Distribution network dispatching instruction ticket generation method based on business-oriented large model

Abstract

The application discloses a kind of based on service guide big model's distribution network dispatching instruction ticket generation method, comprising the following steps: multi-source business data acquisition and standardization analysis;Build power maintenance service special prompt word engineering and generate special prompt word;Load and initialize the pre-training special big model of electric power industry, inject prompt word and establish business semantic alignment mechanism;Rely on prompt word guide and big model reasoning complete instruction core element intelligent matching;Intelligently generate dispatching instruction ticket and pass through prompt word compliance verification, business logic double check, check qualified and output standard instruction ticket.The application builds the exclusive semantic bridge of electric power business and big model, effectively suppresses model hallucination, accurately matches business core element, solves the problems of poor business adaptability, semantic deviation and insufficient intelligence in traditional technology, improves the compliance and accuracy of instruction ticket generation, with low business iteration adaptation cost and strong expansibility.

Description

Technical Field

[0001] This invention relates to the field of distribution network scheduling automation technology, and more specifically to a method for generating distribution network scheduling instruction tickets based on a business-oriented large model. Background Technology

[0002] Currently, the generation of distribution network dispatching instruction tickets mainly relies on the following three types of technologies: 1. Rule Engine and Template Matching Technology: By pre-setting a "five-prevention" logic library and an operation ticket template library, and based on the parsing of key fields in the maintenance application form, a fixed template is matched to generate an instruction ticket. This method relies on hard-coded rules, which has poor adaptability to complex and non-standardized business scenarios and cannot handle ambiguous semantics and changing business requirements.

[0003] 2. Traditional machine learning / simple neural network techniques: These techniques train shallow models based on historical instruction ticket samples, extracting features such as device and operation type to generate instruction sequences. However, these models have weak generalization ability, are insufficiently adaptable to new businesses and devices, and are prone to "model illusion" problems such as instruction logic errors and non-standard terminology.

[0004] 3. Direct Application of General-Purpose Large Models: This method directly calls open-source / commercial general-purpose large models and generates content by inputting simple text commands. However, these general-purpose large models lack professional knowledge of power distribution network dispatching, and the generated content does not conform to business specifications such as the "Power Safety Work Regulations" and the "Distribution Network Dispatch Operation and Management Regulations," resulting in low accuracy and poor practicality. Therefore, the above three methods have the following problems: 1. Insufficient business adaptability: The rule engine is rigid and the general model lacks professional guidance, making it impossible to deeply integrate with the exclusive business logic such as distribution network maintenance, changes, and mode arrangements.

[0005] 2. Semantic comprehension bias: Lacking a semantic bridge in the field of power industry, the model is prone to misunderstanding business terms and operational logic, resulting in the generation of incorrect instructions.

[0006] 3. Lack of intelligent matching: Unable to automatically and accurately match instruction type, target, and assigned work group based on route information and service type.

[0007] 4. Severe model illusion: Without professional prompts, the model is prone to fabricating non-existent operation steps and equipment names, resulting in low compliance and security.

[0008] 5. Poor scalability: Changes to business rules require recoding and development, resulting in long adaptation cycles, high costs, and an inability to quickly respond to the iterative needs of the power grid business. Summary of the Invention

[0009] To address the shortcomings of existing technologies, this invention aims to provide a method for generating distribution network dispatching instruction tickets based on a business-oriented large-scale model and a prompt word engineering approach, solving the following core problems: 1. Constructing a dedicated prompt word engineering approach for power maintenance business, serving as a "semantic bridge" between the business and the large-scale model to eliminate semantic bias and suppress model illusions. 2. Achieving intelligent selection of instruction types, precise filtering of recipients, and automatic matching of jurisdictional work teams based on the large-scale model and prompt word engineering approach. 3. Deeply integrating multi-source business data (maintenance application forms, mode arrangements, change orders, etc.) to generate dispatching instruction tickets that fully comply with power business specifications. 4. Enhancing the system's adaptability to new businesses and scenarios, reducing the development cost of rule changes, and shortening the adaptation cycle.

[0010] To achieve the above objectives, the present invention provides the following technical solution: a method for generating distribution network scheduling instruction tickets based on a business-oriented large model, comprising the following steps: Step 1: Multi-source business data collection and standardized parsing: Collect core business data for distribution network dispatching, including maintenance application forms, mode arrangement forms, historical dispatching instruction tickets, equipment change orders, power grid topology data, and line jurisdiction information; perform structured parsing on unstructured text data to extract core elements: equipment elements, business elements, and management elements; construct standardized data vectors.

[0011] in, For business datasets, For a single business data item, For the first The first data item One standardized feature; Step 2: Construct a dedicated prompt word project for power maintenance operations, and then use the dedicated prompt word project for power maintenance operations to generate a dedicated prompt word P; Step 3: Loading and initializing the business-oriented large model. A dedicated large model pre-trained with power industry data is used, and model weights and industry knowledge base are loaded. The dedicated prompt word P generated in Step 2 is used as a system-level prefix prompt word and injected into the model input layer to establish a business semantic alignment mechanism to ensure that all model outputs are constrained by the professional prompt word. Step four: Based on prompt words and large-scale model reasoning, achieve intelligent matching of core elements; Step 5: Intelligent generation and compliance verification of dispatch instruction tickets. First, the model generates a complete dispatch instruction ticket text based on the matched instruction type, the target of the instruction, the jurisdiction and the work group, combined with the prompt word constraints. Then, a dual verification mechanism is constructed: prompt word compliance verification: verifying whether the output conforms to all constraint rules in P; business logic verification: verifying the security, logic and compliance of the operation steps based on the "five-prevention" rule base. If the verification passes, a standard instruction ticket is output; if it fails, steps 4 and 5 are retried until the verification is successful.

[0012] As a further improvement to this invention, the specific steps for constructing the dedicated prompt word project for power maintenance business in step two are as follows: Based on business logic analysis and sample data training, a hierarchical prompt word framework is designed as a semantic bridge between business and model, containing a 4-layer core structure: Basic constraint layer: Defines power industry standards, safety regulations, and terminology standards, and forces the model to output compliant content; Business logic layer: Embeds distribution network dispatching operation rules, maintenance procedures, and "five-prevention" logic, clearly defining business boundaries; Element guidance layer: Specifies the core elements of the instruction to be extracted; Format specification layer: Defines the standard format, field requirements, and writing style of instruction tickets to ensure standardized output; obtains the following prompt word generation formula:

[0013] in: The final generated dedicated prompt word, : Weight vector of basic constraint layer , For the first The weight of each basic specification, : Business logic layer weight vector , For the first The weight of each business logic item : Element guidance layer weight vector , For the first The weight of each instruction element, : Format specification layer weight vector , For the first The weights required by the item format. The prompt word layer weighted fusion operator enables the organic combination of multi-layer rules.

[0014] As a further improvement of the present invention, the specific method for achieving intelligent matching of core elements based on prompt words and large model reasoning in step four is as follows: 1. Intelligent selection of instruction type The model is based on parsed business data. With prompt words Calculate the matching degree for each instruction type:

[0015] in, For the first The degree of matching of instruction types, For feature splicing operators, For the first Standard feature library for various instruction types; selection The instruction type corresponding to the maximum value; 2. Precise selection of recipients Based on business scope, voltage level, and operation permissions, construct a formula for filtering the objects to be distributed:

[0016] in, Distribute a set of objects to the target. For object Access permissions, The scope of work corresponding to the business data. For object level, The corresponding level requirements for the business; 3. Automatic matching of work teams within the jurisdiction Based on the route name, topological region, and location information, calculate the matching index between the route and the work team:

[0017] in, For the line With the work team The matching index; Weighting coefficients ( ); For regional matching degree, For line affiliation matching degree, For permission matching degree; select The team with the highest value.

[0018] Beneficial effects: By standardizing and parsing multi-source business data, building a dedicated prompt word project, aligning the semantics of a large model, and implementing dual compliance verification, this invention fundamentally addresses various shortcomings of existing technologies. First, the collection and structured processing of multi-source, multi-dimensional business data breaks through the limitations of traditional rule engines, which rely on single data sources and fixed application scenarios. It is compatible with various distribution network dispatching business scenarios, significantly improving business adaptability and solving the problem of traditional technologies being unable to adapt to complex, non-standardized businesses. Second, by building a power-specific business-oriented architecture and establishing a precise semantic alignment mechanism between the model and power business, it compensates for the lack of professional semantics in general-purpose large models, completely eliminating semantic comprehension bias. Simultaneously, the entire solution embeds multi-layered business constraints and compliance verification logic, effectively suppressing the illusion of fabricated information in large models and ensuring the security and compliance of instruction tickets. Furthermore, this invention does not require hard-coding of fixed rules. Relying on the dynamic adaptation capabilities of the model and prompt words, it can quickly respond to power grid business iteration updates, significantly reducing adaptation costs and shortening iteration cycles, thus solving the pain point of poor scalability in traditional technologies.

[0019] The layered prompt word weighted fusion mechanism and multi-dimensional core element intelligent matching algorithm subsequently added to this invention further optimize the standardization and matching accuracy of instruction ticket generation, refine the business constraint boundaries, make the model output more in line with power industry standards and on-site practical needs, further improve the intelligence and refinement of instruction ticket generation, and assist the core solution in achieving optimal technical results. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to the given embodiments.

[0021] The distribution network scheduling instruction ticket generation method based on a business-oriented large model in this embodiment includes the following steps: Step 1: Multi-source business data collection and standardized parsing: Collect core business data for distribution network dispatching, including maintenance application forms, mode arrangement forms, historical dispatching instruction tickets, equipment change orders, power grid topology data, and line jurisdiction information; perform structured parsing on unstructured text data to extract core elements: equipment elements, business elements, and management elements; construct standardized data vectors.

[0022] in, For business datasets, For a single business data item, For the first The first data item One standardized feature; Step 2: Construct a dedicated prompt word project for power maintenance operations, and then use the dedicated prompt word project for power maintenance operations to generate a dedicated prompt word P; Step 3: Loading and initializing the business-oriented large model. A dedicated large model pre-trained with power industry data is used, and model weights and industry knowledge base are loaded. The dedicated prompt word P generated in Step 2 is used as a system-level prefix prompt word and injected into the model input layer to establish a business semantic alignment mechanism to ensure that all model outputs are constrained by the professional prompt word. Step four: Based on prompt words and large-scale model reasoning, achieve intelligent matching of core elements; Step 5: Intelligent generation and compliance verification of dispatch instruction tickets. First, the model generates a complete dispatch instruction ticket text based on the matched instruction type, the target of the instruction, the jurisdiction and the work group, combined with the prompt word constraints. Then, a dual verification mechanism is constructed: prompt word compliance verification: verifying whether the output conforms to all constraint rules in P; business logic verification: verifying the security, logic and compliance of the operation steps based on the "five-prevention" rule base. If the verification passes, a standard instruction ticket is output; if it fails, steps 4 and 5 are retried until the verification is successful.

[0023] In practical application, this solution first integrates core business data across all categories of distribution network dispatching, completing the standardized transformation of unstructured data and resolving the issues of traditional technologies having limited data dimensions and being unable to adapt to complex business processes. Relying on a pre-trained power model and dedicated prompt word constraints, it overcomes the shortcomings of general-purpose models lacking professional constraints and prone to generating false impressions, achieving accurate understanding of business semantics. Through intelligent element matching and dual compliance verification, it automatically generates instructions and conducts compliance screening, eliminating the need for manual editing based on fixed templates. This effectively addresses industry pain points such as poor adaptability of traditional rule engines, weak generalization ability of traditional machine learning models, and low accuracy of general-purpose models, achieving automated and compliant generation of dispatching instruction tickets.

[0024] Furthermore, the specific steps for constructing the dedicated prompt word project for power maintenance business in step two are as follows: Based on business logic analysis and sample data training, a hierarchical prompt word framework is designed as a semantic bridge between business and model, containing a 4-layer core structure: Basic constraint layer: Defines power industry standards, safety regulations, and terminology standards, and forces the model to output compliant content; Business logic layer: Embeds distribution network dispatching operation rules, maintenance procedures, and "five-prevention" logic, clearly defining business boundaries; Element guidance layer: Specifies the core elements of the instruction to be extracted; Format specification layer: Defines the standard format, field requirements, and writing style of instruction tickets to ensure standardized output; obtains the following prompt word generation formula:

[0025] in: The final generated dedicated prompt word, : Weight vector of basic constraint layer , For the first The weight of each basic specification, : Business logic layer weight vector , For the first The weight of each business logic item : Element guidance layer weight vector , For the first The weight of each instruction element, : Format specification layer weight vector , For the first The weights required by the item format. The prompt word layer weighted fusion operator enables the organic combination of multi-layer rules.

[0026] This layered weighted prompt word engineering approach, through multi-dimensional rule constraints and semantic alignment, specifically addresses the issues of missing semantics and misunderstandings in general large-scale power industry models. It provides standardized business basis for model inference, suppressing the illusion of models fabricating erroneous information from the source. Furthermore, the weights of each layer can be adjusted according to business rules without reconstructing the overall architecture, significantly improving the solution's scalability and adapting to the iterative needs of power grid business, providing core semantic support for the accurate implementation of core solutions.

[0027] Furthermore, in step four, the intelligent matching of core elements based on prompt words and large-scale model reasoning is achieved in the following specific ways: 1. Intelligent selection of instruction type The model is based on parsed business data. With prompt words Calculate the matching degree for each instruction type:

[0028] in, For the first The degree of matching of instruction types, For feature splicing operators, For the first Standard feature library for various instruction types; selection The instruction type corresponding to the maximum value; 2. Precise selection of recipients Based on business scope, voltage level, and operation permissions, construct a formula for filtering the objects to be distributed:

[0029] in, Distribute a set of objects to the target. For object Access permissions, The scope of work corresponding to the business data. For object level, The corresponding level requirements for the business; 3. Automatic matching of work teams within the jurisdiction Based on the route name, topological region, and location information, calculate the matching index between the route and the work team:

[0030] in, For the line With the work team The matching index; Weighting coefficients ( ); For regional matching degree, For line affiliation matching degree, For permission matching degree; select The team with the highest value.

[0031] This multi-dimensional intelligent matching algorithm overcomes the shortcomings of traditional technologies in automatically matching instruction types, recipients, and work teams. Relying on feature matching, permission verification, and location adaptation, it achieves precise and intelligent matching of core business elements, replacing the crude processing methods of traditional manual screening and template matching. This significantly improves the accuracy and adaptability of instruction ticket generation, further ensuring that the instruction tickets generated by the core solution closely match the actual field operation scenarios. Simultaneously, the weighting coefficients can be flexibly adjusted to adapt to the needs of different regions and types of power grid services, further enhancing the solution's versatility and scalability.

[0032] In summary, this invention establishes an intelligent generation solution for distribution network dispatching instruction tickets based on a business-oriented large model and a dedicated prompt word engineering. Through a complete process of multi-source business data standardization processing, hierarchical power prompt word engineering construction, large model semantic alignment, intelligent element matching, and dual compliance verification, it effectively solves many problems of existing distribution network instruction ticket generation technologies, such as poor business adaptability, semantic understanding bias, insufficient intelligence, serious model illusion, and weak scalability. It realizes automated, standardized, compliant, and accurate generation of dispatching instruction tickets, significantly improving the efficiency and safety of distribution network dispatching work, and can quickly adapt to the iterative needs of various power grid businesses.

[0033] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for generating distribution network dispatch instruction tickets based on a business-oriented large model, characterized in that: Includes the following steps: Step 1: Multi-source business data collection and standardized parsing: Collect core business data for distribution network dispatching, including maintenance application forms, mode arrangement forms, historical dispatching instruction tickets, equipment change orders, power grid topology data, and line jurisdiction information; perform structured parsing on unstructured text data to extract core elements: equipment elements, business elements, and management elements; construct standardized data vectors. in, For business datasets, For a single business data item, For the first The first data item One standardized feature; Step 2: Construct a dedicated prompt word project for power maintenance operations, and then use the dedicated prompt word project for power maintenance operations to generate a dedicated prompt word P; Step 3: Loading and initializing the business-oriented large model. A dedicated large model pre-trained with power industry data is used, and model weights and industry knowledge base are loaded. The dedicated prompt word P generated in Step 2 is used as a system-level prefix prompt word and injected into the model input layer to establish a business semantic alignment mechanism to ensure that all model outputs are constrained by the professional prompt word. Step four: Based on prompt words and large-scale model reasoning, achieve intelligent matching of core elements; Step 5: Intelligent generation and compliance verification of dispatch instruction tickets. First, the model generates a complete dispatch instruction ticket text based on the matched instruction type, the target of the instruction, the jurisdiction and the work group, combined with the prompt word constraints. Then, a dual verification mechanism is constructed: prompt word compliance verification: verifying whether the output conforms to all constraint rules in P; business logic verification: verifying the security, logic and compliance of the operation steps based on the "five-prevention" rule base. If the verification passes, a standard instruction ticket is output; if it fails, steps 4 and 5 are retried until the verification is successful.

2. The method for generating distribution network dispatch instruction tickets based on a business-oriented large model according to claim 1, characterized in that: Step two involves constructing a dedicated prompt word project for power maintenance operations. The specific steps are as follows: Based on business logic analysis and sample data training, a hierarchical prompt word framework is designed as a semantic bridge between the business and the model, comprising a four-layer core structure: Basic constraint layer: Defines power industry standards, safety regulations, and terminology standards, and forces the model to output compliant content; Business logic layer: Embeds distribution network dispatching operation rules, maintenance procedures, and "five-prevention" logic, clearly defining business boundaries; Element guidance layer: Specifies the core elements of the instruction to be extracted; Format specification layer: Defines the standard format, field requirements, and writing style of instruction tickets to ensure standardized output; obtains the following prompt word generation formula: in: The final generated dedicated prompt word, : Weight vector of basic constraint layer , For the first The weight of each basic specification, : Business logic layer weight vector , For the first The weight of each business logic item : Element guidance layer weight vector , For the first The weight of each instruction element, : Format specification layer weight vector , For the first The weights required by the item format. The prompt word layer weighted fusion operator enables the organic combination of multi-layer rules.

3. The method for generating distribution network dispatch instruction tickets based on a business-oriented large model according to claim 2, characterized in that: The specific method for achieving intelligent matching of core elements in step four, based on prompt word guidance and large model reasoning, is as follows: (1) Intelligent selection of instruction type The model is based on parsed business data. With prompt words Calculate the matching degree for each instruction type: in, For the first The degree of matching of instruction types, For feature splicing operators, For the first Standard feature library for various instruction types; selection The instruction type corresponding to the maximum value; (2) Precise selection of recipients Based on business scope, voltage level, and operation permissions, construct a formula for filtering the objects to be distributed: in, Distribute a set of objects to the target. For object Access permissions, The scope of work corresponding to the business data. For object level, The corresponding level requirements for the business; (3) Automatic matching of work groups within the jurisdiction Based on the route name, topological region, and location information, calculate the matching index between the route and the work team: in, For the line With the work team The matching index; Weighting coefficients ( ); For regional matching degree, For line affiliation matching degree, For permission matching degree; select The team with the highest value.

Images