Drainage facility scheduling system and method based on large language model

Abstract

The embodiment of the invention provides a drainage facility scheduling system and method based on a large language model, and the method comprises the steps: extracting a keyword in a request instruction through a scene analysis unit, generating a standardized task instruction according to the keyword, calling a database to verify the authenticity of the instruction, reading the instruction through a task recognition unit, and obtaining a task type; the task decomposition unit generates a planning path according to a task type, the task decomposition unit generates a plurality of task schemes, decomposes each task scheme to obtain a sub-task set of each scheme, identifies a task chain in each sub-task set, and the resource matching unit matches execution models corresponding to a plurality of sub-tasks and data resources called by the execution models according to the task chains. The analysis unit executes the sub-tasks according to the called data resources and generates execution results of a plurality of task schemes, the strategy unit analyzes the execution results to determine a scheduling execution scheme of the drainage facility, the system scheduling scheme and strategy are formulated based on a large language model, manual formulation is not needed, and the generation efficiency of the drainage system scheduling scheme is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a drainage facility scheduling system and method based on a large language model. Background Art

[0002] The urban plant-network-river integrated dispatching system is a comprehensive management system for urban water environment governance and sewage treatment. It realizes integrated dispatching and management by integrating multiple scenes such as sewage treatment plants, drainage networks, and rivers.

[0003] In the existing technology, the urban plant-network-river integrated dispatching system is based on online monitoring equipment and uses technologies such as the Internet, Internet of Things communications and GIS systems to achieve full-process monitoring and intelligent dispatching.

[0004] However, the scheduling system in the existing technology has high technical complexity, great difficulty in data integration, and relies on manual experience to formulate scheduling plans, with a low level of intelligence, resulting in inefficient formulation of drainage system scheduling plans. Summary of the invention

[0005] The embodiments of the present application provide a drainage facility scheduling system and method based on a large language model to solve the problem of low efficiency in formulating scheduling plans in the prior art.

[0006] In a first aspect, an embodiment of the present application provides a drainage facility scheduling system based on a large language model, including: a scenario analysis unit, a task identification unit, a task decomposition unit, a resource matching unit, an analysis unit, and a strategy unit;

[0007] In response to a user inputting a request instruction, the scenario analysis unit extracts keywords from the request instruction to generate instruction keywords;

[0008] The scenario analysis unit performs standardization processing on the instruction keywords to generate standardized task instructions, and calls a database to verify the authenticity of the standardized task instructions;

[0009] The task identification unit reads and parses the standardized task instruction, obtains the task type corresponding to the standardized task instruction, and generates a planned path according to the task type;

[0010] The task decomposition unit identifies the task scope and task subject of the standardized task instruction according to the task planning path, and generates a plurality of task solutions according to the task scope and the task subject;

[0011] The task decomposition unit decomposes the multiple task solutions to obtain subtask sets corresponding to the multiple task solutions, and generates corresponding task chains according to the multiple subtasks in the subtask sets;

[0012] The resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain;

[0013] The analysis unit executes a plurality of subtasks of each scheme according to the data resources called by the execution model, and generates execution results of the plurality of task schemes;

[0014] The strategy unit analyzes the execution results of the multiple task plans to determine a scheduling execution plan for the drainage facility.

[0015] In a possible implementation, the drainage facility scheduling system based on a large language model further includes: a memory unit; the memory unit stores the relationship between the tasks of each unit in the drainage facility scheduling system based on a large language model and standard instructions, task types, planned paths, task subjects, task plans and resource matching, as well as the scheduling plan generated by the strategy unit.

[0016] In a possible implementation, after the scenario analysis unit performs standardization processing on the instruction keywords, it also includes: the scenario analysis unit compares the instruction keywords with preset task division rules, and if the instruction keywords do not meet the task division rules, sends information supplementary instructions to the user end; receives the supplementary instruction keywords sent by the user end, and generates standardized instructions based on the supplementary instruction keywords; and calls a database to verify the authenticity of the standardized instructions.

[0017] In a possible implementation, the resource matching unit matches the execution models corresponding to the multiple subtasks and the data resources called by the execution models according to the task chain, including: the resource matching unit calculates the vector similarities of the multiple subtasks and the model text of each solution according to the task chain, and obtains the corresponding execution model according to the vector similarities of the multiple subtasks and the model text of each solution; the resource matching unit matches the data resources called by the execution model according to a preset matching relationship.

[0018] In a possible implementation, after the analysis unit executes the multiple subtasks of each scheme according to the data resources called by the execution model and generates the execution results of the multiple task schemes, it also includes: the analysis unit reads the operating data of the multiple subtasks of executing each scheme through a large language model, generates calculation results of multiple task schemes according to the operating data of the multiple subtasks of executing each scheme, and performs visual analysis on the calculation results of the multiple task schemes to generate visualization results.

[0019] In a possible implementation, the strategy unit parses the execution results of the multiple task plans to determine the scheduling execution plan for the drainage facilities, including: the strategy unit parses the execution results of the multiple task plans through a large language model, optimizes the execution results of the multiple task plans, and obtains the optimized execution results of the multiple task plans; the strategy unit inputs the optimized execution results of the multiple task plans into the large language model, and generates a scheduling plan for the drainage facilities through the large language model.

[0020] In a second aspect, an embodiment of the present application provides a drainage facility scheduling method based on a large language model, which is applied to a drainage facility scheduling system based on a large language model. The drainage facility scheduling system based on a large language model includes: a scene analysis unit, a task identification unit, a task decomposition unit, a resource matching unit, an analysis unit, and a strategy unit; the method includes:

[0021] In response to a user inputting a request instruction, the scenario analysis unit extracts keywords from the request instruction to generate instruction keywords;

[0022] The scenario analysis unit performs standardization processing on the instruction keywords to generate standardized task instructions, and calls a database to verify the authenticity of the standardized task instructions;

[0023] The task identification unit reads and parses the standardized task instruction, obtains the task type corresponding to the standardized task instruction, and generates a planned path according to the task type;

[0024] The task decomposition unit identifies the task scope and task subject of the standardized task instruction according to the task planning path, and generates a plurality of task solutions according to the task scope and the task subject;

[0025] The task decomposition unit decomposes the multiple task solutions to obtain subtask sets corresponding to the multiple task solutions, and generates corresponding task chains according to the multiple subtasks in the subtask sets;

[0026] The resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain;

[0027] The analysis unit executes a plurality of subtasks of each scheme according to the data resources called by the execution model, and generates execution results of the plurality of task schemes;

[0028] The strategy unit analyzes the execution results of the multiple task plans to determine a scheduling execution plan for the drainage facility.

[0029] In a possible implementation, the drainage facility scheduling system based on a large language model also includes: a memory unit; the method includes: the memory unit stores the relationship between the tasks of each unit in the drainage facility scheduling system based on a large language model and standard instructions, task types, planned paths, task subjects, task plans and resource matching, as well as the scheduling plan generated by the strategy unit.

[0030] In a possible implementation, after the scenario analysis unit performs standardization processing on the instruction keywords, it also includes: the scenario analysis unit compares the instruction keywords with preset task division rules, and if the instruction keywords do not meet the task division rules, sends information supplementary instructions to the user end; receives the supplementary instruction keywords sent by the user end, and generates standardized instructions based on the supplementary instruction keywords; and calls a database to verify the authenticity of the standardized instructions.

[0031] In a possible implementation, the resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain, including: the resource matching unit calculates the vector similarities of the multiple subtasks of each solution and the model text according to the task chain, and obtains the corresponding execution model according to the vector similarities of the multiple subtasks of each solution and the model text; the resource matching unit matches the data resources called by the execution model according to a preset matching relationship.

[0032] In a possible implementation, after the analysis unit executes the multiple subtasks of each scheme according to the data resources called by the execution model and generates the execution results of the multiple task schemes, it also includes: the analysis unit reads the operating data of the multiple subtasks of executing each scheme through a large language model, generates calculation results of multiple task schemes according to the operating data of the multiple subtasks of executing each scheme, and performs visual analysis on the calculation results of the multiple task schemes to generate visualization results.

[0033] In a possible implementation, the strategy unit parses the execution results of the multiple task plans to determine the scheduling execution plan for the drainage facilities, including: the strategy unit parses the execution results of the multiple task plans through a large language model, optimizes the execution results of the multiple task plans, and obtains the optimized execution results of the multiple task plans; the strategy unit inputs the optimized execution results of the multiple task plans into the large language model, and generates a scheduling plan for the drainage facilities through the large language model.

[0034] The drainage facility scheduling system and method based on the large language model provided in the embodiment of the present application extracts keywords in the request instruction through the scenario analysis unit, generates standardized task instructions according to the keywords, and calls the database to verify the authenticity of the instruction. The task identification unit reads the instruction to obtain the task type, generates a planning path according to the task type, and the task decomposition unit generates multiple task plans, decomposes each task plan to obtain a subtask set of each plan, identifies the task chain in each subtask set, and the resource matching unit matches the execution models corresponding to the multiple subtasks and the data resources called by the execution model according to the task chain. The analysis unit executes the subtasks according to the called data resources and generates the execution results of the multiple task plans. The strategy unit parses the execution results to determine the scheduling execution plan of the drainage facility, and formulates the scheduling plan and strategy of the drainage facility scheduling system based on the large language model without manual formulation, thereby improving the generation efficiency of the drainage system scheduling plan. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0036] Figure 1 A schematic diagram of the structure of a drainage facility scheduling system based on a large language model provided in an embodiment of the present application;

[0037] Figure 2 A flow chart of a drainage facility scheduling method based on a large language model provided in an embodiment of the present application;

[0038] Figure 3 Schematic diagram of the knowledge graph provided for the embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0040] First, the terms involved in this application are explained:

[0041] Large Language Models: Large Language Models, or LLMs for short, refer to language models with tens of billions of parameters or more that are trained on large-scale text corpora to understand and generate human language and perform tasks such as text summarization, translation, and classification.

[0042] The urban plant-network-river integrated dispatching system is a comprehensive management system for urban water environment governance and sewage treatment. It realizes integrated dispatching and management by integrating multiple scenes such as sewage treatment plants, drainage networks, and rivers. In the existing technology, the urban plant-network-river integrated dispatching system is based on online monitoring equipment and uses technologies such as the Internet, Internet of Things communications, and GIS systems to achieve full-process monitoring and intelligent dispatching. However, the dispatching system in the existing technology has high technical complexity, great difficulty in data integration, and relies on manual experience to formulate dispatching plans. The level of intelligence is low, resulting in low efficiency in formulating drainage system dispatching plans.

[0043] In order to solve the above technical problems, the embodiment of the present application proposes the following technical concept: the inventor considers designing a drainage facility scheduling system based on a large language model, and the drainage facility scheduling system based on the large language model includes: a scene analysis unit, a task identification unit, a task decomposition unit, a resource matching unit, an analysis unit and a strategy unit. Considering the use of the scene analysis unit to extract the keywords in the request instruction, standardize the instruction keywords, generate standardized task instructions, and call the database to verify the authenticity of the instruction; the task identification unit is used to parse the standardized task instruction to obtain the task type, and generate a planning path according to the task type; multiple task schemes are generated by the task decomposition unit, and each task scheme is decomposed to obtain the subtask set of each scheme, and the task chain of each subtask set is identified; the resource matching unit is used to match the execution model corresponding to multiple subtasks according to the task chain, and the data resources called by the execution model; the analysis unit executes the subtask according to the data resources called by the execution model, generates the execution results of multiple task schemes, and parses the execution results through the strategy unit to determine the scheduling execution plan of the drainage facility, thereby improving the efficiency of formulating the drainage system scheduling plan. The following detailed embodiments are used for detailed description.

[0044] Figure 1 A structural diagram of a drainage facility scheduling system based on a large language model provided in an embodiment of the present application, such as Figure 1 As shown, the drainage facility scheduling system based on the large language model includes: a scene analysis unit 101, a task identification unit 102, a task decomposition unit 103, a resource matching unit 104, an analysis unit 105 and a strategy unit 106.

[0045] In response to the user inputting a request instruction, the scenario analysis unit extracts keywords from the request instruction and generates instruction keywords.

[0046] In this embodiment, the request instruction input by the user may be in a form including, but not limited to, text, voice, and text converted from voice input.

[0047] Specifically, the scene analysis unit retrieves the scene vector library through a large language model, extracts and classifies keywords for request instructions, and generates instruction keywords.

[0048] The types of keyword classification include but are not limited to time, place, object and content.

[0049] Exemplarily, the request instruction input by the user is: "On July 10, equipment Y of X sewage treatment plant failed. Generate an emergency plan to minimize the impact on the water quality of monitoring point Z in the river downstream of the sewage treatment plant outlet during the maintenance period." The scenario analysis unit selects the enhanced retrieval mode of the large language model, retrieves the plant-network-river scheduling scenario in the vector knowledge base, extracts the keywords in the request instruction, classifies the keywords, and generates instruction keywords.

[0050] For example, the classification results of keywords are:

[0051] Time: July 10, during maintenance period.

[0052] Location range: X sewage treatment plant, Z monitoring point in the river downstream of the outlet of X sewage treatment plant.

[0053] Cause: Equipment No. Y at X sewage treatment plant malfunctioned.

[0054] Task object: Water quality of the river downstream of the outlet of X sewage treatment plant.

[0055] Task content: Develop emergency plan.

[0056] Task requirements: Minimize the impact of faults on water quality at this point during equipment maintenance.

[0057] The scenario analysis unit performs standardization processing on the instruction keywords, generates standardized task instructions, and calls the database to verify the authenticity of the standardized task instructions.

[0058] Specifically, the scene analysis unit standardizes ambiguous or diversified instructions through the preset prompt project in the large language model to generate standardized task instructions. After the standardized task instructions are generated, the corresponding task object observation database is called according to the task object keywords to verify the authenticity of the task.

[0059] The task identification unit reads and parses the standardized task instructions, obtains the task type corresponding to the standardized task instructions, and generates a planned path according to the task type.

[0060] In this embodiment, the commonly used task types in the standardized task instruction scenarios are divided into four categories: knowledge query, data query, analysis, and deduction.

[0061] Among them, the knowledge query category involves the use of large language models and vector databases; the data query category involves database queries; the analysis category involves database queries and the calling of analysis methods; and the deduction category involves the collaborative application of multiple tools such as databases and models.

[0062] Specifically, the task identification unit collects standardized task instructions, classifies and vectorizes them according to the above task types, establishes a preset instruction type vector database, performs matching analysis between the standardized task instructions and the preset instruction type vector database, and determines the task type.

[0063] Specifically, a planning path is generated according to different task types.

[0064] Exemplarily, the planning path for the knowledge query class is: vectorize the standardized instructions; calculate the similarity with the vector database data; select the top K similarity data and the standardized instructions and feed them back to the big model based on the preset prompt; the big model generates the answer.

[0065] Exemplarily, the planning path for the data query class is: the large model determines the data to be stored according to the standardized instructions; generates SQL database call instructions to call the data; feeds the data and the standardized instructions back to the large language model based on the preset prompt; and the large language model generates the answer.

[0066] Exemplarily, the planning path for the analysis class is: the large model determines the database to be called and the analysis tool to be called according to the standardized instructions, where the analysis tools include but are not limited to: comparative analysis, correlation analysis, predictive analysis, attribution analysis, penetration analysis, sensitivity analysis and recommendation analysis tools; generates SQL database call instruction call data, and analyzes the data by calling the analysis interface through the API interface; feeds back the standardized instructions and analysis results to the large language model based on the preset prompt; and the large language model generates answers.

[0067] Exemplarily, the planning path for the deduction type is: determine the spatial scope and execution time of the task; determine the subjects involved in the task object and the relationship between the subjects; determine the path to achieve the task goal; determine the type of deduction task; identify the controllable subject; build a solution library; build task chains for each solution; match the models or tools and data resources required for the task units in each solution task chain; execute the task chains of each solution to analyze the results based on rules, generate strategies and visualize them; feed back the optimal solution selected by the staff together with the report template to the large language model, and the large language model generates the final solution report.

[0068] The task decomposition unit identifies the task scope and task subject of the standardized task instruction according to the task planning path, and generates multiple task plans according to the task scope and task subject.

[0069] Specifically, the task decomposition unit decomposes the standardized task instruction, obtains the spatial scope and time scope of the standardized task instruction, identifies the target subject and associated subject of the standardized task instruction, and obtains multiple subtask sets.

[0070] Exemplarily, the spatial scope can be determined by a preset "key node-spatial scope" database, where the "key node-spatial scope" database includes nodes of different levels for determining the scene, and different nodes correspond to pre-built models such as surface hydrodynamic water quality model, pipe network hydrodynamic water quality model, river hydrodynamic water quality model and sewage treatment plant water quality prediction model, and match the corresponding basic database, parameter library, database required for model calculation and calculation result analysis rule library and other resources. When executing a task, by identifying the location of the task target and matching it with the preset key points, the task target range is smaller than the matching point target, and the spatial scope of task execution and the corresponding data are obtained.

[0071] Among them, the time range can be obtained by analyzing the standard task instructions through the large model and performing relevant calculations.

[0072] Figure 3 Schematic diagram of the knowledge graph provided for the embodiment of the present application.

[0073] For example, the start time is July 10, the time when the fault occurred, and the end time is July 10 plus the equipment maintenance time. The equipment maintenance time can be determined by identifying the fault and Figure 3 It can be obtained by using "equipment-fault-type-handling solution-time" in the knowledge graph shown, and then the start time of task execution is obtained.

[0074] For example, the identification of the target subject and the associated subject by the task decomposition unit: Taking "On July 10, X sewage treatment plant, Y equipment failed, and a set of emergency plans were formulated to minimize the impact on the water quality of the Z monitoring point of the river downstream of the sewage treatment plant outlet during the maintenance of Y equipment" as an example, the target subject is the water quality of the Z monitoring point of the river downstream of the sewage treatment plant outlet. Figure 3The knowledge graph shown in the figure has six types of links related to the subjects: 1. Rainfall—rainwater pipe network—rainwater outlet—river—task point water quantity and quality; 2. Rainfall—rainwater pipe network—initial rainwater storage tank—sewage treatment plant—sewage treatment plant outlet—river—task point water quantity and quality; 3. Rainfall—combined sewer network—combined sewer storage tank—sewage treatment plant—sewage treatment plant outlet—river—task point water quantity and quality; 4. Residential water use—combined sewer network—combined sewer storage tank—sewage treatment plant—sewage treatment plant outlet—river—task point water quantity and quality; 5. Residential water use—sewage pipe network—sewage treatment plant—sewage treatment plant outlet—river—task point water quantity and quality; 6. Natural underlying surface—river—hydraulic structures such as sluices / dams—task point water quantity and quality.

[0075] The task decomposition unit decomposes multiple task schemes to obtain subtask sets corresponding to the multiple task schemes, and generates corresponding task chains according to the multiple subtasks in the subtask sets.

[0076] Specifically, the task decomposition unit identifies the controllable subjects in the task body, generates an executable solution library according to the number and type of controllable subjects, and generates a corresponding task chain according to the target subject link.

[0077] Exemplarily, among the six types of links into which the above-mentioned entities associated therewith are divided, the controllable entities are: initial rainwater storage tanks, combined sewage storage tanks, hydraulic structures such as gates / dams, and sewage treatment plants.

[0078] Exemplarily, executable solutions are generated according to the number and type of controllable subjects through the solution library. For the above-mentioned 6 types of links divided by the subjects associated with it, the executable solutions are: 1. Control the initial rainwater storage tank; 2. Control the combined rainwater and sewage storage tank; 3. Control the hydraulic structures such as gates / dams; 4. Control the equipment and processes in the sewage treatment plant; 5. Control the initial rainwater storage tank + sewage storage tank; 6. Control the initial rainwater storage tank + hydraulic structures such as gates and dams; 7. Control the initial rainwater storage tank + equipment and processes in the sewage treatment plant; 8. Control the combined rainwater and sewage storage tank + hydraulic structures such as gates and dams; 9. Regulate the equipment and processes inside the combined rainwater and sewage storage tank + sewage treatment plant; 10. Regulate the hydraulic structures such as sluice gates and dams + sewage treatment plant; 11. Regulate the initial rainwater storage tank + combined rainwater and sewage storage tank + sluice gates and dams and other hydraulic structures; 12. Regulate the equipment and processes inside the initial rainwater storage tank + combined rainwater and sewage storage tank + sewage treatment plant; 13. Regulate the equipment and processes inside the combined rainwater and sewage storage tank + sluice gates and dams and other hydraulic structures + sewage treatment plant; 14. Regulate the equipment and processes inside the initial rainwater storage tank + combined rainwater and sewage storage tank + sluice gates and dams and other hydraulic structures + sewage treatment plant.

[0079] Exemplarily, by combining the 6 types of links into which the associated subjects are divided, the task chain and task unit of each implementation plan are determined. Taking executable plan 14 as an example, the task chain is: model matching; model module identification and activation; controllable subject scheduling rule identification; data resource call required for model calculation; model calculation; model calculation result analysis and visual demonstration; human-computer interaction for solution optimization; emergency plan generation; storage of operation information of each module.

[0080] The resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain.

[0081] Specifically, the resource matching unit calculates the vector similarity between multiple subtasks of each solution and the model introduction text in the model library, obtains the execution model according to the vector similarity, and obtains the data resources required for the model operation from the preset model-resource matching library.

[0082] The analysis unit executes multiple subtasks of each solution according to the data resources called by the execution model, and generates execution results of multiple task solutions.

[0083] Specifically, the analysis unit calculates each solution in the solution library according to the preset calculation rules based on the matching task scope, model and calculation module, data resources and scheduling rules. The calculation process sends instructions through the preset Prompt, uses the large language model to generate programming language from the instructions, and calls and executes the model and database in the form of calling the API interface through the programming language. It performs multiple subtasks and finally realizes the calculation of each solution. It also uses the large language model to call the icon drawing tool for visual analysis and provide decision makers with decision support.

[0084] The strategy unit analyzes the execution results of multiple task plans to determine the scheduling execution plan for the drainage facilities.

[0085] Specifically, the execution results of each plan are analyzed through a large language model, the optimal plan is determined based on the staff's selection and visual analysis results, and then improved through human-computer interaction; the optimal plan for task execution is generated based on the large language model, preset templates and related databases, and decision makers achieve the optimal call of drainage facilities by executing the plan.

[0086] It can be seen from the above embodiments that the scenario analysis unit extracts keywords in the request instruction, generates standardized task instructions based on the keywords, and calls the database to verify the authenticity of the instruction. The task identification unit reads the instruction to obtain the task type, generates a planning path based on the task type, and the task decomposition unit generates multiple task plans, decomposes each task plan to obtain a subtask set of each plan, identifies the task chain in each subtask set, and the resource matching unit matches the execution models corresponding to the multiple subtasks and the data resources called by the execution model according to the task chain. The analysis unit executes the subtasks according to the called data resources and generates the execution results of multiple task plans. The strategy unit parses the execution results to determine the scheduling execution plan of the drainage facilities, and formulates the scheduling plan and strategy of the drainage facility scheduling system based on the large language model without manual formulation, thereby improving the generation efficiency of the drainage system scheduling plan.

[0087] In one embodiment of the present application, the drainage facility scheduling system based on the large language model further includes: a memory unit;

[0088] The memory unit stores the relationship between the tasks and standard instructions, task types, planned paths, task subjects, task plans and resource matching of each unit in the drainage facility scheduling system based on the large language model, as well as the scheduling plan generated by the strategy unit.

[0089] Specifically, the memory unit stores the "task-standard instructions" generated by the scene analysis unit, the "task-task type-planning path" generated by the task identification unit, the "task-task body-task plan-task chain" generated by the task decomposition unit, the "sub-task-model-data" matching relationship determined by the resource matching unit, and the scheduling plan generated by the strategy unit.

[0090] Specifically, the memory unit standardizes the data that can be stored in each link and stores it in the corresponding preset database. When the drainage facility scheduling system receives the same type of request instruction again, it realizes the reuse of solutions to the same type of problems by calling and referring to the solutions and strategies stored in the memory unit.

[0091] Exemplarily, the task "River Channel Z Point - Spatial Range" and the models and data called later are stored in the "Key Node - Spatial Range" database; the generated emergency plan is stored in the emergency plan library.

[0092] It can be seen from the above embodiments that by storing the data and planned paths of each link in the drainage facility scheduling system in the corresponding database, it is convenient for staff to refer to and optimize the solution when encountering similar problems in the future, thereby improving the efficiency and accuracy of task execution.

[0093] In one embodiment of the present application, after the scene analysis unit performs standardization processing on the instruction keywords, the following steps are further included:

[0094] The scenario analysis unit compares the instruction keywords with the preset task division rules. If the instruction keywords do not meet the task division rules, an information supplement instruction is sent to the user end.

[0095] In this embodiment, the preset task division rules are recorded in the standard instruction library.

[0096] Specifically, if the instruction keywords do not satisfy the task division rules, the scene analysis unit sends information supplementary instructions to the user terminal through the large language model.

[0097] Receive the supplementary instruction keywords sent by the user end, and generate standardized instructions according to the supplementary instruction keywords.

[0098] Specifically, the supplementary instruction keywords sent by the user end are received, and the original instruction keywords and the supplementary instruction keywords are combined with the large language model to determine whether the task division rules are met. If the task division rules are met, the task decomposition module instructions are executed, the supplementary instruction keywords are standardized, and standardized instructions are generated.

[0099] Call the database to verify the authenticity of the standardized instructions.

[0100] Specifically, according to the task object keyword in the standardized instruction, the corresponding observation database is called to verify whether the event in the task occurs.

[0101] From the above embodiments, it can be seen that the scene analysis unit is used to determine whether the instruction keywords meet the task division rules. If the task division rules are not met, the user is prompted to complete the instruction keywords and generate standardized instructions. Standardized instructions are generated through the supplemented instruction keywords, and the observation database is called according to the task object keywords in the instructions to verify whether the events in the task occur, thereby improving the accuracy of generated instructions.

[0102] In one embodiment of the present application, the resource matching unit matches the execution models corresponding to the plurality of subtasks and the data resources called by the execution models according to the task chain, including:

[0103] The resource matching unit calculates the vector similarity between the multiple subtasks of each solution and the model text according to the task chain, and obtains the corresponding execution model according to the vector similarity between the multiple subtasks of each solution and the model text.

[0104] Specifically, the resource matching unit uses a large language model to ask questions to the required model according to the preset prompt project, task scope, identified subject and identified scheme, and determines the model to be used and the calculation module in the model through vector similarity matching.

[0105] Among them, the large language model determines the model to be used and the calculation module in the model through the external model knowledge base. The content recorded in the external model knowledge base includes but is not limited to the task scope of each model, the subjects involved in each model, and the standardized series description labels involved in each model.

[0106] Exemplarily, the task to be completed involves the prediction of water volume and quality in urban rivers. Among the identified entities, there are subjects such as "rainfall, rainwater pipe network, sewage pipe network, storage tank, sewage treatment plant, sewage outlet, river, dam, water volume and water quality". Matching analysis is performed through the large language model, and the "plant-network-river integrated scheduling model" is called, which includes surface water dynamics and water quality module, pipe network water dynamics and water quality module, sewage treatment plant water quality prediction module, river water dynamics and water quality module and optimization algorithm module.

[0107] The resource matching unit matches the data resources called by the execution model according to a preset matching relationship.

[0108] Specifically, based on the preset "model-database" matching relationship, combined with information such as task scope, task time, model used and computing module called, data type, storage location and start time, a large language model is used to generate SQL statements, and the corresponding data resources in the database are called through the API interface.

[0109] In the process of analyzing the consumed data resources, the specific name and location of the controllable subject within the scope are identified by determining the task scope, and the scheduling rules of each subject in the preset controllable subject rule library are called through the large language model to optimize the calculation process.

[0110] It can be seen from the above embodiments that the vector similarity between each solution subtask and the model introduction text in the model library is calculated through the resource matching unit, the model and module used are matched according to the vector similarity, the data resources required for the execution model are determined according to the preset matching relationship, and the scheduling rules of each subject in the preset controllable subject rule library are called through the large language model to optimize the calculation process and improve the accuracy of the calculation data resources.

[0111] In one embodiment of the present application, after the analysis unit executes multiple subtasks of each solution according to the data resources called by the execution model and generates the execution results of multiple task solutions, it also includes:

[0112] The analysis unit reads the running data of multiple subtasks of each solution through a large language model, generates calculation results of multiple task solutions according to the running data of multiple subtasks of each solution, and performs visualization analysis on the calculation results of multiple task solutions to generate visualization results.

[0113] Specifically, the analysis unit uses the large language model to determine the required data resources and the data analysis rule base according to the preset emergency plan template, parses the operation data through the data analysis rule base, inputs the parsed operation data and multiple sub-task execution results into the large language model, calls the drawing software or related auxiliary tools through the large language model to draw charts, and generates a corresponding visualization plan library.

[0114] For example, the changes in water quality, the secondary risks and benefits generated under the conditions of executing various plans are obtained, the water quality calculation results of point Z of the river section calculated by the model are read, the energy, chemicals and labor resources used in the use of pumping stations, sewage treatment plants and dams are counted, and the relevant databases and calculation rules are called through the big language model for calculation. At the same time, the operating data of equipment and facilities such as sewage treatment plants, storage tanks and pipelines in the calculation process are read and analyzed, and the standard task instructions, water quality simulation results, accounting results and risk prediction results are fed back to the big language model. The big language model calls the drawing software to generate a visualization plan.

[0115] It can be seen from the above embodiments that the operation data corresponding to each plan is read through a large language model, and a visualization plan is generated according to the operation data and the execution results of each plan, so that the staff can intuitively observe the task execution results and adjust the decision according to the task execution results.

[0116] In one embodiment of the present application, the strategy unit analyzes the execution results of multiple task plans to determine the scheduling execution plan of the drainage facility, including:

[0117] The strategy unit analyzes the execution results of multiple task solutions through a large language model, optimizes the execution results of the multiple task solutions, and obtains the optimized execution results of the multiple task solutions.

[0118] Specifically, the strategy unit analyzes the execution results of each solution through a large language model, conducts human-computer interaction on the analyzed results, optimizes the execution results of each solution, and determines the optimal solution based on the rules.

[0119] Illustratively, through the water quality control conditions, calculation results and risk analysis results generated by the large language model, the staff interacts with the large language model and issues instructions to the large language model, such as "eliminate the plans with higher risks and poor water quality control at point Z in the river, and sort the remaining plans in order of cost from low to high" and "set the water quality target to no more than Class IV of surface water, recalculate, and sort the plans with compound requirements in order of cost from low to high". The large language model is used to optimize the calculation, optimize the execution results of each plan, and determine the optimal scheduling plan according to the scheduling rules.

[0120] The strategy unit inputs the execution results of multiple optimized task plans into the large language model and generates a scheduling plan for the drainage facilities through the large language model.

[0121] Specifically, the template required for the emergency plan is input into the large language model, the data resources required to generate the execution strategy are determined by the large language model, and the plan-related data is input into the large language model through the preset prompt to generate the final emergency plan.

[0122] Exemplarily, the list of staff members in each link of the above-determined optimal plan, the current inventory of emergency materials, meteorological data during equipment failure handling, and the current operating status of the regulating reservoir, dam, etc. are input into the large language model through a preset prompt to generate an emergency plan.

[0123] It can be seen from the above embodiments that the execution results of multiple solutions are parsed through a large language model, the execution results of each solution are optimized, the corresponding execution strategies are generated using the optimized execution results of each solution, and the optimal execution strategy is determined through human-computer interaction, thereby improving the rationality and accuracy of the generated execution strategy.

[0124] Figure 2 A flow chart of the drainage facility scheduling method based on a large language model provided in the embodiment of the present application is provided in Figure 1 The drainage facility scheduling system based on the large language model includes: a scene analysis unit, a task identification unit, a task decomposition unit, a resource matching unit, an analysis unit and a strategy unit. The method includes:

[0125] S201: In response to a user inputting a request instruction, the scenario analysis unit extracts keywords from the request instruction and generates instruction keywords.

[0126] S202: The scenario analysis unit performs standardization processing on the instruction keywords, generates standardized task instructions, and calls a database to verify the authenticity of the standardized task instructions.

[0127] S203: The task identification unit reads and parses the standardized task instruction, obtains the task type corresponding to the standardized task instruction, and generates a planned path according to the task type.

[0128] S204: The task decomposition unit identifies the task scope and task subject of the standardized task instruction according to the task planning path, and generates multiple task plans according to the task scope and task subject.

[0129] S205: The task decomposition unit decomposes the multiple task solutions to obtain subtask sets corresponding to the multiple task solutions, and generates corresponding task chains according to the multiple subtasks in the subtask sets.

[0130] S206: The resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain.

[0131] S207: The analysis unit executes multiple subtasks of each solution according to the data resources called by the execution model, and generates execution results of multiple task solutions.

[0132] S208: The strategy unit analyzes the execution results of multiple task plans to determine the scheduling execution plan of the drainage facilities.

[0133] It can be seen from the above embodiments that the scenario analysis unit extracts keywords in the request instruction, generates standardized task instructions based on the keywords, and calls the database to verify the authenticity of the instruction. The task identification unit reads the instruction to obtain the task type, generates a planning path based on the task type, and the task decomposition unit generates multiple task plans, decomposes each task plan to obtain a subtask set of each plan, identifies the task chain in each subtask set, and the resource matching unit matches the execution models corresponding to the multiple subtasks and the data resources called by the execution model according to the task chain. The analysis unit executes the subtasks according to the called data resources and generates the execution results of multiple task plans. The strategy unit parses the execution results to determine the scheduling execution plan of the drainage facilities, and formulates the scheduling plan and strategy of the drainage facility scheduling system based on the large language model without manual formulation, thereby improving the generation efficiency of the drainage system scheduling plan.

[0134] In one embodiment of the present application, the drainage facility scheduling system based on the large language model further includes: a memory unit; after step S208, further includes:

[0135] S209: The memory unit stores the relationship between the tasks and standard instructions, task types, planned paths, task subjects, task plans and resource matching of each unit in the drainage facility scheduling system based on the large language model, as well as the scheduling plan generated by the strategy unit.

[0136] It can be seen from the above embodiments that by storing the data and planned paths of each link in the drainage facility scheduling system in the corresponding database, it is convenient for staff to refer to and optimize the solution when encountering similar problems in the future, thereby improving the efficiency and accuracy of task execution.

[0137] In one embodiment of the present application, after step S202, the following steps are further included:

[0138] S210: The scene analysis unit compares the instruction keyword with the preset task division rule. If the instruction keyword does not meet the task division rule, an information supplement instruction is sent to the user end.

[0139] S211: receiving a supplementary instruction keyword sent by the user terminal, and generating a standardized instruction according to the supplementary instruction keyword.

[0140] S212: Calling a database to verify the authenticity of the standardized instruction.

[0141] From the above embodiments, it can be seen that the scene analysis unit is used to determine whether the instruction keywords meet the task division rules. If the task division rules are not met, the user is prompted to complete the instruction keywords and generate standardized instructions. Standardized instructions are generated through the supplemented instruction keywords, and the observation database is called according to the task object keywords in the instructions to verify whether the events in the task occur, thereby improving the accuracy of generated instructions.

[0142] In one embodiment of the present application, step S206 includes:

[0143] S2061: The resource matching unit calculates the vector similarity between the multiple subtasks of each solution and the model text according to the task chain, and obtains the corresponding execution model according to the vector similarity between the multiple subtasks of each solution and the model text.

[0144] S2062: The resource matching unit matches the data resources called by the execution model according to a preset matching relationship.

[0145] It can be seen from the above embodiments that the vector similarity between each solution subtask and the model introduction text in the model library is calculated through the resource matching unit, the model and module used are matched according to the vector similarity, the data resources required for the execution model are determined according to the preset matching relationship, and the scheduling rules of each subject in the preset controllable subject rule library are called through the large language model to optimize the calculation process and improve the accuracy of the calculation data resources.

[0146] In one embodiment of the present application, after step S207, the following steps are further included:

[0147] S213: The analysis unit reads the running data of multiple subtasks of each solution through the large language model, generates calculation results of multiple task solutions according to the running data of multiple subtasks of each solution, and performs visualization analysis on the calculation results of multiple task solutions to generate visualization results.

[0148] It can be seen from the above embodiments that the operation data corresponding to each plan is read through a large language model, and a visualization plan is generated according to the operation data and the execution results of each plan, so that the staff can intuitively observe the task execution results and adjust the decision according to the task execution results.

[0149] In one embodiment of the present application, step S208 includes:

[0150] S2081: The strategy unit analyzes the execution results of the multiple task solutions through the large language model, optimizes the execution results of the multiple task solutions, and obtains the optimized execution results of the multiple task solutions.

[0151] S2082: The strategy unit inputs the execution results of the optimized multiple task plans into the large language model, and generates a scheduling plan for the drainage facilities through the large language model.

[0152] It can be seen from the above embodiments that the execution results of multiple solutions are parsed through a large language model, the execution results of each solution are optimized, the corresponding execution strategies are generated using the optimized execution results of each solution, and the optimal execution strategy is determined through human-computer interaction, thereby improving the rationality and accuracy of the generated execution strategy.

[0153] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A drainage facility scheduling system based on a large language model, characterized in that: include: Scenario analysis unit, task identification unit, task decomposition unit, resource matching unit, analysis unit and strategy unit; In response to a user inputting a request instruction, the scenario analysis unit extracts keywords from the request instruction to generate instruction keywords; The scenario analysis unit performs standardization processing on the instruction keywords to generate standardized task instructions, and calls a database to verify the authenticity of the standardized task instructions; The task identification unit reads and parses the standardized task instruction, obtains the task type corresponding to the standardized task instruction, and generates a planned path according to the task type; The task decomposition unit identifies the task scope and task subject of the standardized task instruction according to the task planning path, and generates a plurality of task solutions according to the task scope and the task subject; The task decomposition unit decomposes the multiple task solutions to obtain subtask sets corresponding to the multiple task solutions, and generates corresponding task chains according to the multiple subtasks in the subtask sets; The resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain; The analysis unit executes a plurality of subtasks of each scheme according to the data resources called by the execution model, and generates execution results of the plurality of task schemes; The strategy unit analyzes the execution results of the multiple task plans to determine a scheduling execution plan for the drainage facility.

2. The drainage facility scheduling system based on a large language model according to claim 1 is characterized in that: The drainage facility scheduling system based on the large language model further includes: a memory unit; The memory unit stores the relationship between the tasks and standard instructions, task types, planned paths, task subjects, task plans and resource matching of each unit in the drainage facility scheduling system based on the large language model, as well as the scheduling plan generated by the strategy unit.

3. The drainage facility scheduling system based on a large language model according to claim 1 is characterized in that: After the scene analysis unit performs standardization processing on the instruction keywords, the method further includes: The scenario analysis unit compares the instruction keyword with the preset task division rule, and if the instruction keyword does not meet the task division rule, sends an information supplement instruction to the user terminal; receiving a supplementary instruction keyword sent by the user terminal, and generating a standardized instruction according to the supplementary instruction keyword; A database is called to verify the authenticity of the standardized instruction.

4. The drainage facility scheduling system based on a large language model according to claim 1 is characterized in that: The resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain, including: The resource matching unit calculates the vector similarity between the multiple subtasks of each solution and the model text according to the task chain, and obtains the corresponding execution model according to the vector similarity between the multiple subtasks of each solution and the model text; The resource matching unit matches the data resources called by the execution model according to a preset matching relationship.

5. The drainage facility scheduling system based on a large language model according to claim 1 is characterized in that: After the analysis unit executes the multiple subtasks of the schemes according to the data resources called by the execution model and generates the execution results of the multiple task schemes, it also includes: The analysis unit reads the operating data of multiple subtasks for executing each of the schemes through a large language model, generates calculation results of multiple task schemes based on the operating data of multiple subtasks for executing each of the schemes, and performs visualization analysis on the calculation results of the multiple task schemes to generate visualization results.

6. The drainage facility scheduling system based on a large language model according to any one of claims 1 to 5, characterized in that: The strategy unit analyzes the execution results of the multiple task plans to determine the scheduling execution plan of the drainage facility, including: The strategy unit analyzes the execution results of the multiple task solutions through the large language model, optimizes the execution results of the multiple task solutions, and obtains the optimized execution results of the multiple task solutions; The strategy unit inputs the execution results of the optimized multiple task plans into a large language model, and generates a scheduling execution plan for the drainage facilities through the large language model.

7. A drainage facility scheduling method based on a large language model, characterized in that: Applied to a drainage facility scheduling system based on a large language model, the drainage facility scheduling system based on a large language model comprises: a scene analysis unit, a task identification unit, a task decomposition unit, a resource matching unit, an analysis unit and a strategy unit; the method comprises: In response to a user inputting a request instruction, the scenario analysis unit extracts keywords from the request instruction to generate instruction keywords; The scenario analysis unit performs standardization processing on the instruction keywords to generate standardized task instructions, and calls a database to verify the authenticity of the standardized task instructions; The task identification unit reads and parses the standardized task instruction, obtains the task type corresponding to the standardized task instruction, and generates a planned path according to the task type; The task decomposition unit identifies the task scope and task subject of the standardized task instruction according to the task planning path, and generates a plurality of task solutions according to the task scope and the task subject; The task decomposition unit decomposes the multiple task solutions to obtain subtask sets corresponding to the multiple task solutions, and generates corresponding task chains according to the multiple subtasks in the subtask sets; The resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain; The analysis unit executes a plurality of subtasks of each scheme according to the data resources called by the execution model, and generates execution results of the plurality of task schemes; The strategy unit analyzes the execution results of the multiple task plans to determine a scheduling execution plan for the drainage facility.

8. The method according to claim 7, characterized in that The drainage facility scheduling system based on the large language model also includes: a memory unit; the method includes: The memory unit stores the relationship between the tasks and standard instructions, task types, planned paths, task subjects, task plans and resource matching of each unit in the drainage facility scheduling system based on the large language model, as well as the scheduling plan generated by the strategy unit.

9. The method according to claim 7, characterized in that: After the scene analysis unit performs standardization processing on the instruction keywords, the method further includes: The scenario analysis unit compares the instruction keyword with the preset task division rule, and if the instruction keyword does not meet the task division rule, sends an information supplement instruction to the user terminal; receiving a supplementary instruction keyword sent by the user terminal, and generating a standardized instruction according to the supplementary instruction keyword; A database is called to verify the authenticity of the standardized instruction.

10. The method according to claim 7, characterized in that The resource matching unit matches the execution models corresponding to the multiple subtasks of each solution and the data resources called by the execution models according to the task chain, including: The resource matching unit calculates the vector similarity between the multiple subtasks of each solution and the model text according to the task chain, and obtains the corresponding execution model according to the vector similarity between the multiple subtasks of each solution and the model text; The resource matching unit matches the data resources called by the execution model according to a preset matching relationship.

11. The method according to claim 7, characterized in that After the analysis unit executes the multiple subtasks of the schemes according to the data resources called by the execution model and generates the execution results of the multiple task schemes, it also includes: The analysis unit reads the operating data of multiple subtasks for executing each of the schemes through a large language model, generates calculation results of multiple task schemes based on the operating data of multiple subtasks for executing each of the schemes, and performs visualization analysis on the calculation results of the multiple task schemes to generate visualization results.

12. The method according to any one of claims 7 to 11, characterized in that The strategy unit analyzes the execution results of the multiple task plans to determine the scheduling execution plan of the drainage facility, including: The strategy unit analyzes the execution results of the multiple task solutions through the large language model, optimizes the execution results of the multiple task solutions, and obtains the optimized execution results of the multiple task solutions; The strategy unit inputs the execution results of the optimized multiple task plans into a large language model, and generates a scheduling execution plan for the drainage facilities through the large language model.

Images