Automating Prediction of Roller Mills Performance in a Concentrator Using a Predictive Model by an Artificial Neural Network

In this article, the authors propose to consider a multicomponent approach to evaluating the efficiency of roller mills in an enrichment plant, based on the use of a set of pre-built various mathematical models. The multicomponent approach makes it possible to predict the volume of finished concentrate output reliably during the period of the factory's transition to the use of ore with changed characteristics. Automation of forecasting will make it possible to analyze the effect of the physical and chemical characteristics of the feedstock adequately on the productivity of the main mills of the processing plant and plan the volumes of the concentrate. During the study, a database was created, the array size of which was 432x28: 432 is the number of daily data, and 28 is the total number of factors. Clustering methods were implemented by the software of the MATLAB R2021a application package, which made it possible to divide the generated database into clusters. As a result, the homogeneity of the data array connected by the same type of links was proved and, accordingly, the confirmation of the validity of combining the arrays of sections No. 1, 2 and 3 in order to form a single database for further processing and building a predictive model. Using the parallel use of the Statistica and Matlab packages, an optimal regression model was obtained. Artificial neural networks were formed with a different number of hidden layers.