结合知识图谱与深度神经网络的烟丝松散回潮质量预测方法

Prediction Method of Loosening and Conditioning Quality of Cut Tobacco Based on Knowledge Graph and Deep Neural Network

针对制丝生产工艺中松散回潮质量预测困难,出料中水的质量分数和出料温度波动大等问题,课题组提出了结合知识图谱与深度神经网络的预测方法。该方法首先从工人经验、技术标准文件、生产规范等当中抽取多源异构数据构建出统一化的知识图谱数据库,然后通过词向量转换工具word2vec将知识图数据转换成可表示的二维向量,最后利用构建的BIGRU-Attention-KG模型进行预测、输出结果。经由案例验证表明所提出模型具有有效性和可行性。该方法实现了定性数据到定量数据再到定性输出的转换过程,为松散回潮质量预测提供了一种新的思路和方法。

In order to solve the problems such as difficulty in predicting the quality of loosening and conditioning and large fluctuation of moisture content and temperature of discharge in silk production process,a method combining knowledge graph and deep neural network was proposed.In this method,a unified knowledge graph database was constructed by extracting multi-source heterogeneous data from workers′experience,technical standard documents and production specifications.Then,the word vector conversion tool word2 vec was used to convert knowledge graph data into representable two-dimensional vectors.Next,the constructed BIGRU-Attention-KG model was used to predict the output results.Finally,the validity and feasibility of the proposed model were verified by a case study.The proposed method realizes the conversion process from qualitative data to quantitative data and then to qualitative output,and provides a new idea and method for the prediction of loosening and conditioning quality.