Journal of Northeastern University(Natural Science) ›› 2024, Vol. 45 ›› Issue (3): 314-322.DOI: 10.12068/j.issn.1005-3026.2024.03.002

• Information & Control • Previous Articles     Next Articles

Prediction of Sinter Chemical Indexes Based on GMM-KNN-LSTM

Guang-lei XIA1, Zhao-xia WU1(), Meng-yuan LIU1, Yu-shan JIANG2   

  1. 1.School of Control Engineering,Northeastern University at Qinhuangdao,Qinhuangdao 066004,China
    2.School of Mathematics and Statistics,Northeastern University at Qinhuangdao,Qinhuangdao 066004,China.
  • Received:2022-11-07 Online:2024-03-15 Published:2024-05-17
  • Contact: Zhao-xia WU
  • About author:WU Zhao-xia,E-mail: ysuwzx@126.com

Abstract:

Aiming at the problem that unlabeled samples cannot be utilized by machine learning due to the low detection frequency of sinter chemical indexes,a prediction model for sinter chemical indexes that makes full use of the useful information in the samples is proposed. Firstly,the unlabeled samples are transformed into labeled samples by combining Gaussian mixture model (GMM) and K-nearest neighbor (KNN) algorithm,and then combined with long short-term memory (LSTM) unit for predicting three chemical indexes,namely,total Fe mass fraction,FeO mass fraction,and alkalinity of sinter. By comparing with the three models of back propagation neural network (BPNN),recurrent neural network (RNN), and LSTM,the results show that the proposed model has a low prediction error. The prediction hit rates of total Fe mass fraction and FeO mass fraction reach 98.73% and 95.33%,respectively within the allowable error of ±0.5%,and the prediction hit rate of alkalinity is 98.13% within the allowable error of ±0.05,demonstrating high prediction accuracy.

Key words: chemical indexes of sinter, prediction model, unlabeled samples processing algorithm, LSTM(long short?term memory), data preprocessing

CLC Number: