Data and knowledge driven approach for burden surface optimization in blast furnace,☆☆

https://doi.org/10.1016/j.compeleceng.2021.107191Get rights and content

Highlights

  • A hybrid optimization strategy is designed to find burden surface setting values.

  • Computational-efficiency data-driven process models are established.

  • A multiobjective optimization and CBR based information fusion strategy is performed.

  • Production status evaluation and knowledge-based adjustment model are proposed.

Abstract

This paper presents a hybrid optimization strategy for determining the setting values of burden surface through measured data and domain knowledge integration manner. The proposed hybrid optimization strategy, including broad learning-based soft sensing models for production indicators, novel twin information fusion based pre-setting model, knowledge-mining based feedback compensation model, data-based production status evaluation and knowledge-based adjustment model, can adjust the setting values of burden surface in response to the changes in production status and safe operation can be reached as well. Finally, comprehensive experiments are conducted to verify the effectiveness and feasibility of the proposed method.

Introduction

Blast furnace (BF) represents a fundamental component in iron production. The main purpose of BF ironmaking is to keep smooth and stable operation, and further achieve energy-saving and consumption-reducing production [1], [2]. In order to achieve these goals, operational optimization of production process is necessary, mainly referring to the burden surface profile optimization [3]. However, the related dynamical characteristics are generally complicated due to the complex nonlinearities and strong couplings among process variables [4], [5], leading to the inaccurate mechanism models [6], [7]. Practically, the setting of parameters corresponding to the burden surface are prescribed based on operator’s experience to maintain the production performance. Furthermore, the performance consists of multiple objectives subjecting to actual constraints. Accordingly, operators cannot decide the appropriate setting values fast and accurately for the frequent changes of the production status. Therefore, it is desired to find the optimal setting of burden surface to ensure the key production indicators within their target ranges.

The past few years have witnessed the increasing attention of developing the operational optimization methodologies for complex industrial processes [8], [9]. In fact, data-driven technique is gradually preferred by both researchers and engineers due to the high complexity of those industrial processes. For example, Jiang et al. [10] proposed a model-free data-driven operational optimal control approach based on reinforcement learning and applied to single-cell flotation industrial process. Dai et al. [11] presented a data-driven optimization control approach for safe operation of mineral grinding process. Ding et al. [12] developed an online automated decision procedure for operational indices using measured data from a large mineral processing factory of hematite iron ore. Galvani et al. [13] utilized point estimation method to model the probabilistic nature of power systems and proposed a multiobjective framework for optimal location and parameter setting. Zhao et al. [14] proposed a data-based predictive optimization method for the real-time gas adjusting problem encountered in steel industry. Even though the data-based optimization methods for industrial processes can exhibit relatively satisfactory solution capability, it may still become inaccurate over a long time of operation. The main reason is that those kinds of industrial processes encounter uncertainty, which bring in various operational challenges and lead to increase of difficulty in operational decision. Furthermore, the production status of BF is usually time-varying. Occasionally, the status of BF is not working smoothly. In order to maintain its satisfactory and safe performance, operators are always required to give the proper setting values based on their accumulated experience. Therefore, the status evaluation and knowledge-based adjustment need to be considered to achieve better optimization performance.

Accordingly, the massive measured data and a lot of domain knowledge of production process can be simultaneously used to discover the production principles, which have great potential for operational optimization to improve the robustness of the decision. In this paper, a modified hybrid optimization strategy is presented to find the optimal setting values of burden surface by integrating measured data and domain knowledge. Specifically, the proposed strategy includes four phases: (1) data-driven modeling, it uses the measured data to build process models based on broad learning (BL) algorithm; (2) pre-setting determination, it develops a twin information fusion strategy by combining the multiobjective optimization algorithm and case-based reasoning (CBR) method based on ensemble learning to obtain the baseline setting values; (3) feedback compensation, it employs data mining technique to analyze the implicit knowledge to get corrected values; (4) production status evaluation and knowledge-based adjustment, it detects the production conditions along with incremental adjustments to ensure the safe production. The proposed strategy combines BF production requirements with safe constraints, which we envision to be a powerful theoretical framework. Comprehensive experiments validate the effectiveness and feasibility of the proposed strategy. The main contributions of this paper are highlighted as follows:

(1) The computational-efficiency data-driven process models are established to map the nonlinear relationship between production indicators and burden surface parameters using BL algorithm.

(2) In order to make full use of the characteristics of CBR fusing expert knowledge and dynamic adaptive ability of heuristic algorithm, a twin information fusion strategy by combining the multiobjective optimization algorithm and CBR method is performed to obtain the baseline setting values of burden surface.

(3) The inherent existence of uncertainty may lead to the improper setting values, the production status evaluation and knowledge-based adjustment model are proposed to eliminate fault using the corrected setting values.

It should be noted that the proposed strategy is different from our previous work [15], which focuses on the feedback compensation of the setting values of burden surface. As mentioned above, the present work includes the functions in [15] as well as integrating domain knowledge to make it more robust. To our best knowledge, this is the first attempt to deeply integrate measured data and domain knowledge for burden surface optimization.

The rest of this paper is arranged as follows: Section 2 describes the characteristics of BF ironmaking process and the challenges of burden surface optimization. Section 3 presents the proposed hybrid optimization strategy for optimal setting values of burden surface in detail. In Section 4, the effectiveness and feasibility of the proposed strategy is verified with comprehensive experiments. Finally, conclusions and future works are given in Section 5.

Section snippets

Problem statement

This section focuses on the statement on characteristics of BF ironmaking process and challenges of burden surface optimization, which motivates the problem formulation.

Hybrid optimization strategy

In this section, a hybrid optimization strategy for solving the burden surface optimization problem is presented. The essence is to perform the setting values determination through measured data and domain knowledge integration manner.

Experimental results and analysis

In this section, comprehensive experiments are carried out to validate the effectiveness and feasibility of the proposed hybrid optimization strategy. The samples come from the practice of a medium-size BF and production records in the experiments. We first compare BL with support vector machine (SVM) and extreme learning machine (ELM) to verify the performance of BL-based prediction model. Then, the production status evaluation model is examined. Finally, the experiments of setting value

Conclusions

In this paper, a hybrid optimization strategy is presented for determining the appropriate setting values of burden surface to keep the key production indicators within the corresponding target ranges by integrating measured data and domain knowledge. The proposed strategy has three prominent advantages, including: first, BL-based soft sensing models estimate the production indicators fast and accurately; second, twin information fusion strategy is employed to obtain the baseline setting values

CRediT authorship contribution statement

Yanjiao Li: Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Writing - original draft, Funding acquisition. Huiqi Li: Supervision, Writing - original draft. Jie Zhang: Conceptualization, Validation, Writing - original draft, Funding acquisition. Sen Zhang: Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported in part by the National Natural Science Foundation of China under Grant 62003038, China Postdoctoral Science Foundation under Grants 2019TQ0002 and 2019M660328, and the Open Funding of the Key Laboratory of Knowledge Automation for Industrial Processes of Ministry of Education, University of Science and Technology Beijing .

Yanjiao Li received the Ph.D. degree from University of Science and Technology Beijing, China, in 2019. She was a visiting research fellow with University of Leeds, U.K., in 2018. She is currently a postdoctoral research fellow with the School of Information and Electronics, Beijing Institute of Technology, China. Her research interests include machine learning, modeling and optimization for complex systems.

References (30)

  • ZhouP. et al.

    Data-driven robust M-LS-SVM-based NARX modeling for estimation and control of molten iron quality indices in blast furnace ironmaking

    IEEE Trans Neural Netw Learn Syst

    (2018)
  • LiY. et al.

    Burden surface decision using MODE with TOPSIS in blast furnace ironmaking

    IEEE Access

    (2020)
  • LuX. et al.

    Data-driven optimal control of operational indices for a class of industrial processes

    IET Control Theory Appl

    (2016)
  • ZhuY. et al.

    Dual RBFFNNs-based model-free adaptive control with aspen HYSYS simulation

    IEEE Trans Neural Netw Learn Syst

    (2017)
  • JiangY. et al.

    Data-driven flotation industrial process operational optimal control based on reinforcement learning

    IEEE Trans Ind Inform

    (2018)
  • Cited by (5)

    Yanjiao Li received the Ph.D. degree from University of Science and Technology Beijing, China, in 2019. She was a visiting research fellow with University of Leeds, U.K., in 2018. She is currently a postdoctoral research fellow with the School of Information and Electronics, Beijing Institute of Technology, China. Her research interests include machine learning, modeling and optimization for complex systems.

    Huiqi Li received the Ph.D. degree from Nanyang Technological University, Singapore, in 2003. She is currently a professor with the School of Information and Electronics, Beijing Institute of Technology, China. Her research interests include image/signal processing and computer-aided diagnosis.

    Jie Zhang received the Ph.D. degree from University of Science and Technology Beijing, China, in 2019. He was a visiting research fellow with University of Leeds, U.K., in 2018. He is currently a postdoctoral research fellow with the School of Electronics Engineering and Computer Science, Peking University, China. His research interests include machine learning, deep learning, and wireless sensor networks.

    Sen Zhang received the Ph.D. degree from Nanyang Technological University, Singapore, in 2005. She is currently a professor with the School of Automation and Electrical Engineering, University of Science and Technology Beijing, China. Her research interests include machine learning, target tracking and estimation theory.

    This paper was recommended for publication by associate editor Shiqian Wu.

    ☆☆

    This paper is for special section VSI-aiia. Reviews processed and recommended for publication by Guest Editor Prof. Shiqian Wu.

    View full text