Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transformer-based Drum-level Prediction in a Boiler Plant with Delayed Relations among Multivariates

Gang Su, Sun Yang, Zhishuai Li

TL;DR

The paper addresses the challenge of predicting steam drum water level in boilers, where long, variable delays and interrelated multivariate dynamics complicate forecasting. It introduces a Transformer-based time-series framework enriched with causal relation analysis, optimal time-delay inference, and feature augmentation to capture long-range dependencies. Key contributions include Granger-causality screening of direct drum-level drivers, delay estimation (e.g., around $t_{\text{delay}}$ values like 212 s for some factors), and a MISO Transformer backbone that achieves near-zero bias with a small dispersion (two-sigma about 5 mm) on real plant data comprising 1,048,574 samples. The approach enables more accurate and stable drum-level predictions, facilitating proactive control strategies and improving operational stability in power plants.

Abstract

The steam drum water level is a critical parameter that directly impacts the safety and efficiency of power plant operations. However, predicting the drum water level in boilers is challenging due to complex non-linear process dynamics originating from long-time delays and interrelations, as well as measurement noise. This paper investigates the application of Transformer-based models for predicting drum water levels in a steam boiler plant. Leveraging the capabilities of Transformer architectures, this study aims to develop an accurate and robust predictive framework to anticipate water level fluctuations and facilitate proactive control strategies. To this end, a prudent pipeline is proposed, including 1) data preprocess, 2) causal relation analysis, 3) delay inference, 4) variable augmentation, and 5) prediction. Through extensive experimentation and analysis, the effectiveness of Transformer-based approaches in steam drum water level prediction is evaluated, highlighting their potential to enhance operational stability and optimize plant performance.

Transformer-based Drum-level Prediction in a Boiler Plant with Delayed Relations among Multivariates

TL;DR

The paper addresses the challenge of predicting steam drum water level in boilers, where long, variable delays and interrelated multivariate dynamics complicate forecasting. It introduces a Transformer-based time-series framework enriched with causal relation analysis, optimal time-delay inference, and feature augmentation to capture long-range dependencies. Key contributions include Granger-causality screening of direct drum-level drivers, delay estimation (e.g., around values like 212 s for some factors), and a MISO Transformer backbone that achieves near-zero bias with a small dispersion (two-sigma about 5 mm) on real plant data comprising 1,048,574 samples. The approach enables more accurate and stable drum-level predictions, facilitating proactive control strategies and improving operational stability in power plants.

Abstract

The steam drum water level is a critical parameter that directly impacts the safety and efficiency of power plant operations. However, predicting the drum water level in boilers is challenging due to complex non-linear process dynamics originating from long-time delays and interrelations, as well as measurement noise. This paper investigates the application of Transformer-based models for predicting drum water levels in a steam boiler plant. Leveraging the capabilities of Transformer architectures, this study aims to develop an accurate and robust predictive framework to anticipate water level fluctuations and facilitate proactive control strategies. To this end, a prudent pipeline is proposed, including 1) data preprocess, 2) causal relation analysis, 3) delay inference, 4) variable augmentation, and 5) prediction. Through extensive experimentation and analysis, the effectiveness of Transformer-based approaches in steam drum water level prediction is evaluated, highlighting their potential to enhance operational stability and optimize plant performance.
Paper Structure (14 sections, 2 equations, 8 figures, 2 tables)

This paper contains 14 sections, 2 equations, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Data Preprocessing and Feature Engineering
  5. Causal Relationships
  6. Optimal Time Delay Selection
  7. Variable Augmentation
  8. Model Architecture Design
  9. Model Training and Evaluation
  10. Experiments
  11. Data
  12. Benchmark Methods
  13. Results
  14. Conclusion

Figures (8)

  • Figure 1: Visualization of selected variables.
  • Figure 2: The Proposed Framework.
  • Figure 3: A visualization of the steam drum water level control system with relevant variables positioned accordingly: the variables of the most interest are highlighted with color yellow, and the factors in blue and red with a dotted link need to be further analyzed.
  • Figure 4: The factors that have a direct causal link to drum level.
  • Figure 5: Delay Analysis: drum level is in blue, and the factor whose delay is to be calculated, e.g., feedwater flow, is in red. Top: before the shifting, the feedwater flow (red) arrives the peak earlier than drum level (blue); Bottom: After calculating the shifting according to Eq. (2), the optimal delay is 212 seconds and the two lines are synced (meaning their covariance is maximized).
  • ...and 3 more figures