Power marketing auxiliary work order processing method and system based on large language model

Abstract

The invention relates to an electric power marketing auxiliary work order processing method and system based on a large language model, belongs to the technical field of electric power marketing work order information processing, and is used for solving the problems that unified structured representation and semantic association are lacked and a feedback optimization closed loop is lacked in the prior art. The method comprises the following steps: constructing an interface knowledge graph, collecting and analyzing multi-source interface information and extracting a business entity relationship; analyzing user input based on the pre-training language model, identifying an intention and generating a multi-level dynamic task plan; interface parameters are generated through mapping of the knowledge graph, and safe and reliable interface calling and execution result verification feedback are achieved. By means of the method, unified structured representation and semantic association are achieved, and an optimization closed loop is fed back.

Description

Technical Field

[0001] This invention belongs to the field of information processing technology for power marketing work orders, specifically relating to a power marketing auxiliary work order processing method and system based on a large language model. Background Technology

[0002] Currently, with the widespread adoption of smart grids and the deep implementation of next-generation marketing systems, electricity marketing is rapidly evolving towards digitalization and refinement. The current Electricity Marketing 2.0 system primarily relies on preset rule engines and robotic process automation (Robotic Process Automation) technology for work order processing. This system heavily depends on manually written, fixed process scripts, lacking deep semantic understanding of user natural language requests. It can only perform simple keyword matching, requiring users to input in strict formats, resulting in a poor user experience and limited service efficiency. Furthermore, the knowledge management system has deficiencies, lacking a unified structured representation and semantic association.

[0003] The invention patent with publication number CN120849606A discloses a method for extracting key elements of power marketing business based on natural language processing technology. The method includes: Step 1, defining category labels for power marketing business, labeling business work order texts with category labels, and constructing a labeled work order dataset; extracting feature vectors from the work order texts using the TF-IDF method, inputting them into a support vector machine for training, and achieving automatic identification of work order business types; Step 2, collecting and preprocessing power marketing business work order texts, using Word2Vec to convert each word in the work order text into a fixed-dimensional vector, using a BiLSTM model to obtain the contextual information of the words, and using a conditional random field to apply the BiLS... Step 3 involves labeling the word sequences output by the TM model to automatically identify and label entities of different categories. Step 4 utilizes the contextual embedding representation of the RoBERTa model, combined with a pre-defined business hierarchy classification structure, to construct a multi-level element classification model. This enables fine-grained extraction of key elements in the power marketing business. A deep neural network entity relationship extraction model is built based on an attention mechanism to extract the relationships between key elements. Step 5 uses a similarity algorithm to calculate the similarity between key elements in the power marketing business and key elements in policy documents. The TextRank algorithm is used to extract keywords from the key elements in policy documents. A policy basis summary is generated based on an improved Seq2Seq model, and an optimization mechanism is used to improve the summary quality. This existing technology has the following drawbacks: a lack of unified structured representation and semantic association, and a lack of feedback optimization loops. These are the shortcomings of the existing technology.

[0004] In view of this, it is very necessary to provide a power marketing auxiliary work order processing method and system based on a large language model to solve the above-mentioned defects in the prior art. Summary of the Invention

[0005] To address the technical problems of existing technologies, such as the lack of unified structured representation and semantic association, and the lack of feedback optimization loops, this invention provides a power marketing auxiliary work order processing method and system based on a large language model to solve the above-mentioned technical problems.

[0006] In a first aspect, the present invention provides a method for processing auxiliary work orders in electricity marketing based on a large language model, comprising: Step S1: The steps to build an interface knowledge graph include collecting and parsing multi-source interface information, building a unified interface metadata model, extracting business entities and their interface relationships, and building the knowledge graph. Multi-source interface information includes front-end components, API documentation, and network request logs from the Electricity Marketing 2.0 system; Parsing multi-source interface information includes extracting relevant attributes and data binding relationships from front-end components, API documentation, and interface elements using parsing libraries and tools; Use log analysis tools to identify API call patterns in network request logs and group these patterns accordingly. During the grouping process, parameter passing is analyzed. By calculating and comparing the parameter similarity between the request and response bodies, parameter dependencies are identified. The mathematical expression for parameter similarity is:

[0007] in, These are the parameter sets for comparing the request parameter set and the parameter set for the response body, respectively. To compare request parameters, For response parameters, For the Kronecker delta function, The number of parameters; Building a unified interface metadata model involves using parsing tools to extract interface metadata from API documentation and extracting network request interaction data from network logs, and then integrating these two types of data to construct the interface metadata model. Define the interface metadata model as JSON Schema format. The interface metadata types include interface endpoint, request method, parameter structure, authentication method, and return format. Business entities and their interface relationships are extracted to build a domain entity model. NLP tools are used to identify business entities from work order texts and API documents. The results are analyzed in conjunction with the parameter passing mechanism to establish corresponding relationships between entity types, interfaces, and parameters. Identify task templates and exception handling patterns in historical work order data, group the work orders, and use the support of the task template to represent the frequency of the task in historical work orders. The mathematical expression is:

[0008] Where T is the task template, Number of tickets containing T is the number of work orders containing task template T, and Total number of tickets is the total number of work orders; Based on the extracted interface metadata and business entities, a knowledge graph is constructed, which includes nodes and edges; The node types of a knowledge graph include interfaces, parameters, business objects, and user roles; Edges represent relationships, and relationship types include calling and containing. The natural language query data and interface metadata are converted into vector form, and the cosine similarity between them is calculated to describe their degree of similarity for semantic search. The mathematical expression is as follows:

[0009] in, For natural language query data vectors, This is an interface metadata vector; The interface metadata is standardized and preprocessed, and then converted into hash values ​​using a hash algorithm. The hash values ​​of subsequent interface metadata are compared with the original interface metadata hash values. If different hash values ​​are found, the interface metadata needs to be updated. The update strategy is as follows: for new interfaces, nodes and edges are automatically added; for modified interfaces, historical nodes are marked using version control; for deleted interfaces, the interface is deleted using a soft deletion method, and archived data is retained.

[0010] Step S2: The steps for constructing dialogue understanding and task planning include semantic parsing of user input, identifying contextual user intent and generating structured intent representations, and generating multi-level task planning for dynamic planning. Semantic analysis is performed on the work order description text input by the user based on a pre-trained language model; the work order description text input by the user is converted into vector form, and cosine similarity is calculated; Identifying contextual user intent includes intent classification and entity extraction; Intent classification employs an attention-based neural network model, mathematically expressed as:

[0011] Here, input is the input, and intent is the corresponding intent. An embedded representation of the input text describing the work order. For the encoding layer of the pre-trained language model, W and b are the trainable parameters, respectively; By combining dialogue history information, a gating mechanism is used to integrate the current user's work order description text input with historical context, specifically including: Use a gated loop unit to maintain the dialogue state: ; in, This indicates the current state of the conversation. The embedded representation of the current input; Entity extraction includes using sequence labeling methods to identify key business entities and linking the identified entities with nodes in the knowledge graph; The structured intent representation is encapsulated in JSON format, including intent type, confidence level, entity list, and relationships; Generating multi-level task planning involves using multi-hop queries to retrieve interfaces and operation sequences related to the current intent in the power marketing interface knowledge graph, and generating an abstract task plan, which includes the target state, constraints, and available operations. Define the initial state Target state and intermediate state set Each state transition corresponds to one available operation; The abstract plan is broken down into specific execution steps, and an operation dependency graph G(V,E) is constructed, where the vertex set V represents the specific operation and the edge set E represents the sequential constraints between operations. For each operation, the system extracts complete call parameters, authentication information, and expected response format from the knowledge graph; During the automated execution process, monitor the interface response time, success rate, and error types. When anomalies occur during monitoring, the causes are analyzed, equivalent alternatives are retrieved from the knowledge graph, and dynamic programming is performed, specifically including: The optimal strategy is calculated using an iterative algorithm. The mathematical expression is:

[0012] Where S is the state space of the task execution environment, A is the set of available operations, P(s'|s,a) is the state transition function, and R(s,a) is the reward function. This is the discount factor.

[0013] Step S3: Construct the automatically executed steps, based on the knowledge graph mapping relationship, identify the parameter information input by the user, generate the parameters used before the interface call, construct a safe and reliable interface call, and verify and provide feedback on the execution results; The knowledge graph-based mapping relationship includes matching the entities and attributes input by the user with the target interface parameters; The parameters used before generating the API call include generating default values ​​based on the context user intent in step S2. If a specific format parameter is required, a format converter is used to convert the default value into the standardized format required by the API. Building secure and reliable API calls involves constructing API requests that conform to the API specifications stored in the knowledge graph; An error handling mechanism is introduced to check the validity of parameters before sending a request, monitor the response status in real time during the API call, and automatically retry if a temporary error occurs. Execution result verification includes comparing the actual results returned by the API with the expected result patterns stored in the knowledge graph. The comparison includes field integrity checks, data format verification, and business logic verification. The execution result feedback transforms the structured API response into a user-friendly expression, and the feedback collection sources obtain user evaluations of the execution effect for updating the knowledge graph.

[0014] Secondly, the technical solution of the present invention also provides a power marketing auxiliary work order processing system based on a large language model, including an interface knowledge graph construction module, a dialogue understanding and task planning construction module, and an automatic execution construction module; The interface knowledge graph construction module collects and parses multi-source interface information, constructs a unified interface metadata model, extracts business entities and their interface relationships, and constructs a knowledge graph. The dialogue understanding and task planning construction module performs semantic parsing on user input, identifies contextual user intent and generates a structured intent representation, and generates multi-level task planning for dynamic planning. The automatic execution construction module, based on the knowledge graph mapping relationship, identifies the parameter information input by the user, generates the parameters used before the interface call, constructs a safe and reliable interface call, and verifies and feeds back the execution result.

[0015] The beneficial effects of this invention are that it provides a power marketing auxiliary work order processing method and system based on a large language model. It constructs a knowledge graph through a unified interface metadata model, associates unified structured representation with semantics based on user semantic analysis, and constructs a safe and reliable interface call by generating parameters used before interface calls and optimizes the knowledge graph by providing feedback on execution results.

[0016] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1This is a flowchart of a power marketing auxiliary work order processing method based on a large language model provided by the present invention.

[0019] Figure 2 This is a schematic diagram of a power marketing auxiliary work order processing system based on a large language model provided by the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention.

[0022] Example 1: like Figure 1 As shown, this embodiment of the invention provides a method for processing auxiliary work orders in electricity marketing based on a large language model, including the following steps: Step S1: The steps to build an interface knowledge graph include collecting and parsing multi-source interface information, building a unified interface metadata model, extracting business entities and their interface relationships, and building the knowledge graph. Multi-source interface information includes front-end components, API documentation, and network request logs from the Electricity Marketing 2.0 system; Front-end components include the HTML DOM tree, forms, and field definitions; For API documentation, if it is in Swagger or OpenAPI format, parse the JSON / YAML file directly; otherwise, use Natural Language Processing (NLP) tools to extract key information.

[0023] Parsing multi-source interface information includes extracting relevant attributes and data binding relationships from front-end components, API documentation, and interface elements using parsing libraries and tools; UI element related attributes include operation attributes, such as ID, class, and event listeners.

[0024] This embodiment provides a parsing rule table, as shown in Table 1: Table 1

[0025] Use log analysis tools to identify API call patterns in network request logs and group these patterns accordingly. During the grouping process, parameter passing is analyzed. By calculating and comparing the parameter similarity between the request and response bodies, parameter dependencies are identified. The mathematical expression for parameter similarity is:

[0026] in, These are the parameter sets for comparing the request parameter set and the parameter set for the response body, respectively. To compare request parameters, For response parameters, For the Kronecker delta function, The number of parameters; Building a unified interface metadata model involves using parsing tools to extract interface metadata from API documentation and extracting network request interaction data from network logs, and then integrating these two types of data to construct the interface metadata model. Define the interface metadata model as JSON Schema format. The interface metadata types include interface endpoint, request method, parameter structure, authentication method, and return format. This embodiment provides an interface metadata model table, as shown in Table 2: Table 2

[0027] Business entities and their interface relationships are extracted to build a domain entity model. NLP tools are used to identify business entities from work order texts and API documents. The results are analyzed in conjunction with the parameter passing mechanism to establish corresponding relationships between entity types, interfaces, and parameters. Business entities include "users", "electricity meters", and "bills".

[0028] Identify task templates and exception handling patterns in historical work order data, group the work orders, and use the support of the task template to represent the frequency of the task in historical work orders. The mathematical expression is:

[0029] Where T is the task template, Number of tickets containing T is the number of work orders containing task template T, and Total number of tickets is the total number of work orders; The task template includes user bill inquiry; the exception handling mode includes authentication failure retry.

[0030] Based on the extracted interface metadata and business entities, a knowledge graph is constructed, which includes nodes and edges; The node types of a knowledge graph include Interface, Parameter, BusinessObject, and UserRole. Edges represent relationships, and relationship types include calling and containing. The natural language query data and interface metadata are converted into vector form, and the cosine similarity between them is calculated to describe their degree of similarity for semantic search. The mathematical expression is as follows:

[0031] in, For natural language query data vectors, This is an interface metadata vector; Users can input natural language, such as "How to obtain user information", and the system will return relevant interface nodes.

[0032] The interface metadata is standardized and preprocessed, and then converted into hash values ​​using a hash algorithm. The hash values ​​of subsequent interface metadata are compared with the original interface metadata hash values. If different hash values ​​are found, the interface metadata needs to be updated. The update strategy is as follows: for new interfaces, nodes and edges are automatically added; for modified interfaces, historical nodes are marked using version control; for deleted interfaces, the interface is deleted using a soft deletion method, and archived data is retained.

[0033] Step S2: The steps for constructing dialogue understanding and task planning include semantic parsing of user input, identifying contextual user intent and generating structured intent representations, and generating multi-level task planning for dynamic planning. Semantic analysis is performed on the work order description text input by the user based on a pre-trained language model; the work order description text input by the user is converted into vector form, and cosine similarity is calculated; The work order description text entered by the user, such as "The user reported that the electricity balance inquiry failed, please assist in processing".

[0034] Identifying contextual user intent includes intent classification and entity extraction; Intent classification employs an attention-based neural network model, mathematically expressed as:

[0035] Here, input is the input, and intent is the corresponding intent. An embedded representation of the input text describing the work order. For the encoding layer of the pre-trained language model, W and b are the trainable parameters, respectively; By combining dialogue history information, a gating mechanism is used to integrate the current user's work order description text input with historical context, specifically including: Use a gated loop unit to maintain the dialogue state: ; in, This indicates the current state of the conversation. The embedded representation of the current input; Entity extraction includes using sequence labeling methods to identify key business entities and linking the identified entities with nodes in the knowledge graph; The structured intent representation is encapsulated in JSON format, including intent type, confidence level, entity list, and relationships; Generating multi-level task planning involves using multi-hop queries to retrieve interfaces and operation sequences related to the current intent in the power marketing interface knowledge graph, and generating an abstract task plan, which includes the target state, constraints, and available operations. Define the initial state Target state and intermediate state set Each state transition corresponds to an available operation, and the preconditions and effects of the operation are defined through interface constraints in the knowledge graph; In this embodiment, descriptions of the elements corresponding to the abstract task plan are given, as shown in Table 3: Table 3

[0036] The abstract plan is broken down into specific execution steps, and an operation dependency graph G(V,E) is constructed, where the vertex set (V) represents the specific operation and the edge set (E) represents the sequential constraints between operations. For each operation, the system extracts complete call parameters, authentication information, and expected response format from the knowledge graph; During the automated execution process, monitor the interface response time, success rate, and error types. When anomalies occur during monitoring, the causes are analyzed, equivalent alternatives are retrieved from the knowledge graph, and dynamic programming is performed, specifically including: The optimal strategy is calculated using an iterative algorithm. The mathematical expression is:

[0037] Where S is the state space of the task execution environment, A is the set of available operations, P(s'|s,a) is the state transition function, and R(s,a) is the reward function. This is the discount factor.

[0038] Execution order optimization is achieved through an optimal strategy, which identifies dependencies in the task graph, learns an operation duration model based on historical execution data, predicts the execution time of each step, and dynamically adjusts the scheduling strategy.

[0039] This embodiment presents the anomaly types and handling strategies in dynamic programming adjustments, as shown in Table 4: Table 4

[0040] Step S3: Construct the automatically executed steps, based on the knowledge graph mapping relationship, identify the parameter information input by the user, generate the parameters used before the interface call, construct a safe and reliable interface call, and verify and provide feedback on the execution results; The knowledge graph-based mapping relationship includes matching the entities and attributes input by the user with the target interface parameters; When a user mentions "last month's electricity bill", the system can automatically map it to the "bill_date" parameter in the bill query interface and calculate the specific date range "2024-04-01 to 2024-04-30".

[0041] The parameters used before generating the API call include generating default values ​​based on the context user intent in step S2. If a specific format parameter is required, a format converter is used to convert the default value into the standardized format required by the API. Building secure and reliable API calls involves constructing API requests that conform to the API specifications stored in the knowledge graph; Interface specifications include endpoint addresses, request methods, authentication methods, and parameter structures.

[0042] An error handling mechanism is introduced to check the validity of parameters before sending a request, monitor the response status in real time during the API call, and automatically retry if a temporary error occurs. Temporary errors include network timeouts and service unavailability; The retry strategy is to wait 1 second before the first retry, 2 seconds before the second, 4 seconds before the third, and so on.

[0043] Execution result verification includes comparing the actual results returned by the API with the expected result patterns stored in the knowledge graph. The comparison includes field integrity checks, data format verification, and business logic verification. In electricity bill inquiry scenarios, the returned data is verified to include necessary fields such as amount and date. The rule engine checks the reasonableness of the amount value, such as whether it is within the normal fluctuation range of historical consumption. The execution result feedback transforms the structured API response into a user-friendly expression, and the feedback collection sources obtain user evaluations of the execution effect for updating the knowledge graph.

[0044] Example 2: like Figure 2 As shown, this embodiment also provides a power marketing auxiliary work order processing system based on a large language model, including an interface knowledge graph construction module 1, a dialogue understanding and task planning construction module 2, and an automatic execution construction module 3; Interface knowledge graph construction module 1 collects and parses multi-source interface information, constructs a unified interface metadata model, extracts business entities and their interface relationships, and constructs a knowledge graph; Dialogue understanding and task planning module 2 performs semantic parsing of user input, identifies contextual user intent and generates structured intent representation, and generates multi-level task planning for dynamic planning; The automatic execution module 3, based on the knowledge graph mapping relationship, identifies the parameter information input by the user, generates the parameters used before the interface call, builds a safe and reliable interface call, and verifies and provides feedback on the execution results.

[0045] The Electricity Marketing 2.0 system in this technical solution refers to a new generation of energy internet marketing service system created by power grid companies to adapt to the reform of the electricity market and the construction of a new power system.

[0046] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the systems disclosed in the embodiments; relevant details can be found in the method section.

[0047] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0048] In the embodiments provided by this invention, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0049] The units described as separate components may or may not be physically separate. 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0050] In addition, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit.

[0051] Similarly, in the various embodiments of the present invention, each processing unit can be integrated into a functional module, or each processing unit can exist physically, or two or more processing units can be integrated into a functional module.

[0052] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0053] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0054] The above-disclosed embodiments are merely preferred embodiments of the present invention, but the present invention is not limited thereto. Any non-creative variations that can be conceived by those skilled in the art, as well as any improvements and modifications made without departing from the principles of the present invention, should fall within the protection scope of the present invention.

Claims

1. A method for processing auxiliary work orders in electricity marketing based on a large language model, characterized in that, Includes the following steps: Step S1: The steps to build an interface knowledge graph include collecting and parsing multi-source interface information, building a unified interface metadata model, extracting business entities and their interface relationships, and building the knowledge graph. Step S2: The steps for constructing dialogue understanding and task planning include semantic parsing of user input, identifying contextual user intent and generating structured intent representations, and generating multi-level task planning for dynamic planning. Step S3: Construct automatically executed steps, based on knowledge graph mapping relationships, identify user input parameter information, generate parameters used before API calls, construct secure and reliable API calls, and verify and provide execution result feedback.

2. The method for processing auxiliary work orders in electricity marketing based on a large language model according to claim 1, characterized in that, The multi-source interface information includes front-end components, API documentation, and network request logs in the Electricity Marketing 2.0 system; Parsing multi-source interface information includes extracting relevant attributes and data binding relationships from front-end components, API documentation, and interface elements using parsing libraries and tools; Use log analysis tools to identify API call patterns in network request logs and group these patterns accordingly. During the grouping process, parameter passing is analyzed. By calculating and comparing the parameter similarity between the request and response bodies, parameter dependencies are identified. The mathematical expression for parameter similarity is: in, These are the parameter sets for comparing the request parameter set and the parameter set for the response body, respectively. To compare request parameters, For response parameters, For the Kronecker delta function, The number of parameters.

3. The method for processing auxiliary work orders in electricity marketing based on a large language model according to claim 1, characterized in that, Building a unified interface metadata model involves using parsing tools to extract interface metadata from API documentation and extracting network request interaction data from network logs, and then integrating these two types of data to construct the interface metadata model. Define the interface metadata model as a JSON Schema format. The interface metadata types include interface endpoints, request methods, parameter structures, authentication methods, and return formats.

4. The method for processing auxiliary work orders in electricity marketing based on a large language model according to claim 1, characterized in that, Business entities and their interface relationships are extracted to build a domain entity model. NLP tools are used to identify business entities from work order texts and API documents. The results are analyzed in conjunction with the parameter passing mechanism to establish corresponding relationships between entity types, interfaces, and parameters. Identify task templates and exception handling patterns in historical work order data, group the work orders, and use the support of the task template to represent the frequency of the task in historical work orders. The mathematical expression is: Where T is the task template, Number of tickets containing T is the number of work orders containing task template T, and Total number of tickets is the total number of work orders; A knowledge graph is constructed based on the extracted interface metadata and business entities. The knowledge graph includes nodes and edges. The types of knowledge graph nodes include interfaces, parameters, business objects, and user roles; Edges represent relationships, and the types of relationships include calling and containing. The natural language query data and interface metadata are converted into vector form, and the cosine similarity between them is calculated to describe their degree of similarity. The mathematical expression is as follows: in, For natural language query data vectors, This is an interface metadata vector; The interface metadata is standardized and preprocessed, and then converted into hash values ​​using a hash algorithm. The hash values ​​of subsequent interface metadata are compared with the original interface metadata hash values. If different hash values ​​are found, the interface metadata needs to be updated. The update strategy is as follows: for new interfaces, nodes and edges are automatically added; for modified interfaces, historical nodes are marked using version control; for deleted interfaces, the interface is deleted using a soft deletion method, and archived data is retained.

5. A method for processing auxiliary work orders in electricity marketing based on a large language model, as described in claim 1 or 4, characterized in that, Semantic parsing of user input includes: performing semantic analysis on the work order description text input by the user based on a pre-trained language model, converting the work order description text input by the user into vector form, and calculating cosine similarity; Identifying contextual user intent includes intent classification and entity extraction; Intent classification employs an attention-based neural network model, mathematically expressed as: Here, input is the input, and intent is the corresponding intent. An embedded representation of the input text describing the work order. For the encoding layer of the pre-trained language model, W and b are the trainable parameters, respectively; By combining dialogue history information, a gating mechanism is used to integrate the current user's work order description text input with historical context, specifically including: Use a gated loop unit to maintain the dialogue state: ; in, This indicates the current state of the conversation. The embedded representation of the current input; Entity extraction includes using sequence labeling methods to identify key business entities and linking the identified entities with nodes in the knowledge graph; The structured intent representation is encapsulated in JSON format, including intent type, confidence level, entity list, and relationships.

6. The method for processing auxiliary work orders in electricity marketing based on a large language model according to claim 1, characterized in that, Generating multi-level task planning involves using multi-hop queries to retrieve interfaces and operation sequences related to the current intent in the power marketing interface knowledge graph, and generating an abstract task plan, which includes the target state, constraints, and available operations. Define the initial state Target state and intermediate state set Each state transition corresponds to one available operation; The abstract plan is broken down into specific execution steps, and an operation dependency graph G(V,E) is constructed, where the vertex set (V) represents the specific operation and the edge set (E) represents the sequential constraints between operations. For each operation, the system extracts complete call parameters, authentication information, and expected response format from the knowledge graph; During the automated execution process, monitor the interface response time, success rate, and error type; When anomalies occur during monitoring, the causes are analyzed, equivalent alternatives are retrieved from the knowledge graph, and dynamic programming is performed, specifically including: The optimal strategy is calculated using an iterative algorithm. The mathematical expression is: Where S is the state space of the task execution environment, A is the set of available operations, P(s'|s,a) is the state transition function, and R(s,a) is the reward function. This is the discount factor.

7. The method for processing auxiliary work orders in electricity marketing based on a large language model according to claim 1, characterized in that, The knowledge graph-based mapping relationship includes matching the entities and attributes input by the user with the target interface parameters; The parameters used before generating the API call include generating default values ​​based on the context user intent in step S2. If a specific format parameter is required, a format converter is used to convert the default value into the standardized format required by the API. Building secure and reliable API calls involves constructing API requests that conform to the API specifications stored in the knowledge graph; An error handling mechanism is introduced to perform parameter validity checks before sending requests and to monitor the response status in real time during the API call process. If a temporary error occurs, an automatic retry is implemented.

8. The method for processing auxiliary work orders in electricity marketing based on a large language model according to claim 1, characterized in that, Execution result verification includes comparing the actual results returned by the API with the expected result patterns stored in the knowledge graph. The comparison includes field integrity checks, data format verification, and business logic verification. The execution result feedback transforms the structured API response into a user-friendly representation. Feedback sources include user evaluations of the execution performance for updating the knowledge graph.

9. A power marketing auxiliary work order processing system based on a large language model, characterized in that, This includes modules for building an interface knowledge graph, building a dialogue understanding and task planning system, and building an automated execution system. The interface knowledge graph construction module collects and parses multi-source interface information, constructs a unified interface metadata model, extracts business entities and their interface relationships, and constructs a knowledge graph. The dialogue understanding and task planning construction module performs semantic parsing on user input, identifies contextual user intent and generates a structured intent representation, and generates multi-level task planning for dynamic planning. The automatic execution construction module, based on the knowledge graph mapping relationship, identifies the parameter information input by the user, generates the parameters used before the interface call, constructs a safe and reliable interface call, and verifies and feeds back the execution result.

10. A power marketing auxiliary work order processing system based on a large language model according to claim 9, characterized in that, The multi-source interface information includes front-end components, API documentation, and network request logs in the Electricity Marketing 2.0 system; Parsing multi-source interface information includes extracting relevant attributes and data binding relationships from front-end components, API documentation, and interface elements using parsing libraries and tools; Use log analysis tools to identify API call patterns in network request logs and group these patterns accordingly. During the grouping process, parameter passing is analyzed. By calculating and comparing the parameter similarity between the request and response bodies, parameter dependencies are identified. The mathematical expression for parameter similarity is: in, These are the parameter sets for comparing the request parameter set and the parameter set for the response body, respectively. To compare request parameters, For response parameters, For the Kronecker delta function, The number of parameters.

Images