Real-Time Detection and Short-Term Prediction of Blast Furnace Burden Level Based on Space-Time Fusion Features

Yanli Chen, Zhipeng Chen, Weihua Gui, Chunhua YangSensors 2022 · OpenReview

TL;DR

This paper introduces a novel method for real-time monitoring and short-term prediction of blast furnace burden levels by fusing data from mechanical probes and radar sensors. By combining space-time features using a Long-Term Focus Memory Network (LFMN) and an Efficient Structure Self-Tuning RBF Neural Network (ESST-RBFNN), the approach achieves significantly improved accuracy and reliability compared to existing methods, crucial for optimizing industrial processes.

What Problem Does This Paper Address?

The iron and steel industry is a major energy consumer and carbon emitter, with blast furnaces being central to this. The "burden level" – the height of materials inside the furnace – is a critical parameter for efficient operation, energy consumption, and emission control. Current methods for measuring this level have significant drawbacks:

  • Mechanical Stock Rods: Provide accurate but discontinuous measurements, failing to capture real-time changes or detect issues like slips or collapses.
  • Radar Probes: Offer continuous, real-time data but are prone to interference from dust and gas within the harsh furnace environment, leading to accuracy issues, fluctuations, and poor stability.

This gap in reliable, continuous, and accurate burden level information hinders optimal blast furnace operation and process intelligence. The paper aims to bridge this by developing a method that leverages the strengths of both sensor types while mitigating their weaknesses.

Methods

The proposed solution is a multi-stage approach that fuses spatial and temporal features:

  1. Space Dimensional Feature Extraction:
    • A piecewise nonlinear regression fitting (specifically, a fifth-order polynomial) is applied to radar data.
  2. This model is then refined using mechanical probe data as a reference point to ensure continuity and accuracy, creating a smoother, more coherent representation of the burden level. This step aims to extract spatial features like falling speed and acceleration.
  3. Time Dimensional Feature Extraction and Prediction (LFMN):
    • A Long-Term Focus Memory Network (LFMN) is employed. This deep learning architecture is designed to capture slow-varying features in radar data, which are indicative of stable distribution mechanisms.
  4. Crucially, the LFMN incorporates an attention mechanism that uses the more reliable mechanical probe data to improve the prediction accuracy and reliability of the burden level for the next cycle.
  5. It then extracts time-dimensional features from the combined radar and predicted data.
  6. Real-Time Detection (ESST-RBFNN):
    • An Efficient Structure Self-Tuning Radial Basis Function Neural Network (ESST-RBFNN) is developed. This network dynamically adjusts its structure based on input data, making it adaptable to varying blast furnace conditions.
  7. It utilizes a novel Fast Eigenvector Space Clustering (ESC) algorithm to determine clustering centers and widths for input samples, improving training efficiency and enabling real-time detection.
  8. The ESST-RBFNN integrates both the time-dimensional features from LFMN and the space-dimensional features extracted earlier, along with key operating parameters, to provide a reliable and continuous burden level output.

Datasets

The study utilized actual detection data from a blast furnace in a steel plant.

  • Radar Probe Data: Sampled every 10 seconds.
  • Mechanical Stock Rod Data: Sampled every 3-6 minutes.

A total of 76,426 rows of data were used for model training and validation, with 10,000 rows specifically for validation. The width of the sliding window used for data processing was set to N=10.

Key Insights and Ideas

  • Synergy of Disparate Sensors: The core idea is that neither mechanical probes nor radar alone are sufficient. Their complementary strengths (accuracy vs. continuity) can be harnessed by fusing their data intelligently.
  • Space-Time Feature Fusion: The paper highlights that burden level dynamics have both slow-changing temporal patterns (predictable by LFMN) and rapid spatial changes. Effectively fusing these two types of features is key to robust prediction and detection.
  • Attention Mechanism for Reliability: Using the more reliable mechanical probe data as an "attention" signal within the LFMN helps guide the prediction process, improving accuracy even when radar data is noisy.
  • Dynamic Network Structure: The ESST-RBFNN's ability to self-tune its structure is crucial for adapting to the highly variable and harsh environment of a blast furnace, offering greater universality than static models.
  • ESC for Efficiency: The proposed ESC algorithm is vital for making the RBFNN suitable for real-time applications by efficiently determining cluster parameters.

Key Results

  • Prediction Accuracy (LFMN): The LFMN method demonstrated high confidence in predicting blast furnace burden levels and extracting time-dimensional features, showing good consistency with radar probe data trends, even during abnormal conditions. The coincidence degree with the blast furnace charging cycle reached 98.87%.
  • Detection Accuracy (ESST-RBFNN):
    • Under normal working conditions, the ESST-RBFNN showed significantly higher accuracy than radar detection and traditional RBFNN, with absolute errors within ±0.1 m and relative errors within ±5%.
  • Even under abnormal working conditions, the ESST-RBFNN maintained good tracking of mechanical probe data, with significantly improved reliability and accuracy compared to radar and traditional RBFNN.
  • Statistical analysis on 24-hour validation data showed the ESST-RBFNN achieved a hit rate of 91.17% for samples with relative error lower than 2% and 99.33% for errors lower than 5%.
  • Overall: The proposed method provides real-time, continuous, accurate, and reliable burden level information, demonstrating high practical value for industrial production.

Limitations and Future Work

  • Extremely Abnormal Conditions: The paper acknowledges that under some extremely abnormal working conditions with very large fluctuations, the accuracy and reliability of the detection method may decrease.
  • LFMN Reliance on Radar: The LFMN's reliance on slow variation features of radar data means that during significant burden level fluctuations, its predictions might not perfectly track the real trend.
  • Future Work: The authors suggest further study and improvement are needed to address these limitations, particularly in handling extreme abnormal conditions and refining the LFMN's ability to track rapid changes.

Summary updated

Abstract

Real-time, continuous and accurate blast furnace burden level information is of great significance for controlling the charging process, ensuring a smooth operation of a blast furnace, reducing energy consumption and emissions and improving blast furnace output. However, the burden level information measured by conventional mechanical stock rods and radar probes exhibit problems of weak anti-interference ability, large fluctuations in accuracy, poor stability and discontinuity. Therefore, a space-time fusion prediction and detection method of burden level based on a long-term focus memory network (LFMN) and an efficient structure self-tuning RBF neural network (ESST-RBFNN) is proposed. First, the space dimensional features are extracted by the space regression model based on radar data. Then, the LFMN is designed to predict the burden level and extract the time dimensional features. Finally, the ESST-RBFNN based on a proposed fast eigenvector space clustering algorithm (ESC) is constructed to obtain reliable and continuous burden level information with high accuracy. Both the simulation results and industrial verification indicate that the proposed method can provide real-time and continuous burden level information in real-time, which has great practical value for industrial production.
BibTeX
@inproceedings{chen2022real,
  author = {Yanli Chen and Zhipeng Chen and Weihua Gui and Chunhua Yang},
  title = {{Real-Time Detection and Short-Term Prediction of Blast Furnace Burden Level Based on Space-Time Fusion Features}},
  booktitle = {Sensors},
  year = {2022},
  url = {https://openreview.net/forum?id=GIXwcqXQFj},
}
Figure