Building construction carbon emission control method and system based on large model agent

Abstract

The application discloses a building construction carbon emission control method and system based on a large model intelligent agent, and belongs to the technical field of building engineering informatization and energy saving and emission reduction. The method is based on the operation of a large model intelligent agent system driven by a large language model kernel, and comprises the following steps: acquiring multi-source Internet of Things sensing data of a construction site through an edge computing gateway; performing data cleaning by calling a tool based on a reasoning and action cooperative mechanism, and completing standardized feature conversion by a sensing intelligent agent; relying on a domain knowledge graph with construction processes as the core as a constraint base, aligning heterogeneous entities by combining a graph attention network, and dynamically accounting for actual carbon emissions by a calculation intelligent agent; under the premise of meeting the rigid constraints of progress and quality, a decision intelligent agent fuses thought chain reasoning and a multi-objective particle swarm algorithm to perform carbon emission and cost collaborative optimization, and outputs and issues an optimal emission reduction scheduling strategy. The application realizes accurate calculation and intelligent closed-loop control of carbon emissions in a complex construction scene.

Description

Technical Field

[0001] This invention belongs to the field of building engineering informatization and energy conservation and emission reduction technology, specifically involving a method and system for controlling carbon emissions from building construction based on a large-scale intelligent model. Background Technology

[0002] In the global trend of addressing climate change and reducing greenhouse gas emissions, the construction industry, as a key high-emission sector, faces crucial challenges in its low-carbon transformation. The construction phase of building projects, characterized by high work intensity, rapid resource turnover, and highly concentrated energy consumption, is one of the most carbon-intensive stages in the entire building lifecycle. Therefore, precise assessment and dynamic management of carbon emissions at construction sites have become crucial for achieving energy conservation and emission reduction in the construction industry. However, the complex environment and intricate procedures at construction sites result in highly dynamic and uncertain carbon emissions.

[0003] Currently, existing technical solutions for controlling carbon emissions from construction mainly fall into two categories: one is an information-based management and control system based on conventional Internet of Things (IoT), and the other is an auxiliary analysis method that has been initially explored in recent years by introducing a single large language model. However, both of these solutions have significant limitations in practical engineering applications.

[0004] First, conventional IoT management and control systems generally suffer from a disconnect between accounting and control. While these systems can collect data, they lack logical understanding capabilities, making them highly susceptible to distortion of underlying activity data when faced with multi-source, heterogeneous, and noisy sensing data. Furthermore, most rely on static models that multiply activity data by emission factors for post-event calculations, failing to integrate real-time project schedules and environmental data for dynamic attribution analysis of exceedances, thus hindering proactive control.

[0005] Secondly, some studies have attempted to introduce a single large-scale model for carbon emission analysis, but general-purpose large-scale models have serious shortcomings when dealing with complex engineering tasks. On the one hand, when large-scale models directly perform high-precision carbon emission quantification multiplication and division operations, they are prone to numerical calculation bias and factual fabrication. On the other hand, single models lack the constraints of professional knowledge graphs in the construction field, and cannot independently handle the complex and long-chain task decomposition from data perception and cleaning, quantitative accounting to multi-objective scheduling decisions under the premise of meeting multiple engineering constraints such as quality control, cost budgeting, and schedule planning. It is difficult to generate a comprehensive and coordinated management strategy that integrates engineering quality, cost quotas, project schedule, and carbon reduction targets.

[0006] In summary, existing construction carbon emission control technologies cannot effectively handle the intelligent processing and accurate calculation of multi-source heterogeneous data, nor can they autonomously deduce the optimal scheduling decision that can be implemented in a complex dynamic environment that meets multiple constraints of project progress, cost and quality. There is an urgent need for a new intelligent control solution. Summary of the Invention

[0007] The purpose of this invention is to overcome the above-mentioned shortcomings of the prior art and provide a method and system for carbon emission control in building construction based on a large model intelligent agent, aiming to achieve high-precision dynamic closed-loop control of carbon emissions in complex construction environments while taking into account multiple constraints such as schedule, cost and quality.

[0008] The objective of this invention can be achieved through the following technical solutions:

[0009] This invention provides a method for controlling carbon emissions from building construction based on a large-scale intelligent agent system. The method operates on a large-scale intelligent agent system driven by a large language model kernel, which shares the semantic understanding and logical reasoning capabilities of the large language model kernel. The method includes the following steps:

[0010] S1. Multi-source IoT sensing: Acquire multi-source IoT sensing data at the construction site. The sensing data includes at least: energy and water consumption data of equipment and temporary facilities, material consumption data, dynamic load and trajectory data of logistics transportation, and environmental monitoring data.

[0011] S2. Standardized Feature Conversion: The multi-source IoT sensing data is converted into a standardized state feature representation and input into a preset large model intelligent agent system;

[0012] S3. Carbon emission collaborative reasoning: The large model intelligent agent system performs collaborative reasoning on the state feature representation to assess the carbon emission status of the current construction stage and generate carbon emission diagnostic results.

[0013] S4. Carbon emission decision issuance: Based on the carbon emission diagnosis results, the large model intelligent agent system generates corresponding carbon emission control decisions and issues the control decisions to the corresponding construction management terminal or IoT control device.

[0014] Furthermore, the large-scale model intelligent agent system includes a perceptual agent, a computational agent, and a decision-making agent; the perceptual agent, computational agent, and decision-making agent are all driven by the underlying kernel of the large language model; wherein:

[0015] The sensing agent is used to call the underlying large language model, and uses the ReAct mechanism to perform logical cleaning and multimodal feature alignment on the multi-source IoT sensing data, and generate the state feature representation.

[0016] The accounting agent is used to call the underlying large language model and adopts a tool-calling-based computational reasoning mechanism to extract energy and water consumption data, material consumption data, and dynamic load and trajectory data of logistics transportation from the state feature representation of equipment and temporary facilities, and calculates and outputs the actual carbon emissions.

[0017] The decision-making agent calls upon the underlying large language model, combines the current construction progress, project cost constraints, construction quality standards and preset carbon emission budget, performs multi-objective deviation comparison analysis on the actual carbon emissions, and generates specific emission reduction scheduling instructions.

[0018] Furthermore, during the execution of the multi-source IoT sensing in step S1, the multi-source IoT sensing data is acquired in real time through the data acquisition interface deployed on the edge computing gateway at the construction site, and the initial format conversion is completed locally on the edge computing gateway.

[0019] Furthermore, when the perceptual agent performs the standardized feature transformation in step S2 using the ReAct mechanism, it specifically includes the following steps:

[0020] S2.1 Receive the discrete multi-source IoT sensing data as initial observations;

[0021] S2.2, Call the large language model to analyze missing values, outliers and heterogeneous formats in the data, and generate corresponding logical processing ideas;

[0022] S2.3. Generate data cleaning and alignment actions based on the aforementioned logical processing thinking, trigger external data processing tools to perform transformation, and iterate this process until the aligned state feature representation is output.

[0023] Furthermore, prior to performing the carbon emission collaborative reasoning in step S3, the method also includes the step of constructing a knowledge graph in the field of construction carbon emissions:

[0024] S3A.1. Obtain national and local carbon emission standards, carbon emission factor databases, construction quality acceptance specifications, engineering quotas and market price databases, and historical project management archives for the construction industry.

[0025] S3A.2 Extract the construction procedures, construction materials, machinery and equipment, energy and water resources, logistics and transportation, construction specifications and quality points, historical emission reduction cases and schedule as graph node entities, extract the corresponding material carbon emission factor, transportation carbon emission factor, energy and water resource carbon emission factor, benchmark energy consumption, rated power, standard quota consumption, benchmark unit price, default transportation distance, benchmark construction period node, cost budget, carbon emission budget and optimization strategy as attributes of the node entities, and extract the semantic association between each entity as relations;

[0026] S3A.3. The entities, attributes and relationships are mapped to the graph database using an attribute graph model to generate the knowledge graph of the construction carbon emission domain.

[0027] Furthermore, when the accounting agent performs the carbon emission collaborative reasoning in step S3 to calculate the actual carbon emissions, it includes the following steps:

[0028] S3B.1. Convert the state feature representation into a query entity;

[0029] S3B.2 In the pre-constructed knowledge graph of construction carbon emissions, retrieve the carbon emission factor attribute value corresponding to the query entity;

[0030] S3B.3. Trigger an external calculation module with a pre-set carbon quantification calculation formula through a tool call mechanism. The external calculation module performs quantification calculations with the corresponding carbon emission factors based on the energy consumption data, material and water resource consumption data, and dynamic load and trajectory data of logistics transportation in the state feature representation, respectively, to obtain the actual carbon emissions.

[0031] Furthermore, when the decision-making agent issues the carbon emission decision in step S4, it employs the Chainof Thought reasoning mechanism:

[0032] S4.1 Obtain the schedule plan, cost budget, quality control indicators and carbon emission budget targets corresponding to the current construction progress;

[0033] S4.2. Compare the actual carbon emissions with the carbon emission budget target, and combine the environmental monitoring data in the state feature representation to determine whether there is a budget overrun, and generate the corresponding overrun attribution entity;

[0034] S4.3 In the knowledge graph of carbon emissions in the construction field, based on the association relationship of the graph structure, retrieve the construction specifications and quality points entities and historical emission reduction case entities that match the excessive attribution entity;

[0035] S4.4 Combining the aforementioned construction specifications, quality points, and historical emission reduction cases, a thought chain reasoning mechanism is used to decompose the carbon control task into sub-tasks such as material replacement assessment, machinery scheduling optimization, and process improvement. Under the premise of satisfying the multi-objective constraints of the aforementioned schedule plan, cost budget, and quality control indicators, intermediate reasoning steps are output sequentially, and finally the emission reduction scheduling instruction is generated.

[0036] Furthermore, this invention also provides a building construction carbon emission control system based on a large-scale intelligent agent, applied to the building construction carbon emission control system based on a large-scale intelligent agent described in this invention, comprising:

[0037] The IoT sensing module is used to acquire multi-source IoT sensing data at the construction site.

[0038] The feature conversion module is used to convert the multi-source IoT sensing data into a standardized state feature representation;

[0039] The knowledge graph storage module is used to store knowledge graphs in the field of construction carbon emissions, including construction procedures, construction materials, machinery and equipment, energy and water resources, logistics and transportation, construction specifications and quality points, historical emission reduction cases, and progress plan entities.

[0040] The intelligent agent collaboration module has a built-in large model intelligent agent system, which includes a perception intelligent agent, an accounting intelligent agent, and a decision-making intelligent agent. It is used to perform perception cleaning, carbon emission quantification and accounting, and multi-objective management and control decision reasoning that combines progress, cost, quality and budget on the state feature representation.

[0041] The quantization calculation module is communicatively connected to the computing agent and is used to receive calculation instructions and perform numerical quantization calculations.

[0042] The decision execution module is used to issue emission reduction scheduling instructions to the corresponding construction management terminal or IoT control device.

[0043] In addition, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the program to implement the carbon emission control method for building construction based on a large model intelligent agent as described in the present invention.

[0044] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program thereon, characterized in that the program, when executed by a processor, implements the carbon emission control method for building construction based on a large model intelligent agent as described in the present invention.

[0045] Compared with the prior art, the technical solution of the present invention has the following main advantages:

[0046] 1. This invention constructs a large-scale multi-agent collaborative architecture comprising sensing, accounting, and decision-making modules, enabling the system to autonomously operate in complex and dynamic construction environments. This architecture, through the division of labor and cooperation among agents, achieves pre-emptive closed-loop control, from intelligent cleaning and alignment of multi-source heterogeneous sensing data and dynamic assessment of carbon emission evolution to the issuance of scheduling commands. This mechanism overcomes the problems of conventional IoT systems lacking logical understanding and lagging behind in post-event static accounting, thus improving the comprehensive management efficiency of all elements of carbon emissions at construction sites.

[0047] 2. This invention constructs a construction carbon emission domain knowledge graph encompassing construction procedures, construction materials, machinery and equipment, energy and water resources, logistics and transportation, construction specifications and quality points, historical emission reduction cases, and schedule entities, serving as a priori constraint foundation for the system's collaborative reasoning. This graph provides specialized construction engineering domain knowledge supplements to the general large language model, enabling the system to accurately retrieve entity attributes and association rules in carbon emission accounting and emission reduction scheduling. This mechanism effectively overcomes the limitations of a single model lacking specialized domain knowledge and rule constraints, improving the structuring level of the underlying data and the standardization of control strategies.

[0048] 3. This invention introduces a collaborative quantification and accounting mechanism based on a tool invocation mechanism, separating the semantic and logical reasoning of the large model from external mathematical quantification calculations. The accounting agent only utilizes the information extraction capabilities of the large model to retrieve corresponding parameters in the knowledge graph, and then passes them to the external calculation module with pre-defined formulas for execution. This mechanism effectively avoids the computational bias that easily occurs when a general-purpose large language model directly performs high-precision calculations, thus improving the absolute accuracy of carbon emission quantification and accounting.

[0049] 4. This invention introduces a dynamic collaborative decision-making mechanism oriented towards multi-dimensional engineering constraints, transforming emission reduction scheduling into a multi-objective optimization process that coordinates progress, cost, quality, and carbon emissions. The decision-making agent performs budget overrun attribution analysis based on the current construction status and, under the premise of meeting strict project performance requirements, autonomously breaks down complex emission reduction tasks into specific sub-tasks such as material replacement, machinery scheduling, and process improvement. This mechanism solves the problem that single carbon emission control measures can easily lead to project delays or cost overruns, improving the on-site feasibility of comprehensive emission reduction scheduling strategies. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the overall process of a building construction carbon emission control method based on a large model intelligent agent, provided in an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of the collaborative flow and system architecture of a large-scale intelligent agent system provided in an embodiment of the present invention;

[0052] Figure 3 This is a schematic diagram of the ontological structure of a knowledge graph in the field of construction carbon emissions provided in an embodiment of the present invention. Detailed Implementation

[0053] 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0054] Example 1: A method for carbon emission control in building construction based on large-scale intelligent agents

[0055] like Figure 1 As shown, this embodiment of the invention provides a method for controlling carbon emissions from building construction based on a large-scale intelligent agent. This method is executed by a server or distributed computing device with a built-in large-scale intelligent agent system. The large-scale intelligent agent system is driven by a low-level large language model kernel, and each agent within the system shares the semantic understanding and logical reasoning capabilities of this large language model kernel. The method specifically includes the following steps:

[0056] S1. Multi-source IoT sensing: Acquire multi-source IoT sensing data from the construction site.

[0057] In its implementation, the system acquires multi-source IoT sensing data in real time through sensors and data acquisition hardware deployed at various elements of the construction site. This sensing data includes at least:

[0058] 1. Energy and water consumption data for equipment and temporary facilities: Real-time data collected through smart meters, level sensors, and flow meters on the electricity consumption (kWh), water consumption (including construction production water and temporary domestic water, m³), ​​liquid fuel consumption (such as diesel, gasoline, etc., L), and gaseous fuel consumption (such as natural gas, liquefied petroleum gas, etc., m³) of various on-site construction machinery (including tower cranes, excavators, loaders, pump trucks, etc.) and temporary office and living facilities.

[0059] 2. Material consumption data: Real-time measurement data of all building materials, such as ready-mixed concrete, steel bars, and blocks, entering the site, as well as dynamic inventory ledgers, collected through smart weighbridges, RFID electronic tags, and material management systems.

[0060] 3. Dynamic load and trajectory data of logistics transportation: Real-time load (t), driving trajectory point sequence, speed limit and idling status data of all logistics vehicles such as dump trucks and building material transport vehicles are obtained through vehicle-mounted satellite positioning terminals (including but not limited to GPS positioning terminals, Beidou positioning terminals or multi-mode joint positioning terminals) and axle load sensors.

[0061] 4. Environmental monitoring data: Through the integrated dust and meteorological monitoring station deployed on site, environmental indicators including PM2.5, PM10, ambient temperature (°C), relative humidity (%) and wind speed (m / s) are acquired in real time.

[0062] During the multi-source IoT sensing process in step S1, to alleviate the bandwidth and computing power pressure on the cloud server for concurrent inference of large models, the aforementioned multi-source IoT sensing data is acquired in real time through the data acquisition interface deployed on the edge computing gateway at the construction site. The edge computing gateway has a built-in protocol parsing engine that converts discrete industrial serial port or network protocol data such as Modbus, MQTT, and RS485 into standardized JSON streaming data. After completing the initial format conversion, the data is uploaded to the large model intelligent agent system through a secure and encrypted transmission channel.

[0063] S2. Standardized Feature Conversion: The multi-source IoT sensing data is converted into a standardized state feature representation and input into a preset large model intelligent agent system.

[0064] Due to the complex electromagnetic environment and frequent overlap of processes at construction sites, the collected sensing data streams often contain missing data, outlier noise points, and heterogeneous formats. For example... Figure 2 As shown, the system calls the perceptual agent to perform logical cleaning and multimodal feature alignment on the initial data.

[0065] The perceptual agent invokes the underlying large language model and employs the ReAct mechanism for autonomous logical reasoning and tool invocation. In terms of specific underlying mathematical implementation, this embodiment formally models the process of cleaning heterogeneous noisy data faced by the perceptual agent as a partially observable Markov decision process.

[0066] Suppose At any given time, the multi-source discrete observation data stream received by the sensing agent is The underlying large language model of the intelligent agent is based on historical observation sequences. With historical action sequence It generates implicit logical processing state representations of thought through an internal nonlinear reasoning mapping function. :

[0067]

[0068] in, This represents the logical reasoning mapping function of a large language model based on pre-trained weights and attention mechanisms.

[0069] Perceptual agents analyze Data cleaning and alignment actions are generated based on missing values, outliers, and heterogeneous formats. :

[0070]

[0071] in, The system includes a pre-built external data processing toolset action space (containing Lagrange interpolation scripts, mean filtering tools, Laplacian test outlier removal scripts, and a unit conversion module). The intrinsic evaluation function for assessing action rewards in a large model is denoted by argmax(), which represents the value of action rewards in the action space. Selecting from the options makes the evaluation function Optimal action handling when the maximum value is obtained.

[0072] After multiple rounds of iterative interaction between thought and action, the perceptual agent invokes external tools to clean the initial perceptual data. The system calculates the reconstructed information entropy between the initial noisy dataset and the currently cleaned dataset in real time, until the information entropy loss function is reached. Below the preset refinement threshold :

[0073]

[0074] in, This represents the total number of samples in this batch of the perception dataset. For the first One data sample; This represents the prior true probability distribution of this type of perceived data under a noise-free ideal state; This indicates that the perceived data has gone through a historical action sequence. The reconstruction probability distribution that appears after cleaning and transformation.

[0075] After the iteration, the perceptual agent integrates the cleaned multi-source activity data with environmental monitoring data, and outputs a standardized, multimodal, spatiotemporally aligned state feature representation. .

[0076] S3. Carbon emission collaborative reasoning: The large model intelligent agent system performs collaborative reasoning on the state feature representation to assess the carbon emission status of the current construction stage and generate carbon emission diagnostic results.

[0077] Before performing the carbon emission collaborative reasoning in step S3, the system pre-constructs a knowledge graph in the field of construction carbon emissions as a priori constraint. Specifically, it acquires national and local carbon emission standards, carbon emission factor databases, construction quality acceptance specifications, engineering quotas and market price databases, and historical project management files for the construction industry.

[0078] like Figure 3 As shown, this embodiment adopts an activity-centric ontology design to construct a knowledge graph ontology structure for the field of construction carbon emissions. Specifically, construction procedures, construction materials, machinery and equipment, energy and water resources, logistics and transportation, construction specifications and quality points, historical emission reduction cases, and schedule plans are extracted as core concept node entities.

[0079] Semantic relationships between entities are extracted as edges to form a star-shaped topology network with "construction procedures" at its core. Specifically, "construction procedures" consume "construction materials" and use "machinery and equipment"; "schedule plans" and "construction specifications and quality points" constrain "construction procedures"; "historical emission reduction cases" guide "construction procedures"; "machinery and equipment" consume "energy and water resources"; and "construction materials" involve "logistics and transportation".

[0080] Meanwhile, the system attaches multi-dimensional constraint attributes to each node entity, specifically including: material carbon emission factor (attached to construction materials), transportation carbon emission factor (attached to logistics and transportation), energy and water resource carbon emission factor (attached to energy and water resources), benchmark energy consumption (attached to machinery and equipment), rated power (attached to machinery and equipment), standard quota consumption (attached to construction procedures), benchmark unit price (attached to construction materials), default transportation distance (attached to logistics and transportation), benchmark construction period node (attached to the schedule), cost budget (attached to construction procedures), carbon emission budget (attached to construction procedures), and optimization strategy (attached to historical emission reduction cases).

[0081] The entities, attributes, and relationships are mapped to a graph database (such as Neo4j) using an attribute graph model to generate a knowledge graph for the construction carbon emissions domain. This is formally represented as an attribute graph model. ,in For a set of node entities, Let be the set of edges. It is a set of attributes. For any node... Its attribute set can be represented as For example, the attribute set of construction material nodes includes the material's carbon emission factor and benchmark unit price.

[0082] To address the alignment issue between non-standard names at construction sites and standard entities in the knowledge graph, the system employs a joint distance metric of graph attention network and semantic vector space when accessing the graph. When inputting a query entity with colloquial features specific to the construction site... At that time, the system calculates its relationship with candidate entities in the graph. Comprehensive semantic similarity matching score :

[0083]

[0084] in, and These are high-dimensional semantic word vectors of entities generated using large model encoding. For two entities in a knowledge graph The shortest topological path distance in the graph structure. and To adjust the hyperparameters of the weights, the system selects... The entity node with the highest score is used to accurately retrieve and extract matching attributes and rule constraints.

[0085] Subsequently, a computational agent is used to perform collaborative carbon emission reasoning to calculate the actual carbon emissions. To avoid biases caused by direct computation from a large language model, the computational agent represents the state features output from step S2. Transformed into query entities, knowledge graph in the field of construction carbon emissions In the process, retrieve the precise carbon emission factor attribute value corresponding to the queried entity. And through the Tool-Calling mechanism, an external calculation module with a pre-set carbon quantization accounting formula is triggered to execute the following multi-source weighted quantization algebraic model:

[0086]

[0087]

[0088]

[0089]

[0090] in, This represents the total actual carbon emissions during the current construction phase. This refers to the carbon emissions inherent in the materials. For the first Actual consumption data of various building materials The corresponding material carbon emission factor; Carbon emissions from the operation of construction machinery and temporary facilities. and The first The machine's operating time and rated power, Corresponding to carbon emission factors for energy and water resources; Carbon emissions from logistics and transportation For transport load capacity, For the trajectory transportation distance, This is the carbon emission factor for turnover. After calculation, the external calculation module returns the numerical results to the accounting agent as the basis for carbon emission diagnosis and assessment.

[0091] S4. Carbon emission decision issuance: Based on the carbon emission diagnosis results, the large model intelligent agent system generates corresponding carbon emission control decisions and issues the control decisions to the corresponding construction management terminal or IoT control device.

[0092] When executing step S4, the decision-making agent uses the Chain of Thought reasoning mechanism to model the emission reduction scheduling task as a nonlinear multi-objective optimization problem with multiple stringent engineering constraints.

[0093] The decision-making agent first obtains the project schedule, project cost budget, construction quality standards, and preset carbon emission budget targets corresponding to the current construction progress. The actual total carbon emissions output by S3 are then processed. Compare with the carbon emission budget target to determine if there is a budget overrun, and generate corresponding overrun attribution entities (e.g., excessive energy consumption of specific equipment, excessive loss coefficient of a certain material).

[0094] In the knowledge graph of carbon emissions in the construction field, the decision-making agent retrieves related entities such as construction specifications, historical emission reduction cases, and quality control points that match the entity attributing the excessive emissions, based on the association relationship of the attribute graph.

[0095] Guided by thought chain prompts, the large language model transforms the quality control points and project schedule nodes retrieved from the knowledge graph into rigid boundary conditions, and defines the emission reduction scheduling strategy as a set of decision variables. (Corresponding to specific material replacement ratios, mechanical operation timing adjustment parameters, etc.), the following multi-objective optimization control model is constructed:

[0096]

[0097]

[0098] in, and These are the estimated carbon emissions and engineering construction costs after implementing the strategy, respectively. and The pre-set carbon emission budget and construction cost budget, and These are the decision weighting coefficients; The actual estimated duration after implementing the scheduling strategy, The project duration is the planned schedule. This is a quality assessment value. The baseline set for quality acceptance specifications.

[0099] The decision-making agent utilizes particle swarm optimization to perform multi-objective collaborative optimization in thought chain deduction. The strategy particle is in the... Speed ​​in the next iteration With position The updated formula is:

[0100]

[0101]

[0102] in, For inertial weights, , As a learning factor, , It is a random number. This represents the optimal strategy experienced by the particle. The global Pareto optimal strategy is found for the community. The decision-making agent sequentially outputs intermediate reasoning steps (such as first performing an economic assessment of material substitution, then performing time-series optimization of machine scheduling), and finally generates emission reduction scheduling instructions that take into account multiple objective constraints. The control decisions are output and distributed to the corresponding construction management terminal or IoT control device to achieve closed-loop control.

[0103] Example 2: Carbon Emission Control System for Building Construction Based on Large-Scale Intelligent Agents

[0104] Combination Figure 2 This invention also provides a building construction carbon emission control system based on a large-scale intelligent agent model, which corresponds to the method in Embodiment 1. The system specifically includes:

[0105] 1. IoT Sensing Module: A sensor array and data acquisition instrument built into the construction site to acquire multi-source IoT sensing data such as energy and water consumption data, material consumption data, dynamic load and trajectory data of logistics transportation, and environmental monitoring data of equipment and temporary facilities at the construction site;

[0106] 2. Feature Conversion Module: An edge computing gateway deployed at the construction site, including an industrial protocol parsing module, is used to perform preliminary local format conversion on the multi-source IoT sensing data, output a unified structured JSON data stream, and convert it into a standardized state feature representation;

[0107] 3. Knowledge Graph Storage Module: Equipped with a graph database engine (such as Neo4j), it stores a knowledge graph in the field of construction carbon emissions, including construction standards, carbon emission factors, quality specifications, engineering quotas and historical cases, and is configured with a graph attention network entity alignment interface;

[0108] 4. Intelligent Agent Collaboration Module: This module incorporates a large-scale intelligent agent system as the core control hub. It comprises a perception agent, a computational agent, and a decision-making agent. The perception agent uses the ReAct mechanism for logical cleaning of IoT perception data; the computational agent uses a tool-calling mechanism to trigger external quantitative computation; and the decision-making agent uses a thought chain mechanism combined with knowledge graph constraints to execute control decision reasoning under multi-dimensional engineering constraints.

[0109] 5. Quantization Calculation Module: Communicatively connected to the accounting agent, independent of the large model agent system, it has a pre-set deterministic multi-source weighted carbon emission quantification algebraic formula, used to receive parameter instructions from the accounting agent and perform precise numerical quantification calculations to eliminate the numerical illusion of the large model;

[0110] 6. Decision Execution Module: Includes API gateway and control interface for sending emission reduction scheduling instructions generated by decision intelligence to the corresponding construction management terminal (such as APP terminal, PC terminal) or on-site IoT control device (such as intelligent power distribution cabinet, heavy machinery control gateway) in real time.

[0111] Example 3: Electronic Equipment

[0112] This invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor.

[0113] The processor can be a central processing unit (CPU), or a graphics processing unit (GPU), digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device. Memory, as a carrier of computer-readable program code, can be used to store computer programs and / or functional modules. Memory can include high-speed random access memory, and can also include non-volatile memory, such as one or more disk storage devices, flash memory devices, or other non-volatile solid-state storage devices.

[0114] The processor implements the various steps of the building construction carbon emission control method based on large model intelligent agents described in Embodiment 1 above by running or executing computer programs and / or modules stored in memory, and by calling data stored in memory.

[0115] Example 4: Computer-readable storage medium

[0116] This invention also provides a computer-readable storage medium storing a computer program thereon. The computer-readable storage medium can be a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, or any other non-volatile or volatile storage medium capable of storing program code. When the computer program is executed by a processor, it implements the various steps of the building construction carbon emission control method based on a large-scale intelligent agent described in Embodiment 1 above.

Claims

1. A method for controlling carbon emissions from building construction based on a large-scale intelligent agent model, characterized in that, The method is based on a large model intelligent agent system driven by a large language model kernel, which shares the semantic understanding and logical reasoning capabilities of the large language model kernel. The method includes the following steps: S1. Acquire multi-source IoT sensing data at the construction site, wherein the sensing data includes at least: energy and water consumption data of equipment and temporary facilities, material consumption data, dynamic load and trajectory data of logistics transportation, and environmental monitoring data; S2. Convert the multi-source IoT sensing data into a standardized state feature representation and input it into a preset large model intelligent agent system; S3. Through the large model intelligent agent system, perform collaborative reasoning on the state feature representation, assess the carbon emission status of the current construction stage, and generate carbon emission diagnostic results. S4. Based on the carbon emission diagnosis results, the large model intelligent agent system generates corresponding carbon emission control decisions and sends the control decisions to the corresponding construction management terminal or IoT control device.

2. The method for controlling carbon emissions from building construction based on a large-scale intelligent agent as described in claim 1, characterized in that, The large-scale model intelligent agent system includes a perceptual agent, a computational agent, and a decision-making agent; all three agents are driven by the underlying kernel of the large language model; wherein: The sensing agent is used to call the underlying large language model, and uses the ReAct mechanism to perform logical cleaning and multimodal feature alignment on the multi-source IoT sensing data, and generate the state feature representation. The accounting agent is used to call the underlying large language model and adopts a tool-calling-based computational reasoning mechanism to extract energy and water consumption data, material consumption data, and dynamic load and trajectory data of logistics transportation from the state feature representation of equipment and temporary facilities, and calculates and outputs the actual carbon emissions. The decision-making agent calls upon the underlying large language model, combines the current construction progress, project cost constraints, construction quality standards and preset carbon emission budget, performs multi-objective deviation comparison analysis on the actual carbon emissions, and generates specific emission reduction scheduling instructions.

3. The method for controlling carbon emissions from building construction based on a large-scale intelligent agent as described in claim 1, characterized in that, When performing multi-source IoT sensing in step S1, the multi-source IoT sensing data is acquired in real time through the data acquisition interface deployed on the edge computing gateway at the construction site, and the initial format conversion is completed locally on the edge computing gateway.

4. The method for controlling carbon emissions from building construction based on a large-scale intelligent agent as described in claim 2, characterized in that, When the perceptual agent performs the standardized feature transformation in step S2 using the ReAct mechanism, the specific steps include: S2.1 Receive the discrete multi-source IoT sensing data as initial observations; S2.2, Call the large language model to analyze missing values, outliers and heterogeneous formats in the data, and generate corresponding logical processing ideas; S2.

3. Generate data cleaning and alignment actions based on the aforementioned logical processing thinking, trigger external data processing tools to perform transformation, and iterate this process until the aligned state feature representation is output.

5. The method for controlling carbon emissions from building construction based on a large-scale intelligent agent as described in claim 2, characterized in that, Prior to performing the carbon emission collaborative reasoning in step S3, the process also includes the step of constructing a knowledge graph in the field of construction carbon emissions: S3A.1, National and local carbon emission standards for the construction industry, carbon emission factor database, construction quality acceptance specifications, engineering quota and market price database, and historical project management archives; S3A.2 Extract the construction procedures, construction materials, machinery and equipment, energy and water resources, logistics and transportation, construction specifications and quality points, historical emission reduction cases and schedule as graph node entities, extract the corresponding material carbon emission factor, transportation carbon emission factor, energy and water resource carbon emission factor, benchmark energy consumption, rated power, standard quota consumption, benchmark unit price, default transportation distance, benchmark construction period node, cost budget, carbon emission budget and optimization strategy as attributes of the node entities, and extract the semantic association between each entity as relations; S3A.

3. The entities, attributes, and relationships are mapped to the graph database using an attribute graph model to generate the knowledge graph of the construction carbon emission domain.

6. The method for controlling carbon emissions from building construction based on a large-scale intelligent agent as described in claim 5, characterized in that, When the accounting agent performs the carbon emission collaborative reasoning in step S3 to calculate the actual carbon emissions, it includes the following steps: S3B.

1. Convert the state feature representation into a query entity; S3B.

2. In the pre-constructed knowledge graph of construction carbon emissions, retrieve the carbon emission factor attribute value corresponding to the query entity; S3B.

3. Trigger an external calculation module with a pre-set carbon quantification calculation formula through a tool call mechanism. The external calculation module performs quantification calculations with the corresponding carbon emission factors based on the energy consumption data, material and water resource consumption data, and dynamic load and trajectory data of logistics transportation in the state feature representation, respectively, to obtain the actual carbon emissions.

7. The method for controlling carbon emissions from building construction based on a large-scale intelligent agent according to claim 5, characterized in that, When the decision-making agent issues the carbon emission decision in step S4, it uses a Chain of Thought reasoning mechanism: S4.1 Obtain the schedule plan, cost budget, quality control indicators and carbon emission budget targets corresponding to the current construction progress; S4.

2. Compare the actual carbon emissions with the carbon emission budget target, and combine the environmental monitoring data in the state feature representation to determine whether there is a budget overrun, and generate the corresponding overrun attribution entity; S4.3 In the knowledge graph of carbon emissions in the construction field, based on the association relationship of the graph structure, retrieve the construction specifications and quality points entities and historical emission reduction case entities that match the excessive attribution entity; S4.4 Combining the aforementioned construction specifications, quality points, and historical emission reduction cases, a thought chain reasoning mechanism is used to decompose the carbon control task into sub-tasks such as material replacement assessment, machinery scheduling optimization, and process improvement. Under the premise of satisfying the multi-objective constraints of the aforementioned schedule plan, cost budget, and quality control indicators, intermediate reasoning steps are output sequentially, and finally the emission reduction scheduling instruction is generated.

8. A carbon emission control system for building construction based on a large-scale intelligent agent, applied to the carbon emission control method for building construction based on a large-scale intelligent agent as described in claims 1-7, characterized in that, include: The IoT sensing module is used to acquire multi-source IoT sensing data at the construction site. The feature conversion module is used to convert the multi-source IoT sensing data into a standardized state feature representation; The knowledge graph storage module is used to store knowledge graphs in the field of construction carbon emissions, including construction procedures, construction materials, machinery and equipment, energy and water resources, logistics and transportation, construction specifications and quality points, historical emission reduction cases, and progress plan entities. The intelligent agent collaboration module has a built-in large model intelligent agent system, which includes a perception intelligent agent, an accounting intelligent agent, and a decision-making intelligent agent. It is used to perform perception cleaning, carbon emission quantification and accounting, and multi-objective management and control decision reasoning that combines progress, cost, quality and budget on the state feature representation. The quantization calculation module is communicatively connected to the computing agent and is used to receive calculation instructions and perform numerical quantization calculations. The decision execution module is used to issue emission reduction scheduling instructions to the corresponding construction management terminal or IoT control device.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the building construction carbon emission control method based on a large model intelligent agent as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the carbon emission control method for building construction based on a large model intelligent agent as described in any one of claims 1 to 7.

Images