Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Spatio-Temporal Consistent Soft Sensor Modeling and Monitoring of Thermal Power Plants based on Physical Knowledge

Qianchao Wang, Peng Sha, Leena Heistrene, Yuxuan Ding, Yaping Du

TL;DR

This work tackles the challenge of interpretable spatio-temporal soft sensors for thermal power plants by proposing STCIM, a two-stage architecture that first extracts and aligns local and distant spatio-temporal features and then maps them into a physics-informed latent space. The latent space is constrained by a discrete state-space loss, enabling latent variables to carry explicit physical meaning while maintaining predictive accuracy. Across two synthetic datasets and real 330MW and 1000MW plant data, STCIM demonstrates superior generalization, with ablation studies showing that adaptive position encoding and far-topological alignment are critical to performance. The approach offers a practical pathway to safer, more reliable monitoring and control in complex industrial processes where physical laws and data-driven insights must cohere.

Abstract

Data-driven soft sensors have been widely applied in complex industrial processes. However, the interpretable spatio-temporal features extraction by soft sensors remains a challenge. In this light, this work introduces a novel method termed spatio-temporal consistent and interpretable model (STCIM). First, temporal and spatial features are captured and aligned by a far topological spatio-temporal consistency extraction block. Then, the features are mapped into an interpretable latent space for further prediction by explicitly giving physical meanings to latent variables. The efficacy of the proposed STCIM is demonstrated through the modeling of two generated datasets and a real-life dataset of coal-fired power plants. The corresponding experiments show: 1) The generalization of STCIM outperforms other methods, especially in different operation situations. 2) The far topological spatio-temporal consistency is vital for feature alignment. 3) The hyper-parameters of physics-informed interpretable latent space loss decide the performance of STCIM.

Spatio-Temporal Consistent Soft Sensor Modeling and Monitoring of Thermal Power Plants based on Physical Knowledge

TL;DR

This work tackles the challenge of interpretable spatio-temporal soft sensors for thermal power plants by proposing STCIM, a two-stage architecture that first extracts and aligns local and distant spatio-temporal features and then maps them into a physics-informed latent space. The latent space is constrained by a discrete state-space loss, enabling latent variables to carry explicit physical meaning while maintaining predictive accuracy. Across two synthetic datasets and real 330MW and 1000MW plant data, STCIM demonstrates superior generalization, with ablation studies showing that adaptive position encoding and far-topological alignment are critical to performance. The approach offers a practical pathway to safer, more reliable monitoring and control in complex industrial processes where physical laws and data-driven insights must cohere.

Abstract

Data-driven soft sensors have been widely applied in complex industrial processes. However, the interpretable spatio-temporal features extraction by soft sensors remains a challenge. In this light, this work introduces a novel method termed spatio-temporal consistent and interpretable model (STCIM). First, temporal and spatial features are captured and aligned by a far topological spatio-temporal consistency extraction block. Then, the features are mapped into an interpretable latent space for further prediction by explicitly giving physical meanings to latent variables. The efficacy of the proposed STCIM is demonstrated through the modeling of two generated datasets and a real-life dataset of coal-fired power plants. The corresponding experiments show: 1) The generalization of STCIM outperforms other methods, especially in different operation situations. 2) The far topological spatio-temporal consistency is vital for feature alignment. 3) The hyper-parameters of physics-informed interpretable latent space loss decide the performance of STCIM.

Paper Structure

This paper contains 22 sections, 10 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Brief Introduction of Thermal Power Plants
  3. Spatio-Temporal Consistent and Interpretable Modeling
  4. Model Structure
  5. Spatio-Temporal Consistent Feature Extraction Block
  6. Local Spatio-Temporal Feature Extraction Block
  7. Far Topological Alignment Block
  8. Physics-informed Interpretable Latent Space Loss Function
  9. Experimental Results
  10. Experiment Setup
  11. Analytical Experiments
  12. Rotor Equation-Reduced Rotary Speed
  13. 3 Order Synchronous Generator
  14. Comparative Experiments
  15. The comparisons of different methods
  16. ...and 7 more sections

Figures (6)

  • Figure 1: 330MW thermal power plant.
  • Figure 2: Overview of the STCIM framework. (a) The flow chart of STCIM. (b) The local spatio-temporal feature extraction block. (c) The far topological alignment block. (d) The physics-informed interpretable latent space loss function
  • Figure 3: Evaluation metrics of rotor equation reduced rotary speed.
  • Figure 4: The output forecasting and state variables monitoring of comparative experiments
  • Figure 5: Evaluation metrics of 1000MW thermal power plant.
  • ...and 1 more figures