Large model-based business process intelligent planning method, system, device and medium

Abstract

The application relates to a large model-based business process intelligent planning method, system, device and medium. The method comprises the following steps: acquiring a business text instruction, inputting the business text instruction into a task demand analysis large language model, identifying task key parameter information, and generating a structured task instruction; inputting the structured task instruction into a task decomposer, combining a business knowledge graph to generate an initial sequence of subtasks; according to the business subtask type information, a target business subtask agent is allocated to each business subtask to construct a target business subtask agent entity; according to the subtask data dependency relationship information, a business chain directed acyclic graph is constructed based on the initial sequence of subtasks in combination with a business constraint rule library, and each target business agent entity is called according to the business chain directed acyclic graph to execute a target business, and target business execution result information is generated. The method can generate an intelligent workflow, and improve the flexibility and accuracy of supervision business.

Description

Technical Field

[0001] This application belongs to the field of business process planning, and in particular relates to a business process intelligent planning method, system, device and medium based on a large model. Background Technology

[0002] Business process planning is typically used to analyze business requirements, design execution steps, configure resource elements, and orchestrate task sequences to form an implementable business process plan. The goal of business process planning is to transform abstract business objectives into specific operational sequences, ensuring that various activities are coordinated in terms of time, resources, and logical relationships, and ultimately achieving the expected business output.

[0003] Currently, in key ecological and environmental supervision areas such as ecological red line supervision and nature reserve protection, the monitoring and diagnosis of human activities faces the following bottlenecks: Fragmented business chains: data processing, change identification, risk diagnosis, and report generation are typically completed by multiple independent systems or through manual processes. Workflows are rigid and inflexible, making it difficult to flexibly reorganize and quickly adapt to diverse regulatory scenarios such as "regular inspections across the entire region," "key area verification," and "emergency response to sudden events." Faced with new regulatory tasks or dynamically changing monitoring needs, technical personnel must rewrite code from the ground up and configure complex processes, resulting in low automation levels and heavy reliance on the intervention of domain experts. This leads to excessively long cycles from problem discovery and analysis to the generation of effective regulatory reports. Insufficient system intelligence: existing systems are mostly rigid "pipelines" based on fixed rules or simple tool sets, lacking the ability to understand business semantics. They cannot parse business requirements described in natural language or complex logical combinations, nor can they dynamically and intelligently call and combine the most suitable algorithm models, data resources, and analysis processes according to task objectives, resulting in low levels of automation and intelligence in decision-making. Summary of the Invention

[0004] Therefore, it is necessary to provide a business process intelligent planning method, system, device, and medium based on a large model that can transform traditional fixed data processing processes into flexible and adaptable intelligent workflows to address the aforementioned technical problems.

[0005] Firstly, this application provides a business process intelligent planning method based on a large model, including:

[0006] Obtain the business text instructions of the target business, input the business text instructions into the task requirement analysis big language model, identify the key task parameter information, and generate structured task instructions based on the key task parameter information;

[0007] The structured task instructions are input into the task decomposer, which combines the business knowledge graph to perform knowledge graph traversal and subtask reasoning to generate an initial sequence of subtasks. Each business subtask in the initial sequence of subtasks includes business subtask type information, business subtask target processing data information, and subtask data dependency information.

[0008] Based on the business subtask type information, target business subtask intelligent agents are assigned to each business subtask from the business subtask intelligent agent library, and business subtask target processing data corresponding to the business subtask target processing data information is bound to each target business subtask intelligent agent to construct the target business subtask intelligent agent entity.

[0009] Based on the data dependency information of the subtasks and combined with the business constraint rule base, a directed acyclic graph of the business chain is constructed based on the initial sequence of the subtasks. Then, the target business intelligent entities are called to execute the target business according to the directed acyclic graph of the business chain, and the target business execution result information is generated.

[0010] In one embodiment, the task requirement analysis large language model includes a semantic feature encoding module, a requirement intent recognition module, and a parameter recognition module. It acquires the business text instructions of the target business, inputs the business text instructions into the task requirement analysis large language model, identifies key task parameter information, and generates structured task instructions based on the key task parameter information, including:

[0011] Obtain the business text instructions for the target business and input the business text instructions into the semantic feature encoding module to generate semantic features of the business text instructions;

[0012] The semantic features of the business text instructions are input into the requirement intent recognition module to extract the semantic features of the task requirement intent;

[0013] By concatenating the semantic features of business text instructions and the semantic features of task requirement intent, the enhanced semantic features of business text instructions are obtained.

[0014] The semantic features of the business text instructions are input into the parameter recognition module to generate key task parameter information, and structured task instructions are generated based on the key task parameter information.

[0015] In one embodiment, the business text instructions are input into the task requirements analysis language model to identify key task parameter information. Prior to this, the process also includes:

[0016] Obtain the task requirement training sample set. The task requirement training samples in the task requirement training sample set include sample business text instructions, task requirement intent semantic feature labels of sample business text instructions, and task key parameter labels of sample business text instructions.

[0017] Input the sample business text instructions into the task requirement analysis big language model to generate the task requirement intent semantic features and key task parameter information of the sample business text instructions;

[0018] The task requirement intent semantic feature labels and task key parameter labels of each task requirement training sample in the task requirement training sample set are used as the training output labels of the task requirement analysis big language model. The sample business text instructions of each task requirement training sample are used as the training input of the task requirement analysis big language model. The task requirement analysis big language model is trained by combining the task requirement intent semantic features and task key parameter information of each task requirement training sample.

[0019] In one embodiment, the task key parameter labels include task key parameter type labels and task key parameter semantic feature labels. The task key parameter information includes task key parameter type prediction probability information, task key parameter text data, and the task key parameter semantic features corresponding to the task key parameter text data. The loss function of the task requirement analysis large language model includes a task requirement loss function and a task key parameter loss function. The expressions for the task requirement loss function and the task key parameter loss function are as follows:

[0020]

[0021]

[0022] In the formula, and These are the task requirement loss function and the task key parameter loss function, respectively. The total number of training samples required for the task. and The first Task requirement training samples include task requirement intent semantic feature labels and task requirement intent semantic features. It is the second norm. and These are the type weights of the key task parameters and the numerical weights of the key task parameters, respectively. The total number of task key parameter types corresponding to the task key parameter type label, and the sum of the values ​​are respectively the first and second. The training samples required for this task belong to the first... Task key parameter type, task key parameter type label and first The training samples required for this task belong to the first... Prediction probability information for key parameter types of tasks. and The first The task requires training samples to include semantic feature labels of key parameters and semantic features of key parameters.

[0023] In one embodiment, based on the subtask data dependency information and combined with the business constraint rule base, a directed acyclic graph of the business chain is constructed based on the initial sequence of the subtasks, including:

[0024] Based on each business subtask in the initial sequence of subtasks, the initial business chain nodes are constructed.

[0025] Based on the subtask data dependency information of each business subtask, the initial business chain directed edges are constructed;

[0026] The initial business chain directed acyclic graph is constructed based on the initial business chain nodes and the initial business chain directed edges;

[0027] The initial business chain directed acyclic graph is statically validated based on the business logic rules in the business constraint rule base, and the initial business chain directed acyclic graph that passes the static validation is set as the business chain directed acyclic graph.

[0028] In one embodiment, based on the business subtask type information, a target business subtask intelligent agent is assigned to each business subtask from the business subtask intelligent agent library, and business subtask processing data information is bound to each target business subtask intelligent agent to construct a target business subtask intelligent agent entity, including:

[0029] Based on the business subtask type information and combined with the business subtask applicable type keywords of each business subtask intelligent agent in the business subtask intelligent agent library, candidate business subtask intelligent agents are identified from the business subtask intelligent agent library.

[0030] Obtain the historical business subtask execution success rate and historical business subtask processing speed of each candidate business subtask agent in the business subtask scenario corresponding to the business subtask type information, and obtain the business subtask processing target data adaptation index of each candidate business subtask agent for the business subtask target processing data.

[0031] Based on the historical business subtask execution success rate, historical business subtask processing speed, and business subtask target processing data adaptation index, the business subtask adaptation index of each candidate business subtask agent is calculated.

[0032] Based on the business subtask adaptation index, the target business subtask intelligent agent is identified from each candidate business subtask intelligent agent.

[0033] Bind the target processing data of the business subtask to the intelligent agent of each target business subtask, and construct the intelligent agent entity of the target business subtask.

[0034] In one embodiment, the business subtask adaptation index is a linear sum of the historical business subtask execution success rate, the historical business subtask processing speed, and the business subtask target processing data adaptation index.

[0035] Secondly, this application also provides a business process intelligent planning system based on a large model, including:

[0036] The business text instruction acquisition module is used to acquire the business text instructions of the target business, input the business text instructions into the task requirement analysis big language model, identify the key parameter information of the task, and generate structured task instructions based on the key parameter information of the task.

[0037] The task initial sequence generation module is used to input structured task instructions into the task decomposer, combine them with the business knowledge graph, perform knowledge graph traversal and subtask reasoning, and generate the subtask initial sequence. Each business subtask in the subtask initial sequence includes business subtask type information, business subtask target processing data information, and subtask data dependency information.

[0038] The subtask agent loading module is used to allocate target business subtask agents to each business subtask from the business subtask agent library according to the business subtask type information, and bind the business subtask target processing data corresponding to the business subtask target processing data information to each target business subtask agent to construct the target business subtask agent entity.

[0039] The target business intelligent execution module is used to construct a directed acyclic graph of the business chain based on the data dependency information of the subtasks and the business constraint rule library, and to call each target business intelligent agent entity to execute the target business according to the directed acyclic graph of the business chain, and generate the target business execution result information.

[0040] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method as described in any of the first aspects of this application.

[0041] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects of this application.

[0042] The aforementioned intelligent business process planning methods, systems, devices, and media based on large-scale models can accurately identify the core intent in business text instructions and extract key parameters supporting business execution through large language models. This generates structured task instructions, providing unified and accurate data for subsequent task decomposition and agent matching. Through knowledge graph-driven task decomposition, it accurately locates the type information, target processing data information, and data dependencies of each sub-task, enabling the orderly decomposition and standardized implementation of complex businesses. Through multi-dimensional optimal matching of agents based on sub-task attributes, it accurately matches agents with business sub-tasks, improving the targeting and efficiency of business sub-task execution and significantly enhancing business processing quality and efficiency. By constructing a directed acyclic graph of the business chain combining sub-task dependency parsing and business constraint rules, it ensures the compliance and logical correctness of business processes, avoiding process conflicts and interruptions during execution. This enables dynamic adaptation and efficient implementation of business processes, allowing target businesses to quickly respond to changing needs under standardized constraints, improving the flexibility and accuracy of regulatory processes. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart illustrating a business process intelligent planning method based on a large model, provided as an embodiment of this application. Figure 1 ;

[0045] Figure 2 A flowchart illustrating a business process intelligent planning method based on a large model, provided as an embodiment of this application. Figure 2 ;

[0046] Figure 3 This is a schematic diagram of the structure of a business process intelligent planning system based on a large model, provided as an embodiment of this application. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0048] In one exemplary embodiment of this application, such as Figure 1As shown, a business process intelligent planning method based on a large model is provided. This embodiment illustrates the application of this method to a process planning terminal. It is understood that this method can also be applied to a process planning server, and further to a process planning system including both a process planning terminal and a process planning server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0049] Step S101: Obtain the business text instructions of the target business, input the business text instructions into the task requirement analysis large language model, identify the key task parameter information, and generate structured task instructions based on the key task parameter information.

[0050] Specifically, the process planning terminal can acquire business text instructions input by supervisors or business personnel. The terminal can then input these instructions into a task requirements analysis language model to identify key task parameters within the instructions. Based on these key parameters, the task requirements analysis language model can generate structured task instructions corresponding to the business text instructions.

[0051] To illustrate, the large language model for task requirements analysis can be mounted on a process planning terminal, a process planning server, or a process planning system.

[0052] Optionally, the process planning terminal can input the acquired business text instructions into the semantic feature encoding module of the task requirement analysis large language model to generate semantic features of the business text instructions. The process planning terminal can then input these semantic features into the requirement intent recognition module of the task requirement analysis large language model to extract the task requirement intent semantic features of the business text instructions. The process planning terminal can then concatenate the semantic features of the business text instructions and the task requirement intent semantic features to obtain enhanced semantic features of the business text instructions. Finally, the process planning terminal can input these enhanced semantic features into the parameter recognition module of the task requirement analysis large language model to generate key task parameter information for the business text instructions, and based on this key task parameter information, generate structured task instructions for the business text instructions.

[0053] Step S102: Input the structured task instructions into the task decomposer, combine them with the business knowledge graph, perform knowledge graph traversal and subtask reasoning, and generate the initial sequence of subtasks.

[0054] Optionally, each business subtask in the initial sequence of subtasks may include business subtask type information, business subtask target processing data information, and subtask data dependency information.

[0055] Specifically, the process planning terminal can input structured task instructions into the task decomposer, which will parse the core execution objectives, key constraints, and core parameter ranges of the target business. The process planning terminal can then match these core execution objectives, key constraints, and core parameter ranges with the business knowledge graph.

[0056] Optionally, the process planning terminal can locate the execution target node corresponding to the core execution target of the target business in the business knowledge graph according to the business domain corresponding to the structured task instruction. Then, it can traverse all sub-concepts, related tasks and data dependencies related to the target business along the association edges between the execution target nodes to comprehensively sort out the execution logic and required links of the target business.

[0057] Furthermore, the process planning terminal can use the task decomposer to break down the core execution target into multiple atomic business subtasks with independent execution attributes based on the logical rules in the business knowledge graph. According to the core execution target, key constraints and core parameter range of the target business, combined with the business knowledge graph, the business subtask type information, business subtask target processing data information and subtask data dependency information of each business subtask are set to create the initial sequence of subtasks.

[0058] To illustrate, a business knowledge graph can store all business rules, domain knowledge, and component information, providing a logical basis for task decomposition and process assembly. A business knowledge graph can store domain knowledge through a graph structure.

[0059] Step S103: Based on the business subtask type information, assign target business subtask intelligent agents to each business subtask from the business subtask intelligent agent library, and bind the business subtask target processing data corresponding to the business subtask target processing data information to each target business subtask intelligent agent to construct the target business subtask intelligent agent entity.

[0060] Specifically, the process planning terminal can allocate target business subtask intelligent agents to each business subtask based on the business subtask type information of each business subtask in the initial sequence of subtasks, combined with the intelligent agent evaluation indicators of the business subtask intelligent agents in the intelligent agent description library corresponding to the business subtask intelligent agent library, and bind the business subtask target processing data corresponding to the business subtask target processing data information of the business subtask to each target business subtask intelligent agent, thereby constructing the target business subtask intelligent agent entity.

[0061] For example, the agent evaluation metrics for business subtask agents may include, but are not limited to, historical business subtask execution success rate, historical business subtask processing speed, and data adaptation metrics.

[0062] Step S104: Based on the subtask data dependency information and the business constraint rule library, a directed acyclic graph of the business chain is constructed based on the initial sequence of the subtasks. Then, the target business intelligent agents are called to execute the target business according to the directed acyclic graph of the business chain, and the target business execution result information is generated.

[0063] Specifically, the process planning terminal can construct a directed acyclic graph (DAG) of the business chain based on the initial sequence of subtasks, according to the subtask data dependency information and the business constraint rule base. It then calls upon each target business intelligent agent entity to execute the target business based on the DAG, generating the target business execution result information. The business constraint rule base can store business logic rules.

[0064] To illustrate, the data format of the target business execution result information can be flexibly configured according to business needs. The data format of the target business execution result information may include, but is not limited to, structured reports, data tables, and visualization charts.

[0065] To illustrate, the target business execution result information may include the execution process data of the target business, the execution conclusion information of the target business, and the execution details of each business sub-task.

[0066] In the aforementioned intelligent business process planning method based on a large model, the core intent in business text instructions can be accurately identified through a large language model, and key parameters supporting business execution can be extracted. This generates structured task instructions, providing a unified and accurate data basis for subsequent task decomposition and agent matching. Through knowledge graph-driven task decomposition, the type information, target processing data information, and data dependencies of each sub-task can be accurately located, enabling the orderly decomposition and standardized implementation of complex businesses. Through multi-dimensional optimal matching of agents based on sub-task attributes, agents and business sub-tasks can be accurately matched, improving the targeting and efficiency of executing business sub-tasks and significantly improving business processing quality and efficiency. Through the construction of a directed acyclic graph of the business chain combining sub-task dependency parsing and business constraint rules, the compliance and logical correctness of the business process can be ensured, avoiding process conflicts and interruptions during execution. This enables dynamic adaptation and efficient implementation of the business process, allowing the target business to quickly respond to changes in demand under standardized constraints, improving the flexibility and accuracy of regulatory business.

[0067] In an optional embodiment of this application, the task requirement analysis large language model may include a semantic feature encoding module, a requirement intent recognition module, and a parameter recognition module. Please refer to [reference needed]. Figure 1 and Figure 2Step S101: Obtain the business text instructions for the target business, input the business text instructions into the task requirement analysis large language model, identify the key task parameter information, and generate structured task instructions based on the key task parameter information, which may include:

[0068] Step S201: Obtain the business text instructions of the target business and input the business text instructions into the semantic feature encoding module to generate semantic features of the business text instructions.

[0069] Step S202: Input the semantic features of the business text instruction into the requirement intent recognition module to extract the semantic features of the task requirement intent.

[0070] Step S203: Concatenate the semantic features of the business text instruction and the semantic features of the task requirement intent to obtain the enhanced semantic features of the business text instruction.

[0071] Step S204: Input the enhanced semantic features of the business text instruction into the parameter recognition module to generate key task parameter information, and generate structured task instructions based on the key task parameter information.

[0072] In an optional embodiment of this application, before inputting business text instructions into the task requirement analysis large language model to identify key task parameter information, the following steps may also be included:

[0073] Specifically, the process planning terminal can obtain a training sample set of task requirements.

[0074] Optionally, the task requirement training samples in the task requirement training sample set include sample business text instructions, task requirement intent semantic feature labels of sample business text instructions, and task key parameter labels of sample business text instructions.

[0075] Specifically, the process planning terminal can input sample business text instructions into the task requirement analysis big language model to generate the task requirement intent semantic features and key task parameter information of the sample business text instructions.

[0076] Specifically, the process planning terminal can use the semantic feature labels of the task requirement intent and the key parameter labels of each task requirement training sample in the task requirement training sample set as the training output labels of the task requirement analysis big language model. The process planning terminal can use the sample business text instructions of each task requirement training sample as the training input of the task requirement analysis big language model, and train the task requirement analysis big language model by combining the semantic feature of the task requirement intent and the key parameter information of each task requirement training sample.

[0077] In an optional embodiment of this application, the task key parameter label may include a task key parameter type label and a task key parameter semantic feature label. The task key parameter information may include task key parameter type prediction probability information, task key parameter text data, and the task key parameter semantic features corresponding to the task key parameter text data. The loss function of the task requirement analysis large language model may include a task requirement loss function and a task key parameter loss function. The expressions for the task requirement loss function and the task key parameter loss function may be:

[0078]

[0079]

[0080] In the formula, and These are the task requirement loss function and the task key parameter loss function, respectively. The total number of training samples required for the task. and The first The task requirements training samples include the task requirements, intent, semantic features, labels, and the first task. The task requires semantic features of the training samples. It is the second norm. and These are the type weights of the key task parameters and the numerical weights of the key task parameters, respectively. The total number of task key parameter types corresponding to the task key parameter type label, and the sum of the values ​​are respectively the first and second. The training samples required for this task belong to the first... Task key parameter type, task key parameter type label and first The training samples required for this task belong to the first... Prediction probability information for key parameter types of tasks. and The first The task requires training samples with key parameters, semantic features, and labels. The task requires semantic features of key parameters of the training samples.

[0081] In an optional embodiment of this application, a directed acyclic graph of the business chain is constructed based on the initial sequence of subtasks, according to the subtask data dependency information and in conjunction with the business constraint rule base. This may include:

[0082] Specifically, the process planning terminal can construct the initial business chain nodes based on each business subtask in the initial sequence of subtasks.

[0083] Specifically, the process planning terminal can construct the initial business chain directed edges based on the subtask data dependency information of each business subtask.

[0084] Specifically, the process planning terminal can construct a directed acyclic graph of the initial business chain based on the initial business chain nodes and the directed edges of the initial business chain.

[0085] Specifically, the process planning terminal can perform static verification on the initial business chain directed acyclic graph based on the business logic rules in the business constraint rule base, and set the initial business chain directed acyclic graph that passes the static verification as the business chain directed acyclic graph.

[0086] In an optional embodiment of this application, please refer to Figure 1 and Figure 2 Step S103: Based on the business subtask type information, assign target business subtask intelligent agents to each business subtask from the business subtask intelligent agent library, and bind business subtask processing data information to each target business subtask intelligent agent to construct the target business subtask intelligent agent entity, which may include:

[0087] Step S206: Based on the business subtask type information and combined with the business subtask applicable type keywords of each business subtask intelligent agent in the business subtask intelligent agent library, candidate business subtask intelligent agents are identified from the business subtask intelligent agent library.

[0088] Step S207: Obtain the historical business subtask execution success rate and historical business subtask processing speed of each candidate business subtask agent in the business subtask scenario corresponding to the business subtask type information, and obtain the business subtask processing target data adaptation index of each candidate business subtask agent for the business subtask target processing data.

[0089] Step S208: Based on the historical business subtask execution success rate, historical business subtask processing speed, and business subtask target processing data adaptation index, calculate the business subtask adaptation index of each candidate business subtask agent.

[0090] Step S209: Identify the target business subtask agent from each candidate business subtask agent based on the business subtask adaptation index.

[0091] Step S210: Bind the business subtask target processing data corresponding to the business subtask target processing data information of the business subtask to each target business subtask intelligent agent, and construct the target business subtask intelligent agent entity.

[0092] In an optional embodiment of this application, the business subtask adaptation index can be a linear sum of the historical business subtask execution success rate, the historical business subtask processing speed, and the business subtask target processing data adaptation index.

[0093] Specifically, the process planning terminal can linearly sum the historical business sub-task execution success rate, historical business sub-task processing speed, and business sub-task target processing data adaptation index to calculate the business sub-task adaptation index.

[0094] In one exemplary embodiment of this application, such as Figure 2 As shown, a business process intelligent planning method based on a large model is provided, including:

[0095] Step S201: Obtain the business text instructions of the target business and input the business text instructions into the semantic feature encoding module to generate semantic features of the business text instructions.

[0096] Step S202: Input the semantic features of the business text instruction into the requirement intent recognition module to extract the semantic features of the task requirement intent.

[0097] Step S203: Concatenate the semantic features of the business text instruction and the semantic features of the task requirement intent to obtain the enhanced semantic features of the business text instruction.

[0098] Step S204: Input the enhanced semantic features of the business text instruction into the parameter recognition module to generate key task parameter information, and generate structured task instructions based on the key task parameter information.

[0099] Step S205: Input the structured task instructions into the task decomposer, combine them with the business knowledge graph, perform knowledge graph traversal and subtask reasoning, and generate the initial sequence of subtasks.

[0100] Step S206: Based on the business subtask type information and combined with the business subtask applicable type keywords of each business subtask intelligent agent in the business subtask intelligent agent library, candidate business subtask intelligent agents are identified from the business subtask intelligent agent library.

[0101] Step S207: Obtain the historical business subtask execution success rate and historical business subtask processing speed of each candidate business subtask agent in the business subtask scenario corresponding to the business subtask type information, and obtain the business subtask processing target data adaptation index of each candidate business subtask agent for the business subtask target processing data.

[0102] Step S208: Based on the historical business subtask execution success rate, historical business subtask processing speed, and business subtask target processing data adaptation index, calculate the business subtask adaptation index of each candidate business subtask agent.

[0103] Step S209: Identify the target business subtask agent from each candidate business subtask agent based on the business subtask adaptation index.

[0104] Step S210: Bind the business subtask target processing data corresponding to the business subtask target processing data information of the business subtask to each target business subtask intelligent agent, and construct the target business subtask intelligent agent entity.

[0105] Step S211: Based on the subtask data dependency information and the business constraint rule library, a directed acyclic graph of the business chain is constructed based on the initial sequence of the subtasks. Then, the target business intelligent entities are called to execute the target business according to the directed acyclic graph of the business chain, and the target business execution result information is generated.

[0106] The aforementioned intelligent business process planning method based on a large model effectively addresses the ambiguity and vagueness of natural language business requirements by accurately extracting key task parameters and generating structured task instructions. It achieves reasonable decomposition and orderly arrangement of complex business processes by combining business knowledge graph traversal and sub-task reasoning. Furthermore, it ensures precise adaptation between business sub-tasks and intelligent agents by selecting target intelligent agents based on their sub-task type information, execution success rate, processing speed, and data adaptability, and binding them to processing data. Finally, it guarantees the logical compliance and robustness of the business process by constructing a directed acyclic graph of the business chain based on sub-task data dependencies and a business constraint rule base, thereby efficiently generating target business execution results and comprehensively improving the accuracy, adaptability, and efficiency of business process planning and execution.

[0107] In one exemplary embodiment of this application, a business process intelligent planning method based on a large model is provided, comprising:

[0108] For example, a supervisor might input a natural language command into the system interface: "Monitor any newly constructed artificial structures in the Yellow River Delta National Nature Reserve within the past month and analyze whether they are located within the core area." The process planning terminal can receive the natural language command for the target business submitted by the supervisor through the multimodal input module of the ecological supervision business system.

[0109] Furthermore, the process planning terminal can parse the natural language instructions submitted by the supervisors through the task requirement analysis big language model, and identify the core intent as "change monitoring and compliance analysis". The process planning terminal can extract the key parameter information of the task by combining the identified core intent "change monitoring and compliance analysis" with the task requirement analysis big language model.

[0110] For example, the key parameters of the task may include, but are not limited to, geographical area information, time range information, target object information, and analysis action information.

[0111] Furthermore, the process planning terminal can generate structured task description objects and structured task instructions based on key task parameter information. For example: "Region: Yellow River Delta National Nature Reserve, Time range: the past month, Target object: newly added artificial buildings, Analysis action: compliance analysis of the core area."

[0112] Furthermore, the process planning terminal can use a task decomposer to match and reason with the structured task instructions and the business knowledge graph, and then generate a sequence of business subtasks.

[0113] For example, a sequence of business subtasks may include:

[0114] Business Subtask A: Obtain the boundary vector data of the protected area.

[0115] Subtask B: Acquire two sets of high-resolution satellite imagery data within the protected area, one from today and one month ago.

[0116] Subtask C: Perform change detection on the two phases of imagery and extract all surface patches that have changed.

[0117] Business Subtask D: Classify the changed patches by land features and select the patches that belong to the "building" category.

[0118] Business Subtask E: Overlay the newly added building patches with the boundary of the core area of ​​the protected area to determine whether each patch is located within the core area.

[0119] Business Subtask F: Integrate all the above results to generate a comprehensive monitoring report that includes information such as the location, area, and compliance status of the map features.

[0120] Furthermore, the process planning terminal can use a tool matcher to assign a target business subtask agent to each business subtask from the business subtask agent library based on the business subtask type information of each business subtask.

[0121] For example, business subtasks A and B can be matched with a spatial data acquisition agent, which is adept at connecting to remote sensing data cloud platforms and GIS databases and automatically downloading data according to their parameters (region, time). Business subtask C can be matched with a remote sensing change detection agent, which can encapsulate pixel-level change detection algorithms. Business subtask D can be matched with a deep learning ground feature classification agent, which can use a pre-trained convolutional neural network model to classify images. Business subtask E can be matched with a GIS spatial analysis agent, which can specifically perform spatial query and overlay analysis operations. Business subtask F can be matched with a report generation agent, which can populate structured data into preset Word or PDF templates.

[0122] Furthermore, the process planning terminal can instantiate each selected target subtask agent and bind specific parameters of the corresponding business subtask (e.g., specifying "Yellow River Delta Nature Reserve" and specific date for the spatial data acquisition agent).

[0123] Furthermore, the process planning terminal can intelligently analyze the dependencies between business subtasks through the process assembler. For example, it can discover that business subtask C requires the outputs of business subtasks A and B, business subtask D requires the output of business subtask C, and business subtask E requires the outputs of business subtasks A and D.

[0124] Furthermore, the process planning terminal can generate a directed acyclic graph of the business chain.

[0125] For example, business subtask A can run in parallel with business subtask B; after business subtasks A and B are completed, business subtask C is executed; after business subtask C is completed, business subtask D is executed; after business subtasks A and D are completed, business subtask E is executed; after business subtask E is completed, business subtask F is executed. The process validator checks this business chain as a directed acyclic graph to confirm that the logic is correct and the parameters are complete.

[0126] Furthermore, the process planning terminal can simultaneously launch target subtask agent A and target subtask agent B. The process planning terminal can use the data flow manager to pass the imagery downloaded by target subtask agent B and the boundary data acquired by target subtask agent A to target subtask agent C. Target subtask agent C performs change detection, and the output changed patches are passed to target subtask agent D for feature classification. The "newly added building patches" selected by target subtask agent D and the "core area boundary" provided by target subtask agent A are passed together to target subtask agent E for overlay analysis. The analysis results output by target subtask agent E (e.g., "3 new buildings were found, 1 of which is located in the core area") are passed to target subtask agent F. Target subtask agent F can automatically call a template, generate a detailed monitoring report, and push it to the supervisors through the user interface.

[0127] In the aforementioned intelligent business process planning method based on large models, by constructing a business knowledge base and an intelligent agent component library, it is possible to achieve intelligent modeling and dynamic assembly of human activity monitoring business processes. This can transform traditional fixed data processing processes into flexible and adaptable intelligent workflows, thereby parsing natural language requirements, automatically planning task sequences based on knowledge graphs, matching the optimal processing intelligent agent for each sub-task, and finally assembling them into an executable workflow, thereby improving the agility and accuracy of regulatory business.

[0128] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0129] Based on the same inventive concept, this application also provides a large-model-based intelligent business process planning system for implementing the above-mentioned large-model-based intelligent business process planning method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the large-model-based intelligent business process planning system provided below can be found in the limitations of the large-model-based intelligent business process planning method described above, and will not be repeated here.

[0130] In one exemplary embodiment, such as Figure 3 As shown, a business process intelligent planning system 300 based on a large model is provided, including:

[0131] The business text instruction acquisition module 301 can be used to acquire the business text instructions of the target business, input the business text instructions into the task requirement analysis large language model, identify the key parameter information of the task, and generate structured task instructions based on the key parameter information of the task.

[0132] The task initial sequence generation module 302 can be used to input structured task instructions into the task decomposer, combine them with a business knowledge graph, perform knowledge graph traversal and subtask reasoning, and generate a subtask initial sequence. Each business subtask in the subtask initial sequence includes business subtask type information, business subtask target processing data information, and subtask data dependency information.

[0133] The subtask agent loading module 303 can be used to allocate target business subtask agents to each business subtask from the business subtask agent library according to the business subtask type information, and bind the business subtask target processing data corresponding to the business subtask target processing data information to each target business subtask agent to construct the target business subtask agent entity.

[0134] The target business intelligent execution module 304 can be used to construct a directed acyclic graph of the business chain based on the initial sequence of the sub-tasks, according to the sub-task data dependency information and the business constraint rule library, and call each target business intelligent agent entity to execute the target business according to the directed acyclic graph of the business chain, and generate target business execution result information.

[0135] In an optional embodiment of this application, the business text instruction acquisition module 301 can also be used for:

[0136] Obtain the business text instructions for the target business and input the business text instructions into the semantic feature encoding module to generate semantic features of the business text instructions.

[0137] The semantic features of the business text instructions are input into the requirement intent recognition module to extract the semantic features of the task requirement intent.

[0138] By concatenating the semantic features of business text instructions and the semantic features of task requirement intent, the enhanced semantic features of business text instructions are obtained.

[0139] The semantic features of the business text instructions are input into the parameter recognition module to generate key task parameter information, and structured task instructions are generated based on the key task parameter information.

[0140] In an optional embodiment of this application, the business process intelligent planning system 300 based on a large model can also be used for:

[0141] Obtain the task requirement training sample set, which includes sample business text instructions, task requirement intent semantic feature labels of sample business text instructions, and task key parameter labels of sample business text instructions.

[0142] Input the sample business text instructions into the task requirement analysis language model to generate the task requirement intent semantic features and key task parameter information of the sample business text instructions.

[0143] The task requirement intent semantic feature labels and task key parameter labels of each task requirement training sample in the task requirement training sample set are used as the training output labels of the task requirement analysis big language model. The sample business text instructions of each task requirement training sample are used as the training input of the task requirement analysis big language model. The task requirement analysis big language model is trained by combining the task requirement intent semantic features and task key parameter information of each task requirement training sample.

[0144] In an optional embodiment of this application, the target business intelligent execution module 304 can also be used for:

[0145] The initial business chain nodes are constructed based on each business subtask in the initial sequence of subtasks.

[0146] Based on the subtask data dependency information of each business subtask, the initial business chain directed edges are constructed.

[0147] The initial business chain is constructed by using the initial business chain nodes and the initial business chain directed edges.

[0148] The initial business chain directed acyclic graph is statically validated based on the business logic rules in the business constraint rule base, and the initial business chain directed acyclic graph that passes the static validation is set as the business chain directed acyclic graph.

[0149] In an optional embodiment of this application, the subtask agent loading module 303 can also be used for:

[0150] Based on the business subtask type information and combined with the business subtask applicable type keywords of each business subtask agent in the business subtask agent library, candidate business subtask agents are identified from the business subtask agent library.

[0151] Obtain the historical business subtask execution success rate and historical business subtask processing speed of each candidate business subtask agent in the business subtask scenario corresponding to the business subtask type information, and obtain the business subtask processing target data adaptation index of each candidate business subtask agent for the business subtask target processing data.

[0152] Based on the historical business subtask execution success rate, historical business subtask processing speed, and business subtask target processing data adaptation index, the business subtask adaptation index of each candidate business subtask agent is calculated.

[0153] Based on the business subtask adaptation index, the target business subtask agent is identified from each candidate business subtask agent.

[0154] Bind the target processing data of the business subtask to the intelligent agent of each target business subtask, and construct the intelligent agent entity of the target business subtask.

[0155] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the business process intelligent planning method based on a large model as described above.

[0156] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0157] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0158] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A business process intelligent planning method based on a large model, characterized in that, The method includes: Obtain the business text instructions of the target business, input the business text instructions into the task requirement analysis large language model, identify the key task parameter information, and generate structured task instructions based on the key task parameter information; The structured task instructions are input into the task decomposer, and combined with the business knowledge graph, knowledge graph traversal and subtask reasoning are performed to generate an initial sequence of subtasks; wherein, each business subtask in the initial sequence of subtasks includes business subtask type information, business subtask target processing data information, and subtask data dependency information. Based on the business subtask type information, a target business subtask intelligent agent is assigned to each business subtask from the business subtask intelligent agent library, and the business subtask target processing data corresponding to the business subtask target processing data information is bound to each target business subtask intelligent agent to construct a target business subtask intelligent agent entity. Based on the subtask data dependency information and combined with the business constraint rule library, a directed acyclic graph of the business chain is constructed based on the initial sequence of the subtasks. Then, each of the target business intelligent agents is called to execute the target business according to the directed acyclic graph of the business chain, and the target business execution result information is generated.

2. The method according to claim 1, characterized in that, The task requirement analysis large language model includes a semantic feature encoding module, a requirement intent recognition module, and a parameter recognition module. The process involves acquiring the target business text instructions, inputting these instructions into the task requirement analysis large language model, identifying key task parameter information, and generating structured task instructions based on this key parameter information. Obtain the business text instructions of the target business and input the business text instructions into the semantic feature encoding module to generate semantic features of the business text instructions; The semantic features of the business text instruction are input into the requirement intent recognition module to extract the semantic features of the task requirement intent; By concatenating the semantic features of the business text instruction and the semantic features of the task requirement intent, an enhanced semantic feature of the business text instruction is obtained. The enhanced semantic features of the business text instruction are input into the parameter recognition module to generate key task parameter information, and the structured task instruction is generated based on the key task parameter information.

3. The method according to claim 2, characterized in that, Before inputting the business text instructions into the task requirement analysis language model to identify key task parameter information, the process also includes: Obtain a task requirement training sample set, wherein the task requirement training samples in the task requirement training sample set include sample business text instructions, task requirement intent semantic feature labels of the sample business text instructions, and task key parameter labels of the sample business text instructions. The sample business text instruction is input into the task requirement analysis big language model to generate the task requirement intent semantic features and the task key parameter information of the sample business text instruction; The task requirement intention semantic feature labels and task key parameter labels of each task requirement training sample in the task requirement training sample set are used as the training output labels of the task requirement analysis big language model. The sample business text instructions of each task requirement training sample are used as the training input of the task requirement analysis big language model. The task requirement analysis big language model is trained by combining the task requirement intention semantic features and task key parameter information of each task requirement training sample.

4. The method according to claim 3, characterized in that, The task key parameter labels include task key parameter type labels and task key parameter semantic feature labels. The task key parameter information includes task key parameter type prediction probability information, task key parameter text data, and the task key parameter semantic features corresponding to the task key parameter text data. The loss function of the task requirement analysis large language model includes a task requirement loss function and a task key parameter loss function. The expressions for the task requirement loss function and the task key parameter loss function are as follows: In the formula, and These are the task requirement loss function and the task key parameter loss function, respectively. The total number of training samples for the task requirements. and The first The task requirement intent semantic feature labels and task requirement intent semantic features of the training samples described above. It is the second norm. and These are the type weights of the key task parameters and the numerical weights of the key task parameters, respectively. The total number of task key parameter types corresponding to the task key parameter type labels, and the sum of these numbers are respectively the first and second. The training samples for the task requirements described in the item belong to the first... The task key parameter type label and the first task key parameter type of the class task key parameter type The training samples for the task requirements described in the item belong to the first... The predicted probability information for the key parameter type of the task. and The first The task requirements training samples described herein include the semantic feature labels of the task key parameters and the semantic features of the task key parameters.

5. The method according to claim 1, characterized in that, The step of constructing a directed acyclic graph of the business chain based on the subtask data dependency information and the business constraint rule base, based on the initial sequence of the subtasks, includes: An initial business chain node is constructed based on each of the business subtasks in the initial sequence of subtasks; Based on the subtask data dependency information of each of the aforementioned business subtasks, an initial business chain directed edge is constructed; A directed acyclic graph of the initial business chain is constructed based on the initial business chain nodes and the initial business chain directed edges; The initial business chain directed acyclic graph is statically validated based on the business logic rules in the business constraint rule base, and the initial business chain directed acyclic graph that passes the static validation is set as the business chain directed acyclic graph.

6. The method according to any one of claims 1 to 5, characterized in that, The step of allocating target business subtask intelligent agents to each business subtask from the business subtask intelligent agent library according to the business subtask type information, and binding the business subtask target processing data corresponding to the business subtask target processing data information to each target business subtask intelligent agent to construct the target business subtask intelligent agent entity includes: Based on the business subtask type information and combined with the business subtask applicable type keywords of each business subtask intelligent agent in the business subtask intelligent agent library, candidate business subtask intelligent agents are identified from the business subtask intelligent agent library. Obtain the historical business subtask execution success rate and historical business subtask processing speed of each candidate business subtask agent in the business subtask scenario corresponding to the business subtask type information, and obtain the business subtask processing target data adaptation index of each candidate business subtask agent for the business subtask target processing data. Based on the historical business sub-task execution success rate, the historical business sub-task processing speed, and the business sub-task target processing data adaptation index, the business sub-task adaptation index of each candidate business sub-task agent is calculated. Based on the business subtask adaptation index, the target business subtask agent of the business subtask is identified from each of the candidate business subtask agents. The target business subtask intelligent agent is constructed by binding the target processing data of the business subtask to the target processing data information of the business subtask for each target business subtask intelligent agent.

7. The method according to claim 6, characterized in that: The business subtask adaptation index is a linear sum of the historical business subtask execution success rate, the historical business subtask processing speed, and the business subtask target processing data adaptation index.

8. A business process intelligent planning system based on a large model, characterized in that, The system includes: The business text instruction acquisition module is used to acquire the business text instructions of the target business, input the business text instructions into the task requirement analysis large language model, identify the key task parameter information, and generate structured task instructions based on the key task parameter information. The task initial sequence generation module is used to input the structured task instructions into the task decomposer, combine the business knowledge graph, perform knowledge graph traversal and subtask reasoning, and generate a subtask initial sequence; wherein, each business subtask in the subtask initial sequence includes business subtask type information, business subtask target processing data information, and subtask data dependency information; The subtask agent loading module is used to allocate target business subtask agents to each business subtask from the business subtask agent library according to the business subtask type information, and bind the business subtask target processing data corresponding to the business subtask target processing data information to each target business subtask agent to construct the target business subtask agent entity. The target business intelligent execution module is used to construct a directed acyclic graph of the business chain based on the data dependency information of the sub-tasks and the business constraint rule library, and to call each of the target business intelligent entities to execute the target business according to the directed acyclic graph of the business chain, thereby generating target business execution result information.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images