CN-121788077-B - Double-track consistency execution instruction compiling method and system summarized by drawing rule

CN121788077BCN 121788077 BCN121788077 BCN 121788077BCN-121788077-B

Abstract

A compiling method and a system for double-track consistency execution instructions summarized by drawing examination rules relate to the field of artificial intelligence. The method comprises the steps of obtaining rule summaries and domain knowledge bases of business experts and carrying out structural analysis to obtain standardized rule objects, extracting applicable formats and triggering conditions, mapping intentions to a predefined examination mode set, instantiating an examination operator chain as an abstract execution sequence, traversing the operator chain to generate variable metadata sets corresponding to each user one by one, and generating two types of products in parallel, namely a rail A readable examination text used for driving a large model to develop and explain reasoning according to fixed logic in the operation period, and a rail B machine readable protocol used for carrying out deterministic comparison and postpositive strong constraint verification in the background. The method is mainly used for compiling the outline of the aesthetic rules into the double-track consistency execution instruction.

Inventors

YANG XINYI
Shao Jiangdong
HU DIANZHONG
WANG JING
LIANG HONGMING
SHI GUOQING
ZHONG ZHAOYAN
YU WENQI

Assignees

烟台海颐软件股份有限公司

Dates

Publication Date: 20260508
Application Date: 20260305

Claims (5)

1. The compiling method of the double-track consistency execution instruction summarized by the drawing rule is characterized by comprising the following steps of: S1, acquiring rule summaries and a domain knowledge base of service experts, and carrying out structural analysis on the original rule summaries to obtain standardized rule objects; S2, extracting applicable formats and triggering conditions from the rule objects and the corresponding domain knowledge base; S3, mapping the intention to a predefined examination mode set according to the intention of the rule object, and instantiating an examination operator chain as an abstract execution sequence; s4, traversing an operator chain, collecting variables involved in the operator chain, and generating variable metadata sets corresponding to each user one by one; S5, taking the operator chain and the variable metadata set as unique fact sources, and generating two types of products in parallel, namely a rail A readable audit text which is used for driving a large model to develop and explain reasoning according to fixed logic in the operation period; After S5, the method further comprises the following steps: s6, extracting and normalizing key elements related to the structure and constraint in the track A and the track B, and carrying out hash consistency verification; S7, checking whether the track A and the track B simultaneously and explicitly contain forbidden constraint declared by the rule in the compiling period, and simultaneously carrying out corresponding post strong verification after the track B obtains an operation result; S8, determining an input and output structure of the current rule in the compiling period, so that the compiling and the running keep the interface consistent; In S5, the readable audit text of the track A adopts a fixed Chinese step genre, and the machine-readable protocol of the track B adopts JSON serialization, including gating, operator chains, variable metadata sets, comparison strategies, post-verification and I/O conventions.
2. The method for compiling double-track consistency execution instructions outlined by the rules of examination of claim 1, wherein in S6, the specific step of hash consistency check comprises: Extracting structural elements which are only associated with execution semantics from the track A and the track B; Carrying out key sequencing, set de-reordering and blank folding on the file according to a standardization rule, unifying regular expression modes, and serializing the file into a standardization JSON; respectively calculating hash values of the track A and the track B; if the hash values are consistent, release is allowed, if the hash values are inconsistent, the caliber drift is judged, and the system blocks compiling release and outputs a difference list.
3. The method of claim 1, wherein in S8, the output uses a unified variable metadata set, and the variable metadata set includes a state, a extraction and computation field, an evidence anchor array, an error code list and a brief reason description.
4. The method according to claim 1, wherein in S1, the presetting of the domain knowledge base includes creating an audit operator base, registering domain entities and synonyms and registering external data sources, and cleaning, entity recognition and unit normalization of natural language summary text of the rule summary to form a standardized rule object.
5. A compiling system for implementing the compiling method of the double-track consistency execution instruction summarized by the examination rules according to any one of claims 1 to 4, wherein the compiling system comprises a structural analysis module, a pre-verification module, an operator chain generation module, a traversal module and a double-track consistency generation module; The structured analysis module is used for acquiring rule summaries and domain knowledge bases of business experts, and carrying out structured analysis on the original rule summaries to obtain standardized rule objects; the front verification module is used for extracting applicable formats and triggering conditions from the rule objects and the corresponding domain knowledge base; The operator chain generation module is used for mapping the intention to a predefined examination mode set according to the intention of the rule object, and instantiating an examination operator chain as an abstract execution sequence; the traversing module is used for traversing the operator chain, collecting the variables involved in the operator chain and generating variable metadata sets corresponding to each user one by one; The double-track consistency generation module is used for taking the operator chain and the variable metadata set as unique fact sources and generating two types of products in parallel, namely a track A readable audit text which is used for driving a large model to develop and explain and reason according to fixed logic in the operation period and a track B machine readable protocol which is used for carrying out deterministic comparison and post-positioned strong constraint verification in the background.

Description

Double-track consistency execution instruction compiling method and system summarized by drawing rule Technical Field The invention relates to the field of artificial intelligence, in particular to a compiling method and a system for double-track consistency execution instructions summarized by an aesthetic drawing rule. Background Along with the improvement of the capability of a Large Language Model (LLM) in terms of semantic understanding and logical reasoning, and the combination of multi-modal processing technologies such as images, OCRs, tables and the like, the automatic auditing of engineering documents and drawings is brought into a new technical path. In the field of power distribution network engineering, particularly in the field of power supply business expansion to the aesthetic drawing business of power receiving engineering, engineering practice has begun to explore intelligent aesthetic drawing platforms combined with rule bases by using multi-mode aesthetic drawing large models, and has obvious potential in terms of interpretability, degree of automation and efficiency. However, one core bottleneck in putting this technology into practice is that engineering rules are often presented in a highly generalized natural language form in actual work, e.g. "incoming cabinet CT transformation ratio should match transformer capacity", "transformer selection should meet energy efficiency standards". The experienced business expert can understand the rules and complement the implicit steps, the threshold values, the units and the forbidden constraints, but if the rule summary is directly input into a large language model as a prompt word, the problems of ambiguous caliber, unclear threshold values and units, difficult reproduction and the like of the auditing result are often caused by the lack of explicit execution steps and necessary field constraints. This "summary→executable" floor gap is a key bottleneck in engineering applications. To solve this problem, the prior art paths have three main types, but all have significant drawbacks: (1) Manually rewriting rules-relying on AI engineers or domain specialists to manually rewrite each rule summary into an exhaustive, structured prompt. The method is high in cost and long in period, and the logic consistency of rules written by different personnel is difficult to ensure, so that version maintenance is extremely difficult. (2) And (3) a pure large model scheme, namely directly relying on a large model to carry out end-to-end auditing. This approach, while flexible, is difficult to meet project audit strong, playable, auditable requirements due to the lack of verifiable deterministic constraints and explicit parameters, units, threshold specifications. (3) The business rule engine has advantages in deterministic execution, but has a semantic gap between a rule language and an expert natural language, still needs long-term intervention of professional IT personnel, and is difficult to be directly applied to complex scenes of drawing auditing due to the fact that the multi-mode information such as images, formats, tables and the like lacks of primary processing capability. Therefore, there is a need for a dual track consistent instruction compilation method and system that can automate, validate, and backtrack the overview of the rules of investigation. Disclosure of Invention The invention provides a double-track consistency execution instruction compiling method and a system thereof capable of automatically verifying and backtracking the outline of an aesthetic rule in order to solve the defects of low efficiency, lack of verification and non backtracking of the existing aesthetic rule compiling method. The invention relates to a compiling method of a double-track consistency execution instruction summarized by an aesthetic drawing rule, which comprises the following steps: S1, acquiring rule summaries and a domain knowledge base of service experts, and carrying out structural analysis on the original rule summaries to obtain standardized rule objects; S2, extracting applicable formats and triggering conditions from the rule objects and the corresponding domain knowledge base; S3, mapping the intention to a predefined examination mode set according to the intention of the rule object, and instantiating an examination operator chain as an abstract execution sequence; s4, traversing an operator chain, collecting variables involved in the operator chain, and generating variable metadata sets corresponding to each user one by one; And S5, taking the operator chain and the variable metadata set as unique fact sources, and generating two types of products in parallel, namely a track A readable audit text for driving a large model to develop and explain reasoning according to fixed logic in the operation period and a track B machine-readable protocol for carrying out deterministic comparison and postpositive strong constraint verification in the background. F