摘要
飞机试飞试验点是基本的飞行试验任务,如何对其进行科学、有效的编排,形成合理的试飞计划,对整个试飞全生命周期的安全、效率、成本目标起到了至关重要的作用。其中,试验点之间的关联关系分析,尤其是前置关系的确定决定了试验点执行顺序,是关键的试飞计划编排因素。因此,文中提出一种基于图卷积神经网络的知识挖掘算法来解决试验点的前置关系预测需求。整个算法模型基于试验点结构化表征的知识图谱开展,随后设计了图知识要素提取、基于图卷积的深层试验点特征挖掘、试验点对逻辑关系回归等模块,挖掘出试验点对间的关联性,实现了较为精准的前置关系预测。在飞行试验中,对试验点数据进行测试,并对比多个经典模型,文中算法的准确率和稳定性具有明显的优势,验证了所提算法的有效性。
The arrangement of flight test points is fundamental in flight testing.How to scientifically and effectively arrange these test points and form a rational flight test plan plays a crucial role in ensuring safety,efficiency,and cost-effectiveness throughout the entire flight test lifecycle.Among them,the analysis of the correlation between test points,especially the determination of the precedence relationship,represents the execution order of the test points and is a key factor in the arrangement of the flight test plan.Therefore,a knowledge mining algorithm based on graph convolutional neural networks is proposed to meet the demand for predicting the precedence relationship of test points.The entire algorithm model is developed in a knowledge graph based on the structural representation of test points.Subsequently,modules such as graph knowledge element extraction,test points'deep feature mining based on graph convolution,and test point pairs'logical relationship regression are designed to explore the correlation relationship between test point pairs and achieve accurate precedence relationship prediction.In the flight test,the test point dataset was tested,and multiple classical models were compared.The proposed algorithm demonstrates significant advantages in terms of accuracy and stability.The effectiveness of the proposed algorithm is verified.
作者
刘鹏
邓晓政
LIU Peng;DENG Xiaozheng(Chinese Flight Test Establishment,Xi’an 710089,China)
出处
《现代电子技术》
北大核心
2025年第17期160-166,共7页
Modern Electronics Technique
基金
中华人民共和国科学技术部国家重点研发计划(2022YFB4300205)。
关键词
飞行试验
试验点执行关系
图学习
图卷积神经网络
自编码器
知识图谱
flight test
test point execution order
graph learning
graph convolutional network
autoencoder
knowledge graph
作者简介
刘鹏(1981-),男,陕西渭南人,硕士研究生,高级工程师,研究方向为数据挖掘、机器学习;通讯作者:邓晓政(1982-),男,陕西渭南人,博士研究生,高级工程师,研究方向为计算智能、机器学习、多目标优化。
