Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A cGAN Ensemble-based Uncertainty-aware Surrogate Model for Offline Model-based Optimization in Industrial Control Problems

Cheng Feng

TL;DR

A novel cGAN ensemble-based uncertainty-aware surrogate model is introduced for reliable offline model-based optimization in industrial control problems and outperforms several competitive baselines in the field of offline model-based optimization for industrial control.

Abstract

This study focuses on two important problems related to applying offline model-based optimization to real-world industrial control problems. The first problem is how to create a reliable probabilistic model that accurately captures the dynamics present in noisy industrial data. The second problem is how to reliably optimize control parameters without actively collecting feedback from industrial systems. Specifically, we introduce a novel cGAN ensemble-based uncertainty-aware surrogate model for reliable offline model-based optimization in industrial control problems. The effectiveness of the proposed method is demonstrated through extensive experiments conducted on two representative cases, namely a discrete control case and a continuous control case. The results of these experiments show that our method outperforms several competitive baselines in the field of offline model-based optimization for industrial control.

A cGAN Ensemble-based Uncertainty-aware Surrogate Model for Offline Model-based Optimization in Industrial Control Problems

TL;DR

A novel cGAN ensemble-based uncertainty-aware surrogate model is introduced for reliable offline model-based optimization in industrial control problems and outperforms several competitive baselines in the field of offline model-based optimization for industrial control.

Abstract

This study focuses on two important problems related to applying offline model-based optimization to real-world industrial control problems. The first problem is how to create a reliable probabilistic model that accurately captures the dynamics present in noisy industrial data. The second problem is how to reliably optimize control parameters without actively collecting feedback from industrial systems. Specifically, we introduce a novel cGAN ensemble-based uncertainty-aware surrogate model for reliable offline model-based optimization in industrial control problems. The effectiveness of the proposed method is demonstrated through extensive experiments conducted on two representative cases, namely a discrete control case and a continuous control case. The results of these experiments show that our method outperforms several competitive baselines in the field of offline model-based optimization for industrial control.
Paper Structure (15 sections, 17 equations, 3 figures, 2 tables, 1 algorithm)

This paper contains 15 sections, 17 equations, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Offline model-based industrial control
  4. Generative adversarial networks
  5. Deep ensemble for uncertainty quantification
  6. Proposed surrogate Model
  7. cGAN ensemble-based probabilistic model
  8. Uncertainty-aware reward penalization
  9. Experiment on Discrete Control
  10. Quantitative evaluation
  11. Evaluation metrics and results
  12. Qualitative evaluation
  13. Experiment on Continuous Control
  14. Related Work
  15. Conclusion

Figures (3)

  • Figure 1: (a) the mean and standard deviation of the discrepancies between distributions of generated control results with different number of randomized dimensions for the inputs of the cGAN ensemble; (b) the mean and standard deviation of the amount of uncertainty for generated control results with different number of randomized dimensions for the inputs of the cGAN ensemble.
  • Figure 2: T-SNE 2D visualization for real and generated control result samples
  • Figure 3: (a,b,c): the distribution of $\kappa$, $\varkappa$ and $r_p$ for the logged inputs compared with the counterpart for the OoD inputs with only control parameter values are randomly generated; (d,e,f): the distribution of $\kappa$, $\varkappa$ and $r_p$ for the logged inputs compared with the counterpart for the OoD inputs whose conditional and control parameters are all randomly generated.