This project is specifically aimed at model identification applied to chemical plants. Here, we present a simplified 'hybrid' modelling approach using neural networks to represent the difficult-to-model parts in the first-principles implementation. The methodology is inspired by the work of Kumar and Rawlings (2023), and we further expand the concept with quantile regression to make the model selectively learn a certain quantile of the data. This approach enhances the model's ability for uncertainty prediction, providing a more robust framework for chemical plant modeling.
- Incentive for Deep Learning (Specifically Hybrid Modelling)
- Case Study for Partial State Measurement
- Towards a Structured 'Greybox' Model
- Quantile Regression for Uncertainty Prediction
- References
The incentive behind adopting a deep learning approach, particularly hybrid modelling, for chemical plants is to...
In this section, we present a case study focusing on partial state measurement...
This section delves into the development of a structured 'greybox' model...
Quantile regression is employed to enable the model to selectively learn a specific quantile of the data, facilitating...
- P. Kumar and J. B. Rawlings. "Structured nonlinear process modeling using neural networks and application to economic optimization." Comput. Chem. Eng., 177, 2023. doi: https://doi.org/10.1016/j.compchemeng.2023.108314.