Prediction Method of Grinding Process Parameters for a Segment Based on Knowledge Graph

A method combining knowledge graph and multivariate neural network is proposed to address the difficulties in predicting the quality of primary grinding and the large fluctuation of overflow particle size in beneficiation processes. Firstly, information was extracted from multi-source heterogeneous data such as worker experience, technical standard documents, and production specifications, and a unified knowledge graph database was constructed. Then, the knowledge graph is transformed into feature vectors using the word vector transformation tool word2vec. Finally, apply the constructed CNN-BiLSTM-Att model for quality prediction and output the results. Case studies have shown that the proposed model is effective and feasible. This method achieves the process of converting qualitative data into quantitative data and further converting it into qualitative output, providing a new approach and method for predicting the quality of a grinding process.