面向知识图谱的智能生产系统工艺知识推理方法

Knowledge Graph-based Process Knowledge Reasoning Method for Intelligent Production System

针对智能生产系统中知识和数据之间存在冗余度高、关联性弱,难以进行知识推理的缺点,提出了面向知识图谱的工艺知识推理方法。对输入信息进行语义标注和分类,提取得到用于信息匹配的特征;利用图卷积方法对提取到的局部特征和全局特征进行关联,整合成带有差异值信息的特征无向图,与构建好的知识图谱进行图匹配操作;根据不同的推理类型采用不同的推理规则,推理得出实例之间的关联和拓扑信息,生成关于输入信息的属性和值。与推荐系统算法、融合知识图谱的图卷积算法相比,该算法在MovieLens-1M、Book-Crossing、Last.FM数据集上取得了较高的预测率,通过工艺实例知识推理验证了所提推理模型的可行性。

Aiming at the disadvantages of high redundancy and weakness between knowledge and data in intelligent production system,and the difficulty to perform knowledge reasoning,a process knowledge reasoning method for knowledge maps is proposed.The input information is semantically labeled and classified,the characteristics of the information match are extracted,the extracted local feature and global feature are associated through graph convolution method,and the feature of the difference value information is integrated and mapped with the constructed knowledge graph.Different reasoning rules are used according to different reasoning types,and the association and topology information between instances are deduced,and the properties and values of input information are generated.Compared with the recommended system algorithm and fusion knowledge map,the algorithm has higher predictive rates on Movielens-1M,Book-Crossing,Last.FM data set.The feasibility of the reasoning model is verified by the process instance knowledge.