基于图检索增强生成的报废动力电池拆解序列规划
王航 , 鄢威 , 张绪美 , 朱硕 , 江志刚 , 朱泽睿
中国机械工程 ›› 2026, Vol. 37 ›› Issue (4) : 967 -976.
基于图检索增强生成的报废动力电池拆解序列规划
GraphRAG-based Disassembly Sequences Planning for End-of-life Power Batteries
针对报废动力电池拆解知识难以复用所导致的序列生成效率低下问题,融合知识图谱(KG)在结构化知识表征与大语言模型在语义推理方面的互补优势,提出了一种基于图检索增强生成的报废动力电池拆解序列规划方法。首先提出基于KG标签匹配的子图生成方法,通过Cypher查询语言形成待拆解特定型号电池的专属拆解序列子图;然后建立基于混合检索和重排序的拆解知识粗粒度检索机制,以实现对目标电池组件的精确定位;最后,构建基于层级约束关系多跳推理的拆解知识细粒度检索模式,通过提取与组件关联的拆解序列信息,利用大语言模型实现拆解序列的智能生成。实验结果显示,所提方法在5种主流动力电池拆解序列生成上实现93.9%的准确率,展现出优异的可行性和有效性。
To address the challenges of low-efficiency disassembly sequence generation caused by the lack of knowledge reusability of end-of-life power batteries, a disassembly sequence planning method was proposed based on GraphRAG, by integrating the complementary strengths of KGs in structured knowledge representation and LLMs in semantic reasoning. First, a KG label-matching-based subgraph generation method was proposed, utilizing the Cypher query language to form custom disassembly sequence subgraphs for specific battery models. Second, a coarse-grained disassembly knowledge retrieval mechanism employing hybrid retrieval and re-ranking was established to locate target battery components precisely. Finally, a fine-grained retrieval model of disassembly knowledge was constructed based on multi-hop reasoning of hierarchical constraint relations. By extracting the disassembly sequence information associated with the components, the intelligent generation of disassembly sequences was achieved using the large language model. The experimental results indicate that the proposed method achieves an accuracy of 93.9% in disassembly sequence generation across five mainstream power batteries, demonstrating the excellent feasibility and effectiveness.
Input:KG、MATCH (n:{Label})-[r]->(m) RETURN n, r, m
Output:拆解序列子图 LabelSubgraph
1:LabelSubgraph ← O ∥ 初始化子图为空
2: for each node n in KG do
3: if hasLabel(n, modelLabel) then
4: LabelSubgraph ← LabelSubgraph ∪ {n} ∥ 将节点添加到子图
5: for each node n1 in LabelSubgraph do
6: for each node n2 in LabelSubgraph do
7: if n1 is connected to n2 in KG then
8: add the relationship between n1 and n2 to LabelSubgraph
9: for each node n in LabelSubgraph do
10: for each attribute a of n do
11: add a to the attributes of n in LabelSubgraph
12: return LabelSubgraph ∥ 返回拆解序列子图
| [1] |
陈铭, 王志刚, 张宇平. 关于退役动力电池机器人拆解智能化的思考[J]. 汽车与配件, 2023(5): 54-58. |
| [2] |
|
| [3] |
程贤福, 周健, 肖人彬, |
| [4] |
|
| [5] |
|
| [6] |
张雷, 彭宏伟, 卞本阳, |
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
|
| [11] |
刘志峰, 杨德军, 顾国刚. 基于模拟退火粒子群优化算法的拆卸序列规划[J]. 合肥工业大学学报(自然科学版), 2011, 34(2): 161-165. |
| [12] |
|
| [13] |
梁佩, 付亚平, 田广东. 面向多产品拆解的拆解序列计划问题的模型与算法[J]. 计算机集成制造系统, 2023, 29(7): 2166-2174. |
| [14] |
|
| [15] |
郭洪飞, 陈志彬, 任亚平, |
| [16] |
|
| [17] |
郭恒, 黎荣, 张海柱, |
| [18] |
|
| [19] |
|
| [20] |
江志刚, 谢彬, 朱硕, |
| [21] |
|
| [22] |
|
| [23] |
|
| [24] |
|
| [25] |
|
| [26] |
|
| [27] |
|
| [28] |
|
国家自然科学基金(52075396)
武汉科技大学“十四五”湖北省优势特色学科项目(2023B0405)
/
| 〈 |
|
〉 |