Effective variable selection and moving window HMM-based approach for iron-making process monitoring

Due to the characteristics of non-linearity, large time delay, time-variety, high-dimension and distributed parameters of blast furnace smelting process, it brought great difficulties to the detection and diagnosis of abnormal furnace conditions. Although there have been some researches on the fault detection of blast furnace smelting process, works concerning on the fault classification are still limited. And there is no exploration to deal with it based on hidden Markov model (HMM). Taking into account the inherent uncertainty and dynamics of the blast furnace smelting process, this article proposes a novel iron-making process monitoring approach based on effective variable selection and moving window HMM (VS-MVHMM). First, considering that not all variables are beneficial to fault identification, a variable selection method based on the coefficient of variation is used. Second, instead of just considering the posterior probability of one single sample, moving window is introduced to utilize the dependency of samples for improving the accuracy of on-line fault identification. Besides, a MVHMM-based threshold statistic is defined to identify the unknown fault in iron-making process. And various known modes are separated based on Viterbi algorithm. The effectiveness of the proposed approach is first validated by a numerical simulation example and Tennessee Eastman (TE) simulation platform. Then it is tested by the real data, which is generated in the No. 2 blast furnace of Liuzhou Iron and Steel Group Corporation.