A Novel Prediction Method for Blast Furnace Gas Utilization Rate Based on Dynamic Weighted Stacked Output‐Relevant Autoencoder

Abstract

The iron and steel industry has the characteristics of high energy consumption and large environmental pollution. Blast furnace (BF), a vital and energy-intensive unit, consumes more than half of energy and cost in the whole iron and steel manufacturing production process. [1][2][3] The energy consumption and operating status of BF are highly related to the gas flow distribution, fuel consumption, and so on. The gas utilization rate (GUR) represents the proportion of CO converted to CO 2 in the BF, which not only determines the rationality of gas flow distribution, but is also a concerned indicator for on-site operators to evaluate BF's energy consumption and operating status. [4] Therefore, it is significant to accurately obtain GUR online. However, BF is a typical black-box system that lacks direct and effective measuring means to understand the internal state of the furnace, [5,6] making accurately obtaining GUR in real-time is not easy.Over the last decades, many experts and scholars have been dedicated to the research for GUR; in general, these researches can be divided into two categories: mechanism-driven [7][8][9][10] and datadriven. [4,[11][12][13][14][15] Mechanism-driven methods have the advantage of strong interpretability, for instance, Kou et al. [7] proposed a 3D mechanism model to analyze the effect of different blast parameters on the GUR. Xiang et al. [8] combined the BF ironmaking mechanism with the actual production data and investigated the association of GUR with other production efficiency indicators. These works mainly focus on analyzing several factors affecting GUR through mechanism models. However, BF ironmaking has the characteristics of high nonlinearity, multivariate, and strong coupling, which makes mechanism-driven methods have inevitable modeling errors. Data-driven methods do not require prior knowledge; as long as the data quality and quantity are guaranteed, accurate prediction models can be built. With the rapid development of computer science and information technology, large amounts of production data have been collected and stored in BF historical database, which provides adequate support for data-driven methods. Therefore, data-driven methods have drawn more and more attention, as well as have been increasingly employed for GUR prediction. Zhang et al. [4] supposed that the operation status of BF is reflected by top gas and proposed a nonlinear identification method to predict GUR. Li et al. [12] proposed a novel GUR prediction method named DU-OS-ELM, which introduced a dynamic forgetting factor and an updated selection strategy into the original extreme learning machine (ELM) model. Besides, an improved ELM based on gray relational analysis and residual modification mechanism is also proposed to predict GUR. [13] Dang et al. [15] investigated the correlation of gas temperature distribution with GUR and developed a GUR prediction model based on support vector regression (SVR), as well as used an optimization algorithm to search for the best model parameters. The

A Deep Learning‐Based Approach to the Estimation of Jominy Profile of Medium‐Carbon Quench Hardenable Steels

“…In data and image preprocessing, AE are applied for various purposes, such as dimensionality reduction and anomaly detection. In the former application, by learning a compressed representation of the data, AE can reduce the dimensionality of high‐dimensional input data, and this is particularly useful for tasks such as, for instance, data visualization, variables selection, [ 51 ] feature extraction, [ 52 ] and denoising. [ 53 ] Within anomaly detection tasks, AE can learn to reconstruct normal patterns in the input data.…”

Evaluation, Prediction, and Feedback of Blast Furnace Hearth Activity Based on Data‐Driven Analysis and Process Metallurgy

“…Data-driven modeling and machine learning algorithms have achieved important and new applications in the metallurgical industry. [16][17][18][19][20][21] However, little research has been conducted on the hearth activity of BFs. Deng et al [22] selected the temperature ratio as the characterization parameter of hearth activity based on data mining and established a quantitative model.…”