Page 105 - 《软件学报》2026年第2期

软件学报 ISSN 1000-9825, CODEN RUXUEW E-mail: jos@iscas.ac.cn
2026,37(2):584−600 [doi: 10.13328/j.cnki.jos.007451] [CSTR: 32375.14.jos.007451] http://www.jos.org.cn
©中国科学院软件研究所版权所有. Tel: +86-10-62562563

*
基于领域知识图谱的框架间 AI 源码自动迁移

丁 嵘 1 , 刘屹洲 2 , 王雨倩 2 , 李一錡 1

1
(北京航空航天大学 人工智能研究院, 北京 100191)
2
(北京航空航天大学 计算机学院, 北京 100191)
通信作者: 丁嵘, E-mail: dingr@buaa.edu.cn

摘 要: 作为人工智能的基础设施, 深度学习框架已经成为人工智能实现跨越发展的重要突破口. 但是由于缺乏统
一标准, 不同框架的兼容水平较差. 忠实模型转换通过将源模型迁移为另一种目标框架下的等价模型, 来增强框架
间的互操作性. 然而, 深度学习框架数量较多且相互间差异较大, 并且自主框架的需求逐渐增多, 互相转换成本较
高. 因此, 提出基于领域知识图谱的框架间 AI 源码自动迁移方法. 该方法基于领域知识图谱和抽象语法树来系统
地处理迁移挑战, 首先将源代码转换为特定的抽象语法树, 提取通用依赖信息和特定算子信息, 然后再利用存储在
领域知识图谱中的框架间算子及参数映射关系来迁移到目标框架下, 形成目标框架下的目标模型代码, 大大降低
了工程复杂度. 对比同类型的代码迁移工具, 所提方法可以在国内外流行深度学习框架如 PyTorch、PaddlePaddle
和 MindSpore 之间进行互相迁移, 达到了较好的成熟度和质量, 部分成果已经开源到百度官方迁移工具 PaConvert 中.
关键词: 深度学习框架; 代码迁移; 知识图谱; 抽象语法树
中图法分类号: TP311

中文引用格式: 丁嵘, 刘屹洲, 王雨倩, 李一錡. 基于领域知识图谱的框架间AI源码自动迁移. 软件学报, 2026, 37(2): 584–600. http://
www.jos.org.cn/1000-9825/7451.htm
英文引用格式: Ding R, Liu YZ, Wang YQ, Li YQ. Automatic Migration of AI Source Code Between Frameworks Based on Domain
Knowledge Graph. Ruan Jian Xue Bao/Journal of Software, 2026, 37(2): 584–600 (in Chinese). http://www.jos.org.cn/1000-9825/7451.
htm

Automatic Migration of AI Source Code Between Frameworks Based on Domain Knowledge
Graph
2
2
1
DING Rong , LIU Yi-Zhou , WANG Yu-Qian , LI Yi-Qi 1
1
(Institute of Artificial Intelligence, Beihang University, Beijing 100191, China)
2
(School of Computer Science and Engineering, Beihang University, Beijing 100191, China)
Abstract: As the foundation of AI, deep learning frameworks play a vital role in driving the rapid progress of AI technologies. However,
due to the lack of unified standards, compatibility across different frameworks remains limited. Faithful model transformation enhances
interoperability by converting a source model into an equivalent model in the target framework. However, the large number and diversity
of deep learning frameworks, combined with the increasing demand for custom frameworks, lead to high conversion costs. To address this
issue, this study proposes an automatic AI source code migration method between frameworks based on a domain knowledge graph. The
method integrates domain knowledge graphs and abstract syntax trees to systematically manage migration challenges. First, the source code
is transformed into a framework-specific abstract syntax tree, from which general dependency information and operator-specific details are
extracted. By applying the operator and parameter mappings stored in the domain knowledge graph, the code is migrated to the target
framework, generating equivalent target model code while significantly reducing engineering complexity. Compared with existing code
migration tools, the proposed method supports mutual migration among widely used deep learning frameworks, such as PyTorch,

* 基金项目: 科技创新 2030—“新一代人工智能”重大项目 (2021ZD0110600); 百度合作项目 (Z231100010323007); 华为合作项目
(22081500805582B)
收稿时间: 2024-06-06; 修改时间: 2024-08-07, 2024-10-19, 2025-03-25; 采用时间: 2025-04-23; jos 在线出版时间: 2025-09-24
CNKI 网络首发时间: 2025-09-25