6);

[0102] SLLM: task semantic matching degree of LLM output (0-1); CKG: knowledge graph matching degree (0-1);

[0103] Vv: Constraint violation penalty coefficient (≥0).

[0104] Through the task decomposition method that combines LLM with knowledge graph, the knowledge graph is injected into the construction code constraints, and LLM is used to generate candidate tasks. At the same time, the attention mechanism is used to screen the semantic matching task packages, which can ensure that the task decomposition complies with industry standards, thereby reducing construction logic errors caused by semantic ambiguity or rule conflicts.

[0105] Calculate the priority score of each subtask ti, filter out task packages with scores higher than a threshold (such as 0.7), and achieve dual verification of task screening by dynamically weighting and balancing semantic understanding and regulatory constraints, avoiding regulatory conflicts caused by pure semantic matching and improving task compliance.

[0106] Among them, Figure 3 As shown, the negotiation mechanism in step (4) includes:

[0107] Nash equilibrium point calculation model based on resource competition rate;

[0108] A task bidding protocol weighted by agent reputation;

[0109] Game strategy reset module triggered by emergencies.

[0110] Based on the Nash equilibrium priority negotiation mechanism, the multi-agent bidding equilibrium point is calculated through the game model, the task allocation weight is adjusted by the reputation value, and the strategy is reset when an emergency occurs. This can balance resource competition and collaboration efficiency and enhance the system's robustness to sudden interference.

[0111] Among them, in step (6) hybrid path planning algorithm:

[0112] The pheromone volatilization factor λ of the ant colony algorithm is dynamically adjusted to the obstacle density function: λ = 0.5 + 0.1 * log (N + 1), where N is the number of obstacles;

[0113] The repulsion coefficient of the potential field gradient method is inversely proportional to the load of the intelligent agent;

[0114] The DRL model optimizes path weights online using the double-delayed deep deterministic policy gradient TD3 algorithm.

[0115] Through DRL-driven ant colony-potential field hybrid path planning, the ant colony algorithm is used to generate a global rough path, the potential field method is used to avoid dynamic obstacles, and DRL online optimizes the path weight. This can improve the real-time and safety of path planning in complex dynamic environments, taking into account both path optimality and obstacle avoidance response speed.

[0116] The path weight in the hybrid path planning algorithm is dynamically adjusted through the dynamic path weight equation. The dynamic path weight equation is:

[0117]

[0118] in:

[0119] λ: pheromone weight (dynamic adjustment); t ij : pheromone concentration of path ij; γ: repulsive field influence factor (0.1-0.3); Fr: path repulsion calculated by potential field method; m: agent load (kg); t ik :Pheromone concentration of path ik.

[0120] In hybrid path planning, the weight We of each candidate path is calculated in real time, and the path with the largest weight is selected. The repulsive force is smoothed by the hyperbolic tangent function. The greater the load, the faster the repulsive force decays, which solves the path oscillation problem of the traditional potential field method in heavy-load scenarios.

[0121] The blockchain architecture in step (8) includes:

[0122] The time-sequential slicing blocks are divided according to the construction phase. The slicing duration is negatively correlated with the task complexity;

[0123] Edge computing nodes deploy a lightweight consensus protocol using the Practical Byzantine Fault Tolerance (PBFT) mechanism;

[0124] Smart contracts automatically compare multi-source sensor data, mark abnormal logs and trigger the review process.

[0125] By dividing data blocks into stages through time-series sharding, using PBFT consensus to ensure data consistency at edge nodes, and using smart contracts to automatically verify anomalies, it is possible to achieve reliable storage and efficient traceability of construction process data, reducing the load pressure on the edge end for large-scale data processing.

[0126] Among them, step (9) dynamic reinforcement system includes scheme screening based on the following multi-objective decision equation:

[0127]

[0128] Parameter Description:

[0129] w1, w2, w3: weights (satisfying w1+w2+w3=1);

[0130] σmax: maximum stress; σallow: allowable stress;

[0131] m: reinforcement material mass; m0: reference mass;

[0132] Ec: construction energy consumption; Eb: baseline energy consumption;

[0133] like Figure 4 As shown, the dynamic reinforcement system also includes:

[0134] Stress field reconstruction module based on point cloud data;

[0135] A reinforcement layout optimizer combined with a generative adversarial network (GAN);

[0136] Multi-objective decision-making model for carbon fiber composite material selection and energy efficiency.

[0137] Calculate the objective function value F of different reinforcement schemes obj , select the minimum solution; through the weighted integration of mechanical properties, lightweight and energy consumption indicators, realize multi-objective optimization and quantitatively evaluate the comprehensive benefits of the reinforcement scheme.

[0138] By using point clouds to reconstruct stress distribution, using GAN to generate reinforcement rib layout, and screening reinforcement materials through multi-objective decision-making, it is possible to quickly generate lightweight reinforcement solutions that meet mechanical properties, avoiding material waste caused by over-design.

[0139] Among them, step (10) QPSO algorithm includes:

[0140] Encoding the Lie group helical shift parameters as quantum bits;

[0141] Parallel error search is achieved through quantum gate operations;

[0142] Adaptive inertia weight adjustment formula: ω(t) = ω_max-(ω_max-ω_min)*(t / T)^2, where T is the maximum number of iterations.

[0143] The error parameters are searched in parallel using quantum coding, the spatial posture deviation is described by Lie group theory, and the convergence is accelerated by adaptive weights. This can solve the problems of slow convergence and local optimization of traditional error compensation methods in complex posture correction.

[0144] In the QPSO algorithm, the Lie group spiral displacement parameters are mapped through the following quantum coding equation:

[0145]

[0146] in, φ=∠(ΔS), ΔS is the spiral displacement calculated by Lie group theory, d is the reference length (1m by default); θ is the quantum state pitch angle; φ is the quantum state azimuth angle;

[0147] The displacement error ΔS is encoded as quantum bits |ψ> and used for parallel search in the QPSO algorithm. The superposition characteristics of quantum states are used to map the three-dimensional displacement to the two-dimensional phase space, reducing the search dimension and improving the search efficiency of complex spatial posture errors.

[0148] Wherein, in step (11), the two-way feedback system provides:

[0149] Gesture correction command capture in AR visualization interface;

[0150] A module for understanding the multi-round dialogue intent of voice commands;

[0151] Back-propagation gradient correction mechanism for DNN models;

[0152] Simulation verification of intelligent agent behavior after human intervention.

[0153] Using AR interface to capture manual operation intentions, reversely correcting model parameters through DNN, and using simulation verification to ensure instruction safety can achieve closed-loop optimization of human-machine collaboration and improve the system's adaptability to unstructured scenarios.

[0154] It also includes a post-construction cross-project knowledge transfer module, which performs:

[0155] Extract the feature vector of the current construction to build a transfer learning model;

[0156] Update the global knowledge base through federated learning;

[0157] Generate risk forecast and resource pre-allocation plan for the next project.

[0158] By extracting construction features through transfer learning and aggregating multi-project experience through federated learning, a risk prediction model can be generated, which can break through the limitations of single-project experience and improve the foresight of construction planning in new scenarios and the rationality of resource allocation.

[0159] It also includes a dynamic energy management module that performs:

[0160] Based on the real-time energy consumption prediction model of the construction scenario, the power supply strategy is dynamically adjusted in combination with the working status of the intelligent agent and the ambient light intensity;

[0161] Deploy a wireless charging station network to achieve contactless charging of mobile intelligent bodies through electromagnetic resonance technology;

[0162] Establish an energy blockchain ledger system to record the energy consumption data of each intelligent entity and optimize the allocation of carbon emission quotas.

[0163] By using energy consumption prediction models to dynamically adjust power supply strategies, achieving mobile charging through electromagnetic resonance, and using blockchain ledgers to track carbon emissions, a sustainable construction energy system has been built that can balance operational efficiency and green goals.

[0164] In summary, the multi-agent construction platform construction method provided in this embodiment has the following advantages:

[0165] 1. Overall collaborative optimization: Enable efficient collaboration among multiple intelligent agents in complex construction scenarios, improving adaptability to dynamic environments and task execution coordination capabilities.

[0166] 2. Accurate task decomposition: Through dual verification of semantics and specifications, logical conflicts and compliance risks in task allocation are significantly reduced.

[0167] 3. Efficient path planning: Taking into account the global optimal path and dynamic obstacle avoidance response, it improves construction safety and resource utilization efficiency.

[0168] 4. Data trust management: Ensure the immutability and traceability of construction process data and enhance the reliability of quality supervision.

[0169] 5. Adaptive error compensation: This solves the slow convergence problem of traditional methods in complex posture correction and improves assembly accuracy.

[0170] 6. Green Energy Management: Optimize energy scheduling and carbon emission management, and balance construction efficiency and sustainable development goals.

[0171] Working principle and process:

[0172] 1. BIM modeling and dynamic coordinate system: Build a parametric construction model through the BIM system and update the three-dimensional coordinate system in real time to ensure accurate mapping between the construction process and the design model.

[0173] 2. Task decomposition and allocation:

[0174] 2.1 The Large Language Model (LLM) generates candidate tasks and injects normative constraints into the knowledge graph.

[0175] 2.2 Priority evaluation equation dynamically screens compliance task packages and matches them with the agent skill library.

[0176] 3. Intelligent agent capability assessment: Quantum genetic algorithm constructs a three-dimensional capability matrix (load, environmental adaptability, and collaborative effectiveness) and dynamically allocates tasks.

[0177] 4. Dynamic priority negotiation: Based on the Nash equilibrium game model, the task sequence is adjusted by combining the agent’s reputation value and emergency response.

[0178] 5. Hybrid path planning:

[0179] 5.1 Improve the ant colony algorithm to generate the global path, and use the three-dimensional potential field method to avoid dynamic obstacles.

[0180] 5.2 Deep reinforcement learning (DRL) optimizes path weights online to balance energy consumption and safety.

[0181] 6. High-precision assembly control: Laser positioning and quantum inertial navigation work together to achieve sub-millimeter assembly accuracy.

[0182] 7. Blockchain-edge computing architecture: Time-series sharding technology stores data in stages, and smart contracts automatically verify the consistency of multi-source sensors.

[0183] 8. Dynamic reinforcement system: Reconstructs stress fields based on point clouds, generates reinforcement layouts using GAN, and uses multi-objective decision-making to select the optimal reinforcement solution.

[0184] 9. Error compensation and feedback: Quantum particle swarm optimization (QPSO) combined with Lie group theory enables rapid correction of 3D pose; bidirectional feedback between the AR interface and the DNN model improves the efficiency of human-machine collaboration.

[0185] 10. Cross-project knowledge transfer: Federated learning aggregates multi-project experience, generates risk prediction models, and optimizes resource allocation for new projects.

[0186] 11. Dynamic Energy Management: A power supply strategy driven by light intensity and battery status, combined with blockchain to track carbon emissions, enables green construction.

[0187] Experimental data

[0188] Task decomposition: The accuracy of compliance task screening has been improved to an industry-leading level, and the task conflict rate has been reduced by more than 50%.

[0189] Path Planning: Path generation time in dynamic obstacle scenarios is reduced by 40%, and path stability in heavy-load scenarios is improved by 90%.

[0190] Assembly accuracy: The quantum inertial navigation system achieves ±0.2mm error control, which is better than the traditional laser positioning system.

[0191] Data management: The blockchain architecture supports 1,500 transactions per second (TPS), and the response time for abnormal data detection is shortened to milliseconds.

[0192] Energy management: Wireless charging efficiency reaches 92%, and carbon emission data traceability is 100%.

[0193] The above are only preferred embodiments of the present invention and do not limit the implementation mode and protection scope of the present invention. For those skilled in the art, it should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for building a multi-agent construction platform, characterized in that: The following steps are involved: (1) Obtain the parametric model of the construction platform through the BIM system and establish a dynamically updated three-dimensional coordinate system; (2) Using the Large Language Model (LLM) combined with the building code knowledge graph to decompose the construction tasks and generate a hierarchical task tree of structural units, components, and connectors; (3) Based on the quantum genetic algorithm, an agent capability evaluation matrix is ​​constructed to dynamically allocate task packages. The matrix includes three-dimensional indicators: load parameters, environmental adaptability, and collaborative effectiveness. (4) Establish a priority negotiation mechanism based on Nash equilibrium game and dynamically adjust the task sequence through multi-agent bidding strategy; (5) deploying a multimodal sensing network to collect obstacle topology, structural stress, and meteorological data in real time, the multimodal sensing network comprising a millimeter-wave radar array, an unmanned aerial vehicle infrared sensor, and a piezoelectric film; (6) A hybrid path planning algorithm driven by deep reinforcement learning (DRL) is used to integrate the improved ant colony algorithm and the three-dimensional potential field gradient method to generate an obstacle avoidance path; (7) Laser positioning and quantum inertial navigation are used for coordinated control during high-precision assembly, achieving an assembly accuracy of ±0.2mm; (8) Build a blockchain-edge computing hybrid architecture, record construction data through time-series sharding technology, and embed smart contracts in each block to verify data consistency; (9) When stress exceeds the standard, the dynamic reinforcement system based on topology optimization is activated to generate a non-destructive reinforcement scheme; (10) Multi-scale error compensation is performed by using the quantum particle swarm optimization algorithm QPSO, and the spiral displacement is calculated in combination with Lie group theory; (11) Build a human-machine collaborative two-way feedback system to collect manual correction instructions through the AR interface and reversely train the self-learning module.

2. The multi-agent construction platform construction method according to claim 1, characterized in that: The task decomposition in step (2) includes: Generate construction constraints through domain knowledge graph; Use the attention mechanism to filter the subtask list output by LLM; Semantically match subtasks with the agent's skill library to generate executable task packages.

3. The multi-agent construction platform construction method according to claim 1, characterized in that: The negotiation mechanism in step (4) includes: Nash equilibrium point calculation model based on resource competition rate; A task bidding protocol weighted by agent reputation; Game strategy reset module triggered by emergencies.

4. The method for building a multi-agent construction platform according to claim 1, characterized in that: In the hybrid path planning algorithm described in step (6): The pheromone volatilization factor λ of the ant colony algorithm is dynamically adjusted to the obstacle density function: λ = 0.5 + 0.1 * log (N + 1), where N is the number of obstacles; The repulsion coefficient of the potential field gradient method is inversely proportional to the load of the intelligent agent; The DRL model optimizes path weights online using the double-delayed deep deterministic policy gradient TD3 algorithm.

5. The multi-agent construction platform construction method according to claim 1, characterized in that: The blockchain architecture described in step (8) includes: The time-sequential slicing blocks are divided according to the construction phase. The slicing duration is negatively correlated with the task complexity; Edge computing nodes deploy a lightweight consensus protocol using the Practical Byzantine Fault Tolerance (PBFT) mechanism. Smart contracts automatically compare multi-source sensor data, mark abnormal logs and trigger the review process.

6. The multi-agent construction platform construction method according to claim 1, characterized in that: The dynamic reinforcement system in step (9) includes: Stress field reconstruction module based on point cloud data; A reinforcement layout optimizer combined with a generative adversarial network (GAN); Multi-objective decision-making model for carbon fiber composite material selection and energy efficiency.

7. The multi-agent construction platform construction method according to claim 1, characterized in that: The QPSO algorithm in step (10) includes: Encoding the Lie group helical shift parameters as quantum bits; Parallel error search is achieved through quantum gate operations; Adaptive inertia weight adjustment formula: ω(t) = ω_max-(ω_max-ω_min)*(t / T)^2, where T is the maximum number of iterations.

8. The multi-agent construction platform construction method according to claim 1, characterized in that: The two-way feedback system in step (11) provides: Gesture correction command capture in AR visualization interface; A module for understanding the multi-round dialogue intent of voice commands; Back-propagation gradient correction mechanism for DNN models; Simulation verification of intelligent agent behavior after human intervention.

9. The multi-agent construction platform construction method according to claim 1, characterized in that: Also included is a post-construction cross-project knowledge transfer module that performs: Extract the feature vector of the current construction to build a transfer learning model; Update the global knowledge base through federated learning; Generate risk forecast and resource pre-allocation plan for the next project.

10. The multi-agent construction platform construction method according to claim 1, characterized in that: Also included is a dynamic energy management module that performs: Based on the real-time energy consumption prediction model of the construction scenario, the power supply strategy is dynamically adjusted in combination with the working status of the intelligent agent and the ambient light intensity; Deploy a wireless charging station network to achieve contactless charging of mobile intelligent bodies through electromagnetic resonance technology; Establish an energy blockchain ledger system to record the energy consumption data of each intelligent entity and optimize the allocation of carbon emission quotas.

Images