Multi-agent collaborative management method and system for whole-process engineering consultation

Abstract

The application provides a multi-agent collaborative management method and system for whole-process engineering consultation, which comprises the following steps: S1, obtaining a whole-process engineering consultation task; S2, performing task decomposition on the obtained whole-process engineering consultation task based on an LLM large language model to obtain task decomposition features of the whole-process engineering consultation task, wherein the task decomposition features comprise one or more subtasks; S3, respectively performing task classification feature extraction on each subtask of the task decomposition features to obtain a task classification label corresponding to each subtask; S4, according to the task classification label corresponding to each subtask, matching a corresponding agent from candidate agents to process the subtask, and obtaining a subtask solution output by the agent; and S5, performing optimization processing according to the obtained subtask solution to obtain a whole-process engineering consultation task solution. The application helps to improve the effect of the whole-process engineering consultation task solution.

Description

A Multi-Agent Collaborative Management Method and System for Full-Process Engineering Consulting Technical Field

[0001] This invention relates to the field of engineering consulting management technology for the entire process of construction projects, and in particular to a multi-agent collaborative management method and system for engineering consulting throughout the entire process. Background Technology

[0002] With the deepening of digital transformation in the construction industry, the application of large language model technology has greatly promoted the development of artificial intelligence technology and driven the deep integration of artificial intelligence technology with engineering management application systems. Currently, engineering management application systems built on large language models use trained large language models as the core component of the artificial intelligence system, combining them with intelligent agents to perform reasoning on natural language and output reasoning conclusions. However, in the implementation of full-process engineering consulting services, it is impossible to achieve task division and collaborative management among multiple intelligent agents for complex multi-objective tasks. This results in the inability to obtain the standardized processing flow and accurate reasoning results required for full-process engineering consulting services, significantly impacting the accuracy of the output reasoning conclusions. Therefore, this invention provides a large-model multi-agent collaborative management method and system technical solution for full-process engineering consulting. Summary of the Invention

[0003] To address the aforementioned problems, this invention aims to provide a multi-agent collaborative management method and system for end-to-end engineering consulting.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] Firstly, this invention proposes a multi-agent collaborative management method for end-to-end engineering consulting, comprising:

[0006] S1 obtains the full-process engineering consulting task, wherein the full-process engineering consulting task is natural language input by the user;

[0007] S2 decomposes the acquired full-process engineering consulting tasks based on the LLM large language model, and obtains the task decomposition features of the full-process engineering consulting tasks, wherein the task decomposition features include one or more sub-tasks.

[0008] S3 extracts task classification features from each subtask of the task decomposition features and obtains the task classification labels corresponding to the subtasks.

[0009] S4 matches the appropriate agent from the candidate agents to process the sub-tasks based on the task classification labels corresponding to each sub-task, and obtains the sub-task solutions output by the agent.

[0010] S5 optimizes the obtained sub-task solutions to obtain a full-process engineering consulting task solution.

[0011] Preferably, the method further includes:

[0012] Based on the obtained full-process engineering consulting task solution, S6 performs prediction verification based on the digital twin model of the engineering project, obtains the solution verification results, and further trains and optimizes the relevant intelligent agents according to the solution verification results.

[0013] Preferably, step S2 includes:

[0014] Based on the LLM (Limited Language Model), the acquired full-process engineering consulting tasks are understood and divided, and the conditional features and target tasks of the full-process engineering consulting tasks are extracted. Further enhanced retrieval is performed based on the acquired conditional features and target tasks to obtain the corresponding potential knowledge requirements. Based on the acquired potential knowledge requirements, the LLM is further used to understand them and obtain potential tasks. The target tasks and potential tasks are respectively regarded as sub-tasks to form the task decomposition features of the full-process engineering consulting tasks. Among them, the conditional features include engineering specifications and constraints.

[0015] Preferably, step S3 includes:

[0016] Extract text feature vectors based on the descriptions of the subtasks and the corresponding engineering project feature information. Matching based on text feature vectors Task category tags corresponding to subtasks ,in , Indicates task category tags, This represents the set of task category tags. Represents text feature vectors The probability that it belongs to task category label j.

[0017] Preferably, step S4 includes:

[0018] Based on the task category tags corresponding to the sub-tasks, the corresponding expert agents are matched to process the sub-tasks and obtain sub-task solutions. Among them, expert agents include investment decision consulting agents, engineering approval and construction service agents, engineering survey management agents, engineering design management agents, engineering bidding and procurement management agents, construction project management agents, engineering supervision agents, engineering cost management agents, engineering operation and maintenance management agents, other consulting service agents, quality management agents, safety management agents, schedule management agents, cost management agents, and other target management agents.

[0019] Preferably, step S5 includes:

[0020] For a subtask solution obtained by a single expert agent, extract key terms information from the subtask solution.

[0021] Based on the corresponding engineering specifications, a compliance analysis is performed on the key clause information in the task solution. The variable content in the key clause is compared with the corresponding standard value range in the engineering specifications. When the key clause information exceeds the standard range, the content in the sub-task solution corresponding to the key clause information is adaptively corrected, and the corresponding variable is adjusted to the value closest to the current value and in line with the standard range, so that the sub-task solution complies with the engineering specifications.

[0022] Preferably, step S5 includes:

[0023] Subtask solutions obtained from multiple expert agents:

[0024] For the sub-task solutions of the target task, extract key clause information from the sub-task solutions. ;

[0025] For each potential task's sub-task solutions, extract key terms information from the sub-task solutions. ;in As variables, , representing the key terms information corresponding to the subtask solution of the nth potential task. Indicates the total number of potential tasks;

[0026] Based on the target task, calculate the impact variation parameters of each subtask on the target task. :

[0027]

[0028] In the formula, The parameter represents the change in impact on the target task, where k is a variable. Each corresponds to a different potential task. That is, the total number of potential tasks. This represents the expert agent corresponding to subtask k; Task category labels indicating the target task; Expert intelligent agents Based on task classification The processing power of the agent is determined by statistical analysis of its performance in task classification. The success rate statistics are as follows; This indicates the key terms information for the corresponding subtask k. Effect on the target task The partial derivative of is used to characterize the degree of influence of key clause information on the effectiveness of the target task.

[0029] Will affect the changing parameters As a target value, adjust the key terms of the current subtask solution until they affect the changing parameters. Less than the set standard threshold , where Z represents the set standard threshold, or the maximum number of adjustments; thus, the optimized subtask solutions for each potential task are obtained;

[0030] By integrating the sub-task solutions of the target task with the optimized sub-task solutions of the potential tasks, a comprehensive engineering consulting task solution is obtained.

[0031] Preferably, step S6 includes:

[0032] The obtained full-process engineering consulting task solution is input into the established BIM-based full-process engineering consulting management model to perform performance analysis and risk analysis, obtain the corresponding performance analysis results and risk analysis results, and record the adaptability and adoption of the task solution to the target task based on the obtained performance analysis results and risk analysis results.

[0033] Secondly, this invention proposes a multi-agent collaborative management system for end-to-end engineering consulting, comprising a receiving module, a processing module, and an output module, wherein...

[0034] The receiving module is used to receive full-process engineering consulting tasks transmitted by users;

[0035] The processing module is used to execute the multi-agent collaborative management method for full-process engineering consulting as shown in any of the embodiments in the first aspect, to obtain a full-process engineering consulting task solution.

[0036] The output module is used to output the solution for the entire process engineering consulting task.

[0037] The beneficial effects of this invention are as follows: It proposes a multi-agent collaborative management method and system for full-process engineering consulting. First, the full-process engineering consulting task is acquired. Based on an LLM (Large Language Model), the task is understood and decomposed. Based on the decomposed sub-tasks, corresponding task classification tags are extracted to label the type of each sub-task. According to the task classification tags, appropriate agents are matched to process the sub-tasks, resulting in targeted sub-task solutions. These solutions are then further optimized to ultimately generate a solution specifically for the engineering consulting task. By using task decomposition, classification tag extraction, and agent matching, targeted processing of corresponding task types based on dedicated agents can be achieved, improving the professionalism of full-process engineering consulting task processing. For sub-task solutions obtained from multiple agents, targeted joint optimization is performed, improving the compatibility of different sub-task solutions and thus enhancing the adaptability and overall performance of the final full-process engineering consulting task solution. Attached Figure Description

[0038] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0039] Figure 1 is a flowchart of a multi-agent collaborative management method for full-process engineering consulting according to an embodiment of the present invention;

[0040] Figure 2 is a framework diagram of a multi-agent collaborative management system for full-process engineering consulting, as shown in an embodiment of the present invention. Detailed Implementation

[0041] The present invention will be further described in conjunction with the following application scenarios.

[0042] Referring to Figure 1, which illustrates a multi-agent collaborative management method for end-to-end engineering consulting, including:

[0043] S1 obtains the full-process engineering consulting task, wherein the full-process engineering consulting task is natural language input by the user;

[0044] S2 decomposes the acquired full-process engineering consulting tasks based on the LLM large language model, and obtains the task decomposition features of the full-process engineering consulting tasks, wherein the task decomposition features include one or more sub-tasks.

[0045] S3 extracts task classification features from each subtask of the task decomposition features and obtains the task classification labels corresponding to the subtasks.

[0046] S4 matches the appropriate agent from the candidate agents to process the sub-tasks based on the task classification labels corresponding to each sub-task, and obtains the sub-task solutions output by the agent.

[0047] S5 optimizes the obtained sub-task solutions to obtain a full-process engineering consulting task solution.

[0048] The above embodiments of the present invention propose a multi-agent collaborative management method for full-process engineering consulting. First, the full-process engineering consulting task is acquired. Then, based on an LLM (Large Language Model), the task is understood and decomposed. Based on the decomposed sub-tasks, corresponding task classification tags are extracted to label the type of each sub-task. According to the task classification tags, appropriate agents are matched to process the sub-tasks, resulting in targeted sub-task solutions. These solutions are then further optimized to ultimately generate a solution specifically for the engineering consulting task. By using task decomposition, classification tag extraction, and agent matching, targeted processing of corresponding task types based on dedicated agents can be achieved, improving the professionalism of full-process engineering consulting task processing. For sub-task solutions obtained from multiple agents, targeted joint optimization is performed, improving the compatibility of quality inspection between different sub-task solutions, thereby enhancing the adaptability and overall performance of the final full-process engineering consulting task solution.

[0049] The multi-agent collaborative management method for full-process engineering consulting described above can be executed and implemented based on servers, smart terminals, etc., and finally output a full-process engineering consulting task solution.

[0050] Preferably, the method further includes:

[0051] Based on the obtained full-process engineering consulting task solution, S6 performs prediction verification based on the digital twin model of the engineering project, obtains the solution verification results, and further trains and optimizes the relevant intelligent agents according to the solution verification results.

[0052] Based on the full-process engineering consulting task solution output by the system, further performance testing and risk analysis are conducted using digital twin technology and predictive analytics to further verify the feasibility and execution effectiveness of the full-process engineering consulting task solution. The prediction results are then used as a reference for further training and setting of the intelligent agent, thereby further improving the performance and effectiveness of the system and the intelligent agent.

[0053] The intelligent agent shown in this invention generally refers to an expert intelligent agent set up for different knowledge categories in the whole process engineering consulting scenario. The expert intelligent agent can be trained based on knowledge and historical data in one or more fields, thereby improving the ability and effectiveness of the expert intelligent agent in handling consulting tasks in a specific field.

[0054] Preferably, step S2 includes:

[0055] Based on the LLM (Limited Language Model), the acquired full-process engineering consulting tasks are understood and divided, and the conditional features and target tasks of the full-process engineering consulting tasks are extracted. Further enhanced retrieval is performed based on the acquired conditional features and target tasks to obtain the corresponding potential knowledge requirements. Based on the acquired potential knowledge requirements, the LLM is further used to understand them and obtain potential tasks. The target tasks and potential tasks are respectively regarded as sub-tasks to form the task decomposition features of the full-process engineering consulting tasks. Among them, the conditional features include engineering specifications and constraints.

[0056] Considering the numerous engineering specifications and standards involved in full-process engineering consulting for construction projects, it is typically necessary to simultaneously consider specifications from different fields when analyzing and processing tasks in a single domain. Therefore, in the above implementation method, when understanding and decomposing the acquired full-process engineering consulting task, the engineering specification information corresponding to the full-process engineering consulting task is first extracted. Enhanced retrieval is then performed based on the engineering specification information, task conditions, and task objectives to determine the potential conditions and potential tasks that must be simultaneously met to achieve the target task under the corresponding engineering specifications. This allows for the expansion and decomposition of the target task into corresponding sub-tasks. This ensures that when generating solutions subsequently, in addition to considering the target task itself, the potential tasks are also taken into account as evaluation criteria, thereby improving the overall quality of the subsequent solutions.

[0057] In one scenario, the current full-process engineering consulting task is understood based on the LLM large language model. This involves extracting engineering project feature information based on the corresponding engineering project for the full-process engineering consulting task (if there is no corresponding engineering project, then the preset general engineering project feature information is used for general full-process engineering consulting tasks), matching the corresponding engineering specifications (such as standards with GB numbers) based on the engineering project feature information, and further performing enhanced retrieval based on the matched engineering specifications, constraints, and target tasks to obtain the potential knowledge requirements that need to be met simultaneously to complete the target task under the constraints. Based on the potential knowledge requirements in the standard knowledge, further understanding and task extraction are performed to obtain potential task information.

[0058] In one scenario, for example, regarding a full-process engineering consulting task of "providing a solution for completing office building construction within budget," the constraint obtained after understanding using the LLM (Limited Language Model) is "within budget," and the objective task is "to complete office building construction within budget." Further, based on the corresponding engineering project, suitable engineering specification information is matched from the knowledge base. Enhanced retrieval is performed on the engineering specification information based on the constraint and objective task to obtain potential knowledge requirements, such as "According to Article z of GB505xx-yyyy, to complete office building construction within budget, the change order rate must be controlled to <5%." Then, based on these potential knowledge requirements, further understanding using the LLM (based on relevant engineering specifications for controlling change order rates) and task division yields the potential tasks for controlling change order rates as "developing a reasonable schedule plan to reduce the change order rate" and "implementing reliable quality control strategies to reduce the change order rate." These are then concatenated with the objective task to obtain the task decomposition features of three corresponding sub-tasks.

[0059] By extracting potential tasks in the above manner, the large language model can compensate for the lack of a comprehensive understanding of "budget" and the need to extend that understanding to "quality" and "planning," thereby improving the accuracy and rationality of understanding and confirming the entire process of engineering consulting tasks.

[0060] Preferably, step S3 includes:

[0061] Extract text feature vectors based on the descriptions of the subtasks and the corresponding engineering project feature information. Matching based on text feature vectors Task category tags corresponding to subtasks ,in , Indicates task category tags, This represents the set of task category tags. Represents text feature vectors The probability that it belongs to task category label j.

[0062] The text feature vectors can be extracted from the description content of the subtask and the corresponding engineering project feature information through text neural network models such as Word2Vec, or through deep learning models such as BERT, or through other existing text feature models. This invention does not make any specific limitations on these methods.

[0063] The task classification tags include task classification information related to the whole process of engineering management, including investment decision consulting, engineering approval and construction services, engineering survey management, engineering design management, engineering bidding and procurement management, construction project management, engineering supervision, engineering cost management, engineering operation and maintenance management, other consulting services, quality management, safety management, schedule management, cost management, and other target management.

[0064] Among them, text feature vectors The probability of belonging to task category label j can be processed based on a trained semantic feature classifier. The semantic feature classifier can be implemented based on classification models such as SVM (Support Vector Machine) or based on the dedicated classification Transformer architecture in the LMM (Large Language Model) to complete the task of text feature vector classification.

[0065] Based on the extracted text feature vectors from the obtained sub-task content, and further matching the corresponding task classification labels based on these text feature vectors, the foundation for subsequently assigning corresponding agents to the sub-tasks can be laid. In particular, classifying sub-tasks through task classification allows for segmentation based on the agent's domain-specific focus on the entire process of engineering consulting and management, thereby improving the generation effect of sub-task solutions.

[0066] Preferably, step S4 includes:

[0067] Based on the task category tags corresponding to the sub-tasks, the corresponding expert agents are matched to process the sub-tasks and obtain sub-task solutions. Among them, expert agents include investment decision consulting agents, engineering approval and construction service agents, engineering survey management agents, engineering design management agents, engineering bidding and procurement management agents, construction project management agents, engineering supervision agents, engineering cost management agents, engineering operation and maintenance management agents, other consulting service agents, quality management agents, safety management agents, schedule management agents, cost management agents, and other target management agents.

[0068] Based on the specific knowledge domains involved in the whole-process engineering consulting scenario, corresponding expert intelligent agents can be set up for specific domains to solve corresponding tasks and provide more targeted solutions.

[0069] In another scenario, expert agents can be configured according to their domain, such as cost agents, safety agents, schedule agents, and quality agents, to meet the needs of specialized task solutions.

[0070] Preferably, step S5 includes:

[0071] For a subtask solution obtained by a single expert agent, extract key terms information from the subtask solution.

[0072] Based on the corresponding engineering specifications, a compliance analysis is performed on the key clause information in the task solution. The variable content in the key clause is compared with the corresponding standard value range in the engineering specifications. When the key clause information exceeds the standard range, the content in the sub-task solution corresponding to the key clause information is adaptively corrected, and the corresponding variable is adjusted to the value closest to the current value and in line with the standard range, so that the sub-task solution complies with the engineering specifications.

[0073] Among them, the key terms in the sub-task solutions can be extracted through keyword retrieval and matching, entity recognition combined with knowledge base, and other methods.

[0074] In one scenario, for example, in a solution for a sub-task related to engineering quality acceptance, the key clauses extracted include "Clause 1: Concrete strength testing ≥ 3 times / layer; Clause 2: Rebar spacing error ≤ 5mm"; for example, in a solution related to budget, the key clauses extracted include "Prepare alternative solutions in case of market price fluctuations exceeding 5%" and "Control the design change rate within 2%".

[0075] In one scenario, the variable content in the key terms information can be a numerical value (such as a specific number) or a trend (such as increase, decrease, or maintenance).

[0076] After obtaining a task solution from a single expert agent, key clauses are extracted from the solution. These key clauses allow for the summarization and generalization of sub-task solutions, improving the accuracy of information extraction. Based on these extracted key clauses, a further comparison is made between the clauses and relevant engineering specifications to ensure compliance, thereby enhancing the feasibility and compliance of the task solution.

[0077] Preferably, step S5 includes:

[0078] Subtask solutions obtained from multiple expert agents:

[0079] For the sub-task solutions of the target task, extract key clause information from the sub-task solutions. ;

[0080] For each potential task's sub-task solutions, extract key terms information from the sub-task solutions. ;in As variables, , representing the key terms information corresponding to the subtask solution of the nth potential task. Indicates the total number of potential tasks;

[0081] Based on the target task, calculate the impact variation parameters of each subtask on the target task. :

[0082]

[0083] In the formula, The parameter represents the change in impact on the target task, where k is a variable. Each corresponds to a different potential task. That is, the total number of potential tasks. This represents the expert agent corresponding to subtask k; Task category labels indicating the target task; Expert intelligent agents Based on task classification The processing power of the agent is determined by statistical analysis of its performance in task classification. The success rate statistics are as follows; This indicates the key terms information for the corresponding subtask k. Effect on the target task The partial derivative of is used to characterize the degree of influence of key clause information on the effectiveness of the target task.

[0084] Will affect the changing parameters As a target value, adjust the key terms of the current subtask solution until they affect the changing parameters. Less than the set standard threshold , where Z represents the set standard threshold, or the maximum number of adjustments; thus, the optimized subtask solutions for each potential task are obtained;

[0085] By integrating the sub-task solutions of the target task with the optimized sub-task solutions of the potential tasks, a comprehensive engineering consulting task solution is obtained.

[0086] In particular, when calculating the success rate of an agent in handling a certain type of task, it can be obtained by statistically analyzing the final acceptance rate (probability, such as the statistical number of final acceptances / total number of processing) of the task solutions provided by the agent when handling collaborative tasks involving a certain type of task.

[0087] The methods for adjusting key terms include changing the plan, deleting clauses, and adding conditions.

[0088] The methods for integrating the sub-task solutions of the target task with the optimized sub-task solutions of the potential task include splicing, overlaying, weighted overlaying, and fusion based on feature similarity, etc., which are not specifically limited in this invention.

[0089] As mentioned in the previous solution, extracting potential tasks and obtaining corresponding sub-task solutions can improve the breadth and comprehensiveness of the solutions. However, since multiple sub-task solutions exist independently, it is still necessary to integrate the sub-task solutions.

[0090] Since the solutions for different sub-tasks are obtained by different expert agents, and different expert agents have different basic knowledge and thinking logic when considering tasks in different domains, the solutions obtained for potential tasks may conflict with or contradict the target task (e.g., contradictory solutions), resulting in unsatisfactory results in the final whole-process engineering consulting task solutions.

[0091] In the actual execution of the solution, the smaller the impact rate of other potential tasks on the result of the target task, the lower the contradiction caused by the potential solution to the target task. The primary task of the solution through potential tasks is to improve the stability of the target task (avoiding oversights). However, the effect of the target task can be improved by the solution of the target task itself (a solution specifically designed to improve the effect of the target). Based on the above ideas, the sub-task solutions can be integrated to improve the stability of the solution while ensuring the effect of the target task.

[0092] Therefore, the above embodiments of the present invention further propose a solution for collaborative optimization of sub-task solutions obtained by different expert agents, thereby improving the overall performance of the solution for the entire process engineering consulting task. First, sub-tasks typically include target tasks and potential tasks. Therefore, the target task is set as the highest priority for final consideration, while the solutions for potential tasks ultimately need to serve the target task. Based on this idea, key clause information of each potential task solution is extracted first. Based on the proposed impact change parameters, the degree of influence of the potential task solutions on the current target task is comprehensively reflected, and this is used as a target parameter to reduce the impact change parameters, avoiding interference from solutions provided by non-expert agents on the target task, thus helping to improve the stability of the solution. Finally, the solutions for the target task and the potential tasks are fused to obtain a more comprehensive and optimized solution (ensuring the effectiveness of the target task while improving the stability and adaptability of the provided solution), thereby improving the quality and effectiveness of the solution.

[0093] In one scenario, when the target task is a solution for "budget optimization," the solution provides relevant technical solutions for budget optimization. However, for potential tasks such as "planning" and "quality," if the provided solutions would cause significant fluctuations in the "budget" in exchange for corresponding "planning improvement" or "quality improvement," then those solutions should be weakened or discarded. This ensures that "budget is the top priority," while supplementing the solutions related to "planning" and "quality," thereby improving the overall stability of the solution (reducing the volatility of the impact on the budget caused by planning and quality).

[0094] Preferably, step S6 includes:

[0095] The obtained full-process engineering consulting task solution is input into the established BIM-based full-process engineering consulting management model to perform performance analysis and risk analysis, obtain the corresponding performance analysis results and risk analysis results, and record the adaptability and adoption of the task solution to the target task based on the obtained performance analysis results and risk analysis results.

[0096] By analyzing and recording fit and adoption rates, we can statistically analyze successful and unsuccessful cases of agents targeting task solutions, thereby collecting relevant success rate characteristic data, which facilitates further adjustments and optimizations to the agent's performance.

[0097] Referring to Figure 2, which illustrates a multi-agent collaborative management system for end-to-end engineering consulting, the system includes a receiving module, a processing module, and an output module.

[0098] The receiving module is used to receive full-process engineering consulting tasks transmitted by users;

[0099] The processing module is used to execute the multi-agent collaborative management method for full-process engineering consulting as shown in Figure 1 above, and obtain a solution for full-process engineering consulting tasks;

[0100] The output module is used to output the solution for the entire process engineering consulting task.

[0101] It should be noted that the processing module is also used to execute the specific implementation methods corresponding to each step in Figure 1, which will not be repeated here.

[0102] It should be noted that the functional units / modules in the various embodiments of the present invention can be integrated into one processing unit / module, or each unit / module can exist physically separately, or two or more units / modules can be integrated into one unit / module. The integrated unit / module described above can be implemented in hardware or in the form of software functional units / modules.

[0103] From the above description of the embodiments, those skilled in the art will clearly understand that the embodiments described herein can be implemented in hardware, software, firmware, middleware, code, or any suitable combination thereof. For hardware implementation, the processor can be implemented in one or more of the following units: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field-Programmable Gate Array (FPGA), processor, controller, microcontroller, microprocessor, other electronic units designed to implement the functions described herein, or combinations thereof. For software implementation, some or all of the processes of the embodiments can be implemented by a computer program instructing the associated hardware. During implementation, the program can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of a computer program from one place to another. Storage media can be any available medium accessible to a computer. Computer-readable media can include, but is not limited to, RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible to a computer.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should be able to analyze that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. A multi-agent collaborative management method for full-process engineering consulting, characterized in that, include: S1 obtains the full-process engineering consulting task, wherein the full-process engineering consulting task is natural language input by the user; S2 decomposes the acquired full-process engineering consulting task based on the LLM large language model to obtain the task decomposition features of the full-process engineering consulting task, wherein the task decomposition features contain one or more sub-tasks; S3 extracts task classification features for each sub-task of the task decomposition features to obtain the task classification labels corresponding to the sub-tasks; S4 matches the corresponding intelligent agents from the candidate intelligent agents to process the sub-tasks according to the task classification labels corresponding to each sub-task, and obtains the sub-task solutions output by the intelligent agents; S5 optimizes the obtained sub-task solutions to obtain the full-process engineering consulting task solution; wherein, step S4 includes: matching the corresponding expert intelligent agents to process the sub-task according to the task classification labels corresponding to the sub-tasks to obtain the sub-task solutions; step S5 includes: for the sub-task solutions obtained by multiple expert intelligent agents: for the sub-task solutions of the target task, extracting key clause information from the sub-task solutions. For each potential task's sub-task solution, extract key terms information from the sub-task solutions. ;in As variables, , representing the key terms information corresponding to the subtask solution of the nth potential task. This represents the total number of potential tasks; based on the target task, calculate the parameter representing the change in the impact of each subtask on the target task. ： In the formula, The parameter represents the change in impact on the target task, where k is a variable. Each corresponds to a different potential task. That is, the total number of potential tasks. This represents the expert agent corresponding to subtask k; Task category labels indicating the target task; Expert intelligent agents Based on task classification The processing power of the agent is determined by statistical analysis of its performance in task classification. The success rate statistics are as follows; This indicates the key terms information for the corresponding subtask k. Effect on the target task The partial derivative is used to characterize the degree of influence of key clause information on the effectiveness of the target task; the influence of changing parameters is used to characterize the degree of influence of key clause information on the effectiveness of the target task. As a target value, adjust the key terms of the current subtask solution until they affect the changing parameters. Less than the set standard threshold Where Z represents the set standard threshold, or the maximum number of adjustments; obtain the optimized sub-task solutions corresponding to each potential task; integrate the sub-task solutions of the target task with the optimized sub-task solutions of the potential tasks to obtain the whole-process engineering consulting task solution.

2. The multi-agent collaborative management method for full-process engineering consulting according to claim 1, characterized in that, The method also includes: S6 performing prediction verification based on the obtained full-process engineering consulting task solution and the digital twin model of the engineering project to obtain the solution verification result, and further training and optimization correction of the relevant intelligent agents based on the solution verification result.

3. The multi-agent collaborative management method for full-process engineering consulting according to claim 1, characterized in that, Step S2 includes: understanding and dividing the acquired full-process engineering consulting tasks based on the LLM large language model, extracting the conditional features and target tasks of the full-process engineering consulting tasks; further enhancing the retrieval based on the acquired conditional features and target tasks to obtain the corresponding potential knowledge requirements, and further understanding the acquired potential knowledge requirements based on the LLM large language model to obtain potential tasks; taking the target tasks and potential tasks as sub-tasks to form the task decomposition features of the full-process engineering consulting tasks; wherein the conditional features include engineering specifications and constraints.

4. The multi-agent collaborative management method for full-process engineering consulting according to claim 3, characterized in that, Step S3 includes: extracting text feature vectors based on the description of the subtask and the corresponding engineering project feature information. Matching based on text feature vectors Task category tags corresponding to subtasks ,in ， Indicates task category tags, This represents the set of task category tags. Represents text feature vectors The probability that it belongs to task category label j.

5. A multi-agent collaborative management method for full-process engineering consulting as described in claim 4, characterized in that, Step S4 includes: the expert intelligent agents include investment decision consulting intelligent agents, engineering approval and construction service intelligent agents, engineering survey management intelligent agents, engineering design management intelligent agents, engineering bidding and procurement management intelligent agents, construction project management intelligent agents, engineering supervision intelligent agents, engineering cost management intelligent agents, engineering operation and maintenance management intelligent agents, other consulting service intelligent agents, quality management intelligent agents, safety management intelligent agents, schedule management intelligent agents, cost management intelligent agents, and other target management intelligent agents.

6. A multi-agent collaborative management method for full-process engineering consulting according to claim 5, characterized in that, Step S5 includes: extracting key clause information from the sub-task solution obtained by a single expert agent; performing compliance analysis on the key clause information in the task solution based on the corresponding engineering specifications, comparing the variable content in the key clause with the corresponding standard value range in the engineering specifications, and when the key clause information exceeds the standard range, adaptively correcting the content in the sub-task solution corresponding to the key clause information, adjusting the corresponding variable to the value closest to the current value and in line with the standard range, so that the sub-task solution complies with the engineering specifications.

7. A multi-agent collaborative management method for full-process engineering consulting according to claim 5, characterized in that, Step S6 includes: inputting the obtained full-process engineering consulting task solution into the established BIM-based full-process engineering consulting management model to perform performance analysis and risk analysis, obtaining the corresponding performance analysis results and risk analysis results, and recording the adaptability and adoption of the task solution to the target task based on the obtained performance analysis results and risk analysis results.

8. A multi-agent collaborative management system for end-to-end engineering consulting, characterized in that, include: The system comprises a receiving module, a processing module, and an output module, wherein the receiving module is used to receive a full-process engineering consulting task transmitted by a user; the processing module is used to execute a multi-agent collaborative management method for full-process engineering consulting as described in any one of claims 1-7 above, to obtain a full-process engineering consulting task solution; The output module is used to output the solution for the entire process engineering consulting task.

Images