CN117649129A - Multi-agent cooperative system and strategy method suitable for industrial digitization - Google Patents
Multi-agent cooperative system and strategy method suitable for industrial digitization Download PDFInfo
- Publication number
- CN117649129A CN117649129A CN202311803086.7A CN202311803086A CN117649129A CN 117649129 A CN117649129 A CN 117649129A CN 202311803086 A CN202311803086 A CN 202311803086A CN 117649129 A CN117649129 A CN 117649129A
- Authority
- CN
- China
- Prior art keywords
- task
- text
- tool
- industrial
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 21
- 238000004458 analytical method Methods 0.000 claims abstract description 30
- 238000007726 management method Methods 0.000 claims abstract description 28
- 230000006870 function Effects 0.000 claims abstract description 21
- 238000004364 calculation method Methods 0.000 claims abstract description 4
- 239000003795 chemical substances by application Substances 0.000 claims description 26
- 238000005457 optimization Methods 0.000 claims description 13
- 238000004519 manufacturing process Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 5
- 238000005065 mining Methods 0.000 abstract description 7
- 230000010365 information processing Effects 0.000 abstract description 4
- 238000011144 upstream manufacturing Methods 0.000 abstract description 4
- 241000282898 Sus scrofa Species 0.000 description 14
- 238000007405 data analysis Methods 0.000 description 7
- 238000005265 energy consumption Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 230000009467 reduction Effects 0.000 description 3
- 206010063385 Intellectualisation Diseases 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000004134 energy conservation Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000005206 flow analysis Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a multi-agent cooperative system and a strategy method suitable for industrial digitization. Comprising the following steps: the large model system is used for realizing the center management function; a tool library for providing basic functions; a functional model library for providing analysis operations; according to the input task, the large model system acquires and understands a user request corresponding to the task, accordingly decomposes the task into a plurality of targets and a plurality of subtasks corresponding to each target, schedules a corresponding tool of each subtask from a tool library, schedules a corresponding functional model of each subtask from a functional model library, completes the combined calculation or decision of data and models, and forms a strategy scheme corresponding to the input task; the problem that when the data center station is used for solving the cooperation of an industrial business system and the data value mining, the generated workload is large and the cooperation with business adjustment cannot be realized is solved, and meanwhile, the dependence of an engineer on the data information processing capability to realize the cooperation of the upstream and the downstream of the industrial chain can be reduced.
Description
Technical Field
The invention relates to the field of multi-agent cooperation, in particular to a multi-agent cooperation system and a strategy method suitable for industrial digitization.
Background
Along with the development of information technology and the development of industrial digitization, at the industrial end, the construction of an information system and the construction of application scenes are gradually perfected, how to mine data value from a complete business system and how to move from business interconnection to data driving decision-making become a new development stage of industrial digitization. At present, a data center is mainly used for bearing the exploration of data aggregation, data analysis and data value mining, and integrating and interconnecting all service systems. But solves the problems of industry service system coordination and data value mining by using a data center, and has the following three problems: 1. the workload is large, mass data are generated in various links of the industry, a large number of data analysis models are constructed by experience of engineers in information processing, and a large amount of time is consumed in data acquisition, cleaning, feature processing, modeling and the like; 2. the positioning of the data center is often focused on the processing of the data, the adjustment of the workload and the business cannot be effectively cooperated, for example, an OA system often makes an approval flow such as a change style, authority and the like of a production report, but the generation of the production report is finished in an ERP system, and the data center cannot cooperate with the business flow and the data flow; 3. the collaboration of the upstream and downstream of the industry chain depends on the processing capacity of engineers on data and information, and cannot actively mine the data value.
Therefore, in order to solve the above problems, the present invention provides a multi-agent collaboration system and a policy method suitable for industrial digitization, which are used for solving the problems that the workload generated by a data center station is large and the collaboration with business adjustment cannot be achieved when the industrial business system collaboration and the data value mining are solved, and simultaneously reducing the dependency of relying on engineers on the data information processing capability to achieve the collaboration between the upstream and the downstream of the industrial chain.
Disclosure of Invention
The invention aims to overcome at least one defect (deficiency) of the prior art, provides a multi-agent cooperation system and a strategy method suitable for industrial digitization, which are used for solving the problems that a data center station has large workload and cannot cooperate with business adjustment when the cooperation of an industrial business system and the mining of data value are solved, and simultaneously can reduce the dependence of an engineer on the data information processing capability to realize the cooperation of the upstream and the downstream of an industrial chain.
The technical scheme adopted by the invention is that the multi-agent cooperative system suitable for industrial digitization comprises:
the large model system is used for realizing the center management function;
a tool library for providing basic functions;
a functional model library for providing analysis operations;
according to the input task, the large model system acquires and understands a user request corresponding to the task, accordingly decomposes the task into a plurality of targets and a plurality of subtasks corresponding to each target, schedules a corresponding tool of each subtask from a tool library, schedules a corresponding functional model of each subtask from a functional model library, completes the combined calculation or decision of data and models, and forms a strategy scheme corresponding to the input task;
the system can divide the text input by the user into tasks, then divide each business process into one intelligent agent, realize the cooperative collaboration of each intelligent agent through a large model system with a powerful central management function, reduce the workload and improve the working efficiency.
Preferably, the system further comprises a domain database related to the domain related to the task and a memory module for realizing long-time and short-time memory;
and the memory module is used for recording a prior strategy scheme and providing a referential basis for the large model system to understand the context of the user request and the planning of each subtask.
The related data of the field database corresponding to the task is obtained through active mining of the data, and the prior strategy scheme is recorded, so that the large model can better understand the field knowledge related to the task, and the task is decomposed into more matched sub-targets, thereby reducing the dependence on data analysis engineers, further reducing the workload and improving the working efficiency.
Preferably, the multi-agent cooperative system suitable for industrial digitization further comprises an industrial business system and a comprehensive analysis management and control system; the industrial business system is used for providing real-time operation data of industry to the large model system; the comprehensive analysis management and control system is used for providing industrial staged analysis data for the large model system. In order to further fine tune and achieve the accuracy of task processing, an industrial business system and a comprehensive analysis management system are further provided in the system, so that a general large model can be matched with domain knowledge, the large model system can be refined to a specific industrial park when targets and tasks are disassembled, and the recent analysis data of the corresponding industrial park are acquired, and therefore the output strategy scheme is higher in matching degree and higher in business value.
Further preferably, the system also comprises a manual feedback module;
the large model system further decomposes the task into a plurality of targets and a plurality of subtasks corresponding to each target according to the prompt words fed back by the user. The completeness of the task described by the user can be further improved through manual feedback to the user for inputting the prompt word, so that the large model system can further understand the task details, and more accurate task strategy planning is realized.
Still further preferably, the artificial feedback module includes:
a knowledge base of all text generating task elements required for completing a task;
generating a vector database of task descriptions by all the texts;
a NOSQL database storing text-generated task descriptions and text-generated knowledge based on the text-generated task descriptions; the manual feedback module is used for feeding back the specific process of the prompt word input to the user as follows:
s41: the large model system acquires and understands a user request corresponding to the task to form an initial Prompt;
s42: vectorizing the initial Prompt text, and matching the text generating task description which best meets the task requirement in a vector database of the text generating task description;
s43: inquiring a NOSQL database of a text generation task knowledge base to complete text generation knowledge of the task according to the text generation description;
s44: constructing a system level Prompt, wherein the system level promt specifically refers to judging whether a user has complete description of a text generation task when a large model system receives knowledge about the text generation task;
s45: generating knowledge according to the text obtained in the step S43, and judging whether a Prompt word input needs to be fed back to a user or not by combining the system level Prompt of the step S44;
s46: if necessary, iterating the text to generate task description according to the prompt word fed back by the user;
s47: s45 and S46 are repeatedly performed until the large model confirms that the user has completely described his text generation task, and then generates text based on the final text generation task description.
Through the steps, update iteration of task description is realized, and the completeness of task text description input into the system by a user is finally improved, so that the output strategy scheme has higher matching degree and higher service value.
Preferably, the multi-agent collaborative system suitable for industrial digitization further comprises a capability judging module, wherein the capability judging module is used for judging the matching degree of the strategy scheme corresponding to the input task between a tool scheduled by the large model system from a tool library and a function model scheduled from a function model library. And the effect of the task completed by the system is evaluated by judging the matching degree, and if the matching degree is poor, the system performs updating learning and further performs self-optimization.
Further preferably, the large model system further comprises a tool production module, and when the large model system judges that the selected tool cannot reach the expected matching degree through the capability judging module, the tool production module is called to produce the tool with higher matching degree. The tool with higher matching degree is automatically produced to execute the required task, so that self-updating is realized, and the universality and the intelligence of the whole large model system are improved.
Preferably, a multi-agent collaborative strategy method suitable for industrial digitization comprises the following steps:
s1: task planning: performing target disassembly on the input task through the language big model to obtain a plurality of sub-targets;
s2: task analysis: understanding user requests of a plurality of sub-targets according to a domain-related domain database related to the task and a strategy scheme recorded in advance by a memory module, so as to generate a plurality of corresponding sub-tasks;
s3: tool and model selection: selecting a tool and a functional model for realizing each subtask;
s4: task execution: respectively executing corresponding subtasks according to the selected tools and the functional models, so as to achieve respective subtasks;
s5: and (3) strategy generation: fusing a plurality of sub-targets to form a total strategy scheme;
the task input by the user can be intelligently disassembled into a plurality of subtasks through the steps, and the corresponding subtasks are generated and executed, so that the dependence on a data analysis engineer is reduced, the overall workload is further reduced, and the working efficiency is improved.
Preferably, in the step S1, further artificial feedback updating is performed, if it is determined that the user does not completely describe the task of the text when the language big model receives knowledge about the task of generating the text, a prompt word input needs to be fed back to the user until the text input by the user can completely describe the task generated, and the task is disassembled into a plurality of sub-targets. The task text description of the language big model input by the user can be more complete through manual feedback update, so that the accuracy of final result generation is ensured.
Preferably, in the step S2, fine tuning learning and expert optimization are further included, fine tuning learning is performed by acquiring real-time operation data of the campus industry and industrial staged data, and an expert is introduced to perform optimization assistance, so that more accurate task analysis is realized.
Preferably, in the step S5, the capability judgment and tool production are further included, if it is judged that the matching degree of the tool scheduled by the language big model from the tool library and the function model scheduled from the function model library is poor, the tool generation is performed, the language big model automatically produces a new tool with higher matching degree, then the step S4 is re-executed, the execution of a plurality of subtasks is performed, the intellectualization of the language big model is further realized, and the universality is improved.
Compared with the prior art, the invention has the beneficial effects that:
the invention improves the synergy among the intelligent agents when the data center station solves the industrial business by providing the multi-agent collaboration system and the strategy method suitable for industrial digitization, improves the data processing capability by carrying out intelligent self data mining and analysis, reduces the dependence on data analysts, greatly reduces the workload and improves the working efficiency.
Drawings
Fig. 1 is a system configuration diagram of the present invention.
Fig. 2 is a flow chart of the system of the present invention.
Fig. 3 is a manual feedback flow chart of the present invention.
FIG. 4 is a schematic diagram of the strategy method of the present invention.
FIG. 5 is a schematic diagram of fine tuning learning and expert optimization according to the present invention.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the invention. For better illustration of the following embodiments, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the actual product dimensions; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
Example 1
As shown in fig. 1, in this embodiment, a multi-agent collaboration system suitable for industrial digitizing is provided, including:
the large model system is used for realizing the center management function;
a tool library for providing basic functions;
a functional model library for providing analysis operations;
according to the input task, the large model system acquires and understands a user request corresponding to the task, accordingly decomposes the task into a plurality of targets and a plurality of subtasks corresponding to each target, schedules a corresponding tool of each subtask from a tool library, schedules a corresponding functional model of each subtask from a functional model library, completes the combined calculation or decision of data and models, and forms a strategy scheme corresponding to the input task;
the system can divide the text input by the user into tasks, then divide each business process into one intelligent agent, realize the cooperative collaboration of each intelligent agent through a large model system with a powerful central management function, reduce the workload and improve the working efficiency.
Preferably, as shown in fig. 1, the system further comprises a domain database related to the domain related to the task and a memory module for realizing long and short time memory; the field database is transversely stored, and contains data of fields where each industry is located, including transportation management system, construction standard, vendor information, development planning and the like, so that more data references are provided for task analysis of the large model system; and meanwhile, the memory module is longitudinally stored, and can be called to find out the strategy corresponding to the task before the task when the task is analyzed, and the strategy is used as a reference to further perfect the final task analysis result.
And the memory module is used for recording a prior strategy scheme and providing a referential basis for the large model system to understand the context of the user request and the planning of each subtask.
The related data of the field database corresponding to the task is obtained through active mining of the data, and the prior strategy scheme is recorded, so that the large model can better understand the field knowledge related to the task, and the task is decomposed into more matched sub-targets, thereby reducing the dependence on data analysis engineers, further reducing the workload and improving the working efficiency.
Preferably, as shown in fig. 1, the multi-agent collaboration system suitable for industrial digitization further comprises an industrial business system and a comprehensive analysis management and control system, wherein the industrial business system comprises information such as energy consumption, property, office, security and the like of each industry, and the comprehensive analysis management and control system also performs comprehensive analysis management and control such as energy conservation and emission reduction, operation flow analysis, business and property analysis and the like on the information stored in the industrial business system.
Further, the industrial business system provides real-time operation data of the industry to the large model system, and the comprehensive analysis management and control system provides staged analysis data of the industry to the large model system. The task processing is further refined, so that the general large model can be matched with knowledge in various fields, a large model system can be refined to a specific industrial park when targets and tasks are disassembled, and the recent analysis data of the corresponding industrial park are acquired, so that the output strategy scheme has higher matching degree and higher service value.
Further preferably, as shown in fig. 2, a manual feedback module is further included;
the large model system further decomposes the task into a plurality of targets and a plurality of subtasks corresponding to each target according to the prompt words fed back by the user. The completeness of the task described by the user can be further improved through manual feedback to the user for inputting the prompt word, so that the large model system can further understand the task details, and more accurate task strategy planning is realized.
Still further preferably, the artificial feedback module includes:
a knowledge base of all text generating task elements required for completing a task;
generating a vector database of task descriptions by all the texts;
a NOSQL database storing text-generated task descriptions and text-generated knowledge based on the text-generated task descriptions; as shown in fig. 3, the specific process of the manual feedback module for feeding back the prompt word input to the user is as follows:
s41: the large model system acquires and understands a user request corresponding to the task to form an initial Prompt;
s42: vectorizing the initial Prompt text, and matching the text generating task description which best meets the task requirement in a vector database of the text generating task description;
s43: inquiring a NOSQL database of a text generation task knowledge base to complete text generation knowledge of the task according to the text generation description;
s44: constructing a system level Prompt, wherein the system level promt specifically refers to judging whether a user has complete description of a text generation task when a large model system receives knowledge about the text generation task;
s45: generating knowledge according to the text obtained in the step S43, and judging whether a Prompt word input needs to be fed back to a user or not by combining the system level Prompt of the step S44;
s46: if necessary, iterating the text to generate task description according to the prompt word fed back by the user;
s47: s45 and S46 are repeatedly performed until the large model confirms that the user has completely described his text generation task, and then generates text based on the final text generation task description.
Taking the example that the task description text of the system input by the user is 'live pig market quotation',
in step S41, the large model system first understands the acquired "live pig market quote" and forms an initial promt;
in step S42, text vectorization is performed on the initial campt, and the text generation task description which best meets the task requirement is matched in the vector database of the text generation task description to be "live pig, live pig market quotation";
in step S43, according to the text generation description, querying a NOSQL database of a text generation task knowledge base for the text generation knowledge of "live pig production place or brand, live pig breed, market price, quotation provider, quotation release time, etc";
in step S44, a system level promt is constructed to determine whether the user has a complete description of the text generation task;
in step S45, according to the task description "live pig, live pig market quotation" generated in step S42, and the text generated in step S43, generating knowledge "live pig producing area or brand, live pig variety, market price, quotation provider, quotation release time, etc." in combination with the system level Prompt of step S44, determining that a Prompt word input needs to be fed back to the user;
in step S46, iterating the text to generate task descriptions according to the prompt words fed back by the user, such as adding the prompt words of specific live pig producing places or brands, live pig breeds, quotation release time and the like;
in step S47, S45 and S46 are repeatedly performed until the large model confirms that the user has completely described its text generation task, such as: the live pig producing area or brand is Hunan Changsha, the live pig variety is external ternary, the quotation provider is Hunan Changsha market, the release time is 2023 and 9 months, and the like, but the method is not limited to the text input, the prompt word input can be carried out according to actual conditions, and finally, a task description is generated based on a final text to generate a text.
Through the steps, update iteration of task description is realized, and the completeness of task text description input into the system by a user is finally improved, so that the output strategy scheme has higher matching degree and higher service value.
Preferably, as shown in fig. 2, the multi-agent collaboration system suitable for industrial digitizing further includes a capability judging module, where the capability judging module is configured to judge a degree of matching between a tool scheduled by the large model system from the tool library and a function model scheduled from the function model library, and a policy scheme corresponding to the inputted task. The matching degree is judged so as to evaluate the effect of the task completed by the system, the system has learning capability, iterative updating capability and the like, and if the matching degree of the tool scheduled from the tool library and the strategy scheme corresponding to the input task is poor, the large model system performs updating learning, further performs self optimization, generates the tool meeting the task requirement and outputs the tool.
Further preferably, the large model system further comprises a tool production module, and when the large model system judges that the selected tool cannot reach the expected matching degree through the capability judging module, the tool production module is called to produce the tool with the higher matching degree. In this embodiment, the large model system may be used as a tool manufacturer to create tools for given tasks, the tools are usually implemented in the form of python utility functions, and perform required tasks by self-producing tools with higher matching degree, so as to implement self-updating iteration.
Preferably, as shown in fig. 4, a multi-agent collaborative strategy method suitable for industrial digitizing includes the following steps:
s1: task planning: performing target disassembly on the input task through the language big model to obtain a plurality of sub-targets;
s2: task analysis: understanding user requests of a plurality of sub-targets according to a domain-related domain database related to the task and a strategy scheme recorded in advance by a memory module, so as to generate a plurality of corresponding sub-tasks;
s3: tool and model selection: selecting a tool and a functional model for realizing each subtask;
s4: task execution: respectively executing corresponding subtasks according to the selected tools and the functional models, so as to achieve respective subtasks;
s5: and (3) strategy generation: fusing a plurality of sub-targets to form a total strategy scheme;
the task input by the user can be intelligently disassembled into a plurality of subtasks through the steps, and the corresponding subtasks are generated and executed, so that the dependence on a data analysis engineer is reduced, the overall workload is further reduced, and the working efficiency is improved.
Preferably, in the step S1, further artificial feedback updating is performed, if it is determined that the user does not completely describe the task of the text when the language big model receives knowledge about the task of generating the text, a prompt word input needs to be fed back to the user until the text input by the user can completely describe the task generated, and the task is disassembled into a plurality of sub-targets. The task text description of the language big model input by the user can be more complete through manual feedback update, so that the accuracy of final result generation is ensured.
Preferably, as shown in fig. 5, in the step S2, fine tuning learning and expert optimization are further included, fine tuning learning is performed by acquiring real-time operation data of the campus industry and industrial staged data, and an expert is introduced to perform optimization assistance, so that more accurate task analysis is realized.
As in the present embodiment, for the energy management flow, our task is: and providing energy conservation and emission reduction suggestions according to the energy policy, analyzing the energy consumption condition and optimally adjusting the energy supply of the park. The task description is submitted to a large language model, the large language model learns specific details of the energy management SOP, the energy analysis index system, the energy management standard specification and the like through fine adjustment learning, and simultaneously, key components of the energy management flow are found out by utilizing expert optimization assistance, detailed information of all the components of the energy management flow is obtained, and comprehensive analysis is performed on the energy management according to the provided information.
Then, according to the obtained information related to the energy management flow, the task is disassembled into four sub-tasks:
step1: the whole network searches for the latest energy policy and the energy management system of the park to generate an energy policy report.
step2: the system is connected with an energy management system, analyzes energy consumption trend of the park in the last five years, links a property relation system, associates property, the situation of the office enterprise of the park and the like, analyzes the energy consumption situation and generates a report;
step3: linking a park energy management expert model, an industry energy saving and emission reduction model and the like, associating energy situations, and generating an energy supply optimization suggestion;
step4: and the associated park equipment management system and the enterprise management system send out an energy supply optimization instruction.
Corresponding tools are then scheduled or generated according to each step to complete the corresponding tasks. And finally generating an output result: the energy policy analysis report, the energy consumption analysis report, the energy supply optimization report, the energy supply adjustment (execution/non-execution/demonstration) and the like.
By the fine tuning learning and expert introduction, the task disassembly accuracy of the large language model is higher, and the final output result meets the requirements of users.
Preferably, in the step S5, the capability judgment and tool production are further included, if it is judged that the matching degree of the tool scheduled by the language big model from the tool library and the function model scheduled from the function model library is poor, the tool generation is performed, the language big model automatically produces a new tool with higher matching degree, then the step S4 is re-executed, the execution of a plurality of subtasks is performed, the intellectualization of the language big model is further realized, and the universality is improved.
It should be understood that the foregoing examples of the present invention are merely illustrative of the present invention and are not intended to limit the present invention to the specific embodiments thereof. Any modification, equivalent replacement, improvement, etc. that comes within the spirit and principle of the claims of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A multi-agent collaboration system suitable for industrial digitizing, comprising:
the large model system is used for realizing the center management function;
a tool library for providing basic functions;
a functional model library for providing analysis operations;
according to the input task, the large model system acquires and understands a user request corresponding to the task, accordingly decomposes the task into a plurality of targets and a plurality of subtasks corresponding to each target, schedules a corresponding tool of each subtask from a tool library, schedules a corresponding functional model of each subtask from a functional model library, completes the combined calculation or decision of data and models, and forms a strategy scheme corresponding to the input task;
it is characterized in that the method comprises the steps of,
the system also comprises a domain database related to the domain related to the task and a memory module for realizing long-time and short-time memory;
and the memory module is used for recording a prior strategy scheme and providing a referential basis for the large model system to understand the context of the user request and the planning of each subtask.
2. The multi-agent co-system suitable for industrial digitizing as claimed in claim 1, wherein,
the system also comprises an industrial business system and a comprehensive analysis management and control system;
the industrial business system is used for providing real-time operation data of industry to the large model system;
the comprehensive analysis management and control system is used for providing industrial staged analysis data for the large model system.
3. The multi-agent co-system suitable for industrial digitizing as claimed in claim 1, wherein,
the system also comprises a manual feedback module;
the large model system further decomposes the task into a plurality of targets and a plurality of subtasks corresponding to each target according to the prompt words fed back by the user.
4. The multi-agent co-system suitable for industrial digitizing as claimed in claim 3, wherein,
the artificial feedback module comprises:
a knowledge base of all text generating task elements required for completing a task;
generating a vector database of task descriptions by all the texts;
a NOSQL database storing text-generated task descriptions and text-generated knowledge based on the text-generated task descriptions; the manual feedback module is used for feeding back the specific process of the prompt word input to the user as follows:
s41: the large model system acquires and understands a user request corresponding to the task to form an initial Prompt;
s42: vectorizing the initial Prompt text, and matching the text generating task description which best meets the task requirement in a vector database of the text generating task description;
s43: inquiring a NOSQL database of a text generation task knowledge base to complete text generation knowledge of the task according to the text generation description;
s44: constructing a system level Prompt, wherein the system level promt specifically refers to judging whether a user has complete description of a text generation task when a large model system receives knowledge about the text generation task;
s45: generating knowledge according to the text obtained in the step S43, and judging whether a Prompt word input needs to be fed back to a user or not by combining the system level Prompt of the step S44;
s46: if necessary, iterating the text to generate task description according to the prompt word fed back by the user;
s47: s45 and S46 are repeatedly performed until the large model confirms that the user has completely described his text generation task, and then generates text based on the final text generation task description.
5. The multi-agent co-system suitable for industrial digitizing as claimed in any of claims 1 to 4, wherein,
the system also comprises a capability judging module, wherein the capability judging module is used for judging the matching degree of the strategy scheme corresponding to the input task between the tools scheduled in the tool library and the functional models scheduled in the functional model library of the large model system.
6. The multi-agent co-system suitable for industrial digitizing as claimed in claim 5, wherein,
and the tool production module is used for calling the tool production module to produce the tool with higher matching degree when the large model system judges that the selected tool cannot reach the expected matching degree through the capability judgment module.
7. A multi-agent collaborative strategy method applicable to industrial digitization according to any one of claims 1-6, comprising the steps of:
s1: task planning: performing target disassembly on the input task through the language big model to obtain a plurality of sub-targets;
s2: task analysis: understanding user requests of a plurality of sub-targets according to a domain-related domain database related to the task and a strategy scheme recorded in advance by a memory module, so as to generate a plurality of corresponding sub-tasks;
s3: tool and model selection: selecting a tool and a functional model for realizing each subtask;
s4: task execution: respectively executing corresponding subtasks according to the selected tools and the functional models, so as to achieve respective subtasks;
s5: and (3) strategy generation: merging multiple sub-targets forms a total policy scheme.
8. The method according to claim 7, wherein the step S1 further comprises performing artificial feedback update, and if the language big model receives knowledge about the text generation task, determining that the user does not completely describe the text task, feeding back a prompt word input to the user until the text input by the user can completely describe the generated task, and performing object disassembly to disassemble the generated task into a plurality of sub-objects.
9. The method according to claim 8, wherein the step S2 further comprises fine tuning learning and expert optimization, fine tuning learning is performed by acquiring real-time operation data of the campus industry and industrial staged data, and expert is introduced to assist optimization.
10. The method according to claim 8, wherein the step S5 further comprises capability determination and tool production, and if it is determined that the matching degree of the language big model from the tools scheduled in the tool library and the function model scheduled in the function model library is poor, the tool generation is performed, the language big model automatically produces a new tool with a higher matching degree, and then the step S4 is re-executed to perform the execution of the plurality of subtasks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311803086.7A CN117649129A (en) | 2023-12-25 | 2023-12-25 | Multi-agent cooperative system and strategy method suitable for industrial digitization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311803086.7A CN117649129A (en) | 2023-12-25 | 2023-12-25 | Multi-agent cooperative system and strategy method suitable for industrial digitization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117649129A true CN117649129A (en) | 2024-03-05 |
Family
ID=90043418
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311803086.7A Pending CN117649129A (en) | 2023-12-25 | 2023-12-25 | Multi-agent cooperative system and strategy method suitable for industrial digitization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117649129A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220350965A1 (en) * | 2021-08-13 | 2022-11-03 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method for generating pre-trained language model, electronic device and storage medium |
| CN116976306A (en) * | 2023-08-01 | 2023-10-31 | 珠海市卓轩科技有限公司 | Multi-model collaboration method based on large-scale language model |
| CN117056531A (en) * | 2023-08-29 | 2023-11-14 | 上海大学 | Domain knowledge driven large language model fine tuning method, system, equipment and storage medium |
| CN117094376A (en) * | 2023-10-19 | 2023-11-21 | 浪潮电子信息产业股份有限公司 | Task processing method, device, system, equipment and readable storage medium |
| CN117194624A (en) * | 2023-08-17 | 2023-12-08 | 北京航空航天大学 | Traffic data analysis complex task intelligent disassembling and completing method based on large language model |
-
2023
- 2023-12-25 CN CN202311803086.7A patent/CN117649129A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220350965A1 (en) * | 2021-08-13 | 2022-11-03 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method for generating pre-trained language model, electronic device and storage medium |
| CN116976306A (en) * | 2023-08-01 | 2023-10-31 | 珠海市卓轩科技有限公司 | Multi-model collaboration method based on large-scale language model |
| CN117194624A (en) * | 2023-08-17 | 2023-12-08 | 北京航空航天大学 | Traffic data analysis complex task intelligent disassembling and completing method based on large language model |
| CN117056531A (en) * | 2023-08-29 | 2023-11-14 | 上海大学 | Domain knowledge driven large language model fine tuning method, system, equipment and storage medium |
| CN117094376A (en) * | 2023-10-19 | 2023-11-21 | 浪潮电子信息产业股份有限公司 | Task processing method, device, system, equipment and readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Weckenborg et al. | Balancing of assembly lines with collaborative robots | |
| Moser et al. | Semantic tool interoperability for engineering manufacturing systems | |
| Xie et al. | A new algorithm for complex product flexible scheduling with constraint between jobs | |
| Ke et al. | Solving project scheduling problem with the philosophy of fuzzy random programming | |
| Çil et al. | Two-sided disassembly line balancing problem with sequence-dependent setup time: A constraint programming model and artificial bee colony algorithm | |
| Breckle et al. | Engineering of assembly systems using graph-based design languages | |
| Li et al. | Two-sided assembly line balancing that considers uncertain task time attributes and incompatible task sets | |
| Cheng et al. | A comparison of optimal control and stochastic programming from a formulation and computation perspective | |
| Aslan | Mathematical model and a variable neighborhood search algorithm for mixed-model robotic two-sided assembly line balancing problems with sequence-dependent setup times | |
| CN111401427A (en) | Product cost evaluation method and system based on industrial big data | |
| Cong et al. | Value recovery from end-of-use products facilitated by automated dismantling planning | |
| Zhang et al. | Evolutionary algorithm incorporating reinforcement learning for energy-conscious flexible job-shop scheduling problem with transportation and setup times | |
| CN101331505A (en) | Method and apparatus for an algorithm development environment for solving a class of real-life combinatorial optimization problems | |
| CN117763943A (en) | Multi-stage production resource scheduling simulation optimization oriented algorithm configurable system | |
| Batova et al. | Informatisation of project activities performed by innovation clusters of the Kamsky agglomeration in Russia | |
| Zhang et al. | Configuration design of reconfigurable single-product robotic assembly line for capacity scalability | |
| CN117649129A (en) | Multi-agent cooperative system and strategy method suitable for industrial digitization | |
| CN115796527A (en) | Production scheduling management method and equipment based on knowledge graph | |
| Dewhurst et al. | Towards integrated manufacturing planning with common tool and information sets | |
| Wu et al. | A flexible planning methodology for product family assembly line based on improved NSGA_II | |
| Shi et al. | Evolutionary game analysis of vehicle procurement in the courier industry from the perspective of green supply chain | |
| CN111078203B (en) | 4M1T innovation system supporting incubation of creators and small micro-enterprises | |
| Kühn et al. | Investigation Of Genetic Operators And Priority Heuristics for Simulation Based Optimization Of Multi-Mode Resource Constrained Multi-Project Scheduling Problems (MMRCMPSP). | |
| Budzko et al. | Mathematical models of control in Digital Economy platforms | |
| Terrada et al. | Conceptualization and Configuration of a Generic SCM 4.0 Business Process Model Using BPMN Based on SCOR |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |