The Journal of China Universities of Posts and Telecommunications ›› 2019, Vol. 26 ›› Issue (6): 63-72.doi: 10.19682/j.cnki.1005-8885.2019.1027

• Artificial intelligence • Previous Articles     Next Articles

Underflow concentration prediction model of deep-cone thickener based on data-driven

Wang Huan, Liu Ting, Cao Yuning, Wu Aixiang   

  1. 1. School of Civil and Environmental Engineering, University of Science and Technology Beijing, Beijing 100083, China
    2. School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing 100083, China
  • Received:2018-08-07 Revised:2019-12-23 Online:2019-12-31 Published:2020-03-10
  • Contact: Wu Aixiang, E-mail:
  • About author:Wu Aixiang, E-mail:
  • Supported by:
    This work was supported by the National Key Research and Development Program of China (2016YFB0700500), the National Science Foundation of China (61572075, 61702036), Fundamental Research Funds for the Central Universities (FRF-TP-17-012A1), and China Postdoctoral Science Foundation (2017M620619).


The underflow concentration prediction of deep-cone thickener is a difficult problem in paste filling. The existing prediction model only determines the influence of some parameters on the underflow concentration, but lacks a prediction model that comprehensively considers the thickening process and various factors. This paper proposed a model which analyzed the variation of the underflow concentration from a number of influencing factors in the
concentrating process. It can accurately predict the underflow concentration. After preprocessing and feature selection of the history data set of the deep-cone thickener, this model uses the eXtreme gradient boosting (XGBOOST) in machine learning to deal with the relationship between the influencing factors and the underflow concentration, so as to achieve a more comprehensive prediction of the underflow concentration of the deep-cone thickener. The experimental results show that the underflow concentration prediction model based on XGBOOST shows a mean absolute error (MAE) of 0.31% and a running time of 1.6 s on the test set constructed in this paper, which fully meet the demand. By comparing the following three classical algorithms: back propagation (BP) neural network, support vector regression (SVR) and linear regression, we further verified the superiority of XGBOOST under the conditions of this study.

Key words:

paste filling, underflow concentration, machine learning, XGBOOST, prediction model

CLC Number: