| 引用本文: | 牟天昊,邹媛媛,李少远.工业过程关键指标预测的知识协同进化增强图卷积网络方法(英文)[J].控制理论与应用,2024,41(3):416~427.[点击复制] |
| MOU Tian-hao,ZOU Yuan-yuan,LI Shao-yuan.Enhancing graph convolutional network of knowledge-based co-evolution for industrial process key variable prediction[J].Control Theory and Technology,2024,41(3):416~427.[点击复制] |
|
| 工业过程关键指标预测的知识协同进化增强图卷积网络方法(英文) |
| Enhancing graph convolutional network of knowledge-based co-evolution for industrial process key variable prediction |
| 摘要点击 2892 全文点击 268 投稿时间：2022-05-06 修订日期：2023-05-27 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2023.20352 |
| 2024,41(3):416-427 |
| 中文关键词 关键指标预测 流程工业 知识挖掘 图卷积神经网络 数据–知识驱动建模 脱丁烷塔 |
| 英文关键词 key variable prediction process industry knowledge mining graph convolutional network data-knowledge driven modeling debutanizer column process |
| 基金项目 国家重点基础研究发展计划 (2018AAA0101701),国家自然科学基金(61833012, 62173224). |
|
| 中文摘要 |
| 在流程工业关键变量预测领域, 已有研究致力于将过程知识与大数据相结合, 以实现更高的准确性, 降低过拟
合风险和提高可解释性. 然而, 现有工作存在准确的先验知识构建成本高、无法从丰富的数据中挖掘知识等问题, 限制
了这些方法在实际工业过程中的广泛应用. 为了解决这些挑战, 本文提出了一种基于知识协同进化的增强图卷积网络方
法. 首先, 利用易获取的过程流图构建低成本的粗粒度流程知识. 然后, 在图卷积神经网络模型训练中引入图探索, 实现
知识更新. 最后, 为了降低知识复杂度并保持一致性, 设计了一种知识过滤机制. 所提出的方法在基准的脱丁烷塔工艺
过程上进行了验证. 实验结果表明, 该方法具有出色的预测准确性, 并获得高质量的新知识. |
| 英文摘要 |
| Efforts have been made in the field of industrial process key variable prediction to integrate process knowledge
with big data in order to achieve higher accuracy, reduce the risk of overfitting, and improve interpretability. However,
existing approaches face challenges such as the high cost of constructing accurate prior knowledge and the inability to
extract knowledge from abundant data, which limits their applicability to real industrial processes. To address these issues,
this study proposes an enhanced graph convolutional network of knowledge-based co-evolution (KBCE-GCN) method for
industrial process key variable prediction. Initially, a coarse-grained process knowledge is constructed from an easily accessible
process flow diagram, requiring minimal construction cost. Subsequently, graph exploration is introduced in GCN
model training to update the knowledge. Finally, a knowledge filtering mechanism is designed to reduce the complexity of
the knowledge and maintain consistency. The proposed KBCE-GCN method is validated using a benchmark debutanizer
column process. The experimental results demonstrate excellent prediction accuracy and the acquisition of high-quality
new knowledge. |
|
|
|
|
|