Abstract: In view of the strong temporal and associative coupling characteristics of process knowledge,a method of process quality prediction based on temporal knowledge graph and Convolutional Neural Network—Long Short Term Memory (CNN-LSTM) was proposed.The knowledge graph embedding was used to extract the semantic features of the knowledge graph,such as heterogeneous data,process standards and specification requirements,and complex process standards were modeled as the implicit correlation between complex triples according to representation of entities and relations.On this basis,the set of triples of subgraphs were mapped to the low-dimensional vector space to capture the semantics indirectly,and the feature fusion was used to strengthen the temporal features as inputs.Then,the combinational neural network model based on attention mechanism was constructed to extract the significant temporal features,and finally the quality link prediction of knowledge graph was realized.The experimental results showed that the accuracy of the process quality prediction method based on temporal knowledge graph and CNN-LSTM were superior to other methods,which verified the effectiveness and efficiency of the proposed model.In view of the strong temporal and associative coupling characteristics of process knowledge,a method of process quality prediction based on temporal knowledge graph and Convolutional Neural Network—Long Short Term Memory (CNN-LSTM) was proposed.The knowledge graph embedding was used to extract the semantic features of the knowledge graph,such as heterogeneous data,process standards and specification requirements,and complex process standards were modeled as the implicit correlation between complex triples according to representation of entities and relations.On this basis,the set of triples of subgraphs were mapped to the low-dimensional vector space to capture the semantics indirectly,and the feature fusion was used to strengthen the temporal features as inputs.Then,the combinational neural network model based on attention mechanism was constructed to extract the significant temporal features,and finally the quality link prediction of knowledge graph was realized.The experimental results showed that the accuracy of the process quality prediction method based on temporal knowledge graph and CNN-LSTM were superior to other methods,which verified the effectiveness and efficiency of the proposed model.

Key words: knowledge graph, TransH model, feature fusion, convolutional neural network—long short term memory neural network, quality prediction

摘要： 针对流程生产工艺知识的强时序性和关联耦合特征,提出一种融合时序知识图谱与CNN-LSTM的工艺质量预测模型。首先利用知识嵌入技术提取多源异构数据、工艺标准、规范要求等知识图谱语义特征,根据实体和关系表示将复杂工艺标准建模为复合的三元组间的隐含关联关系;在此基础上,将子图三元组集映射至低维向量空间以间接捕获语义,通过特征融合强化时序特征作为输入,构建基于注意力机制的组合神经网络模型以提取显著时序特征,最终实现面向流程生产工艺的质量预测。实验结果表明,基于时序知识图谱与CNN-LSTM的流程生产工艺质量预测方法精确率优于其他方法,验证了所提模型的有效性与高效性。针对流程生产工艺知识的强时序性和关联耦合特征,提出一种融合时序知识图谱与CNN-LSTM的工艺质量预测模型。首先利用知识嵌入技术提取多源异构数据、工艺标准、规范要求等知识图谱语义特征,根据实体和关系表示将复杂工艺标准建模为复合的三元组间的隐含关联关系;在此基础上,将子图三元组集映射至低维向量空间以间接捕获语义,通过特征融合强化时序特征作为输入,构建基于注意力机制的组合神经网络模型以提取显著时序特征,最终实现面向流程生产工艺的质量预测。实验结果表明,基于时序知识图谱与CNN-LSTM的流程生产工艺质量预测方法精确率优于其他方法,验证了所提模型的有效性与高效性。

关键词: 知识图谱, TransH模型, 特征融合, CNN-LSTM神经网络, 质量预测