An agent decision rule extraction method based on logical reasoning

By using a logical reasoning-based approach, the discrete and continuous features of intelligent agent decision-making behavior data are vectorized, and deep learning models are used to filter and predict rule antecedents and consequents. This solves the black-box problem of the intelligent agent decision-making process, achieves white-box interpretation and high-accuracy reproduction, and is applicable to autonomous driving, intelligent manufacturing and smart energy management.

CN122334475APending Publication Date: 2026-07-03709TH RESEARCH INSTITUTE CHINA STATE SHIPBUILDING CORP LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
709TH RESEARCH INSTITUTE CHINA STATE SHIPBUILDING CORP LTD
Filing Date
2026-03-30
Publication Date
2026-07-03

AI Technical Summary

Technical Problem

Existing technologies struggle to efficiently extract the objective physical conditions that trigger decisions and the corresponding action instructions from the complex decision-making behavior data of intelligent agents. This makes it difficult to intuitively understand and trace the decision-making process of intelligent agents, affecting the security and interpretability of the system.

Method used

A logic-based reasoning approach is adopted, which uses a rule antecedent filter and a rule consequent predictor to vectorize the discrete and continuous features of the agent's decision-making behavior data. Then, a deep learning model is used to filter out the target rule antecedent and predict the rule consequent, thereby achieving a white-box interpretation of the agent's decision-making process.

Benefits of technology

It enables white-box interpretation and high-accuracy reproduction of the decision-making process of intelligent agents, improves the security and interpretability of the system, and is applicable to complex intelligent agent application scenarios such as autonomous driving, intelligent manufacturing and smart energy management.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122334475A_ABST
    Figure CN122334475A_ABST
Patent Text Reader

Abstract

The application belongs to the technical field of data mining, and specifically discloses an agent decision rule extraction method based on logical reasoning, which comprises the following steps: obtaining agent decision behavior data; performing vectorization processing and splicing on discrete features and continuous features in the agent decision behavior data to obtain vectorized data; based on the vectorized data and a rule antecedent candidate set, one or more sub-antecedents are screened out from the rule antecedent candidate set by a rule antecedent filter to generate a target rule antecedent, and the target rule antecedent is composed of the one or more sub-antecedents; based on the target rule antecedent, a target rule consequent is predicted by a rule consequent predictor, and the target rule antecedent and the target rule consequent are taken as a rule extraction result of the agent decision behavior data. Through the application, the agent decision rule can be effectively extracted, and the white-box explanation and high-accuracy reproduction of the black-box decision process of the agent can be realized.
Need to check novelty before this filing date? Find Prior Art