(374d) Development of Hybrid First Principles – Artificial Intelligence Models: Applications to an Industrial Steam Superheater System

First-principles models can provide very good prediction even for cases where there are no data at all, or data are limited in certain range of operating conditions or for cases where data collection is infeasible. However, first-principles physics-based models for complex nonlinear dynamic systems are often time consuming to construct, computationally expensive, and often intractable for online adaptation. It can also be difficult, if not impossible, to develop accurate models for some complex phenomena, that are poorly understood. On the contrary, artificial intelligence (AI) or data-driven models can be relatively easier to develop, simulate, and adapt online¹, even for complex and ill-defined dynamic systems. However, development of AI models requires large amount of data, and therefore can be infeasible where data acquisition is prohibitive or collection of certain type of data is practically impossible provided the current state of the measurement technology. Furthermore, AI models may not be predictive especially when they are extrapolated and/or if the data used for developing such data-driven models suffer from information gap. This work develops several approaches where the first-principles models are synergistically coupled with the machine learning models, thus exploiting their key strengths.

The integration of physics-based (first-principles) models with machine learning approaches have found numerous implementations in the form of lumped parameter models², residual modeling³, physics informed models⁴ and digital twin development⁵. These models can lead to improved interpretability and extrapolation capabilities⁶. However there are significant opportunities for exploiting the synergistic hybridization of first-principles models with data-driven models. Several novel coupling approaches are developed as part of this work including series, parallel, and cross-coupling of first-principles and data-driven models. For such hybrid models to be successful and yield desired outcomes, various other aspects need to be considered other than the coupling approaches. In particular, it is important to consider what information need to be exchanged between first-principles models and data-driven models and at what interval, how to select the specific data-driven model for the desired outcome, and how to adapt the hybrid model. The presence of nonlinear transients in both the first-principles and AI models makes it notably difficult for the optimal synthesis of such hybrid models with due consideration of complexity, computational expense, and accuracy. Furthermore, different AI models may have specific advantages and limitations, which become critical while addressing any particular phenomena pertaining to the operation of the superheater system under different load conditions.

The proposed algorithms have been developed for modeling boiler components where complex dynamics associated with reactive-diffusive processes leading to oxide scale formation in the superheater tube banks coupled with mass and heat transfer pose to be considerably challenging. Despite many studies on hybrid modeling⁵ and process monitoring⁷ of thermal power plants using AI models to estimate certain key parameters from historical operational data, accurate modeling of heat transfer dynamics of in-service power plants over the full working range still remains a challenge. A first-principles model of an industrial steam superheater system as part of a coal-fired power plant is developed. The first-principles model is a three-dimensional differential-algebraic equation (DAE) model with a detailed model of heat transfer through thick-walled tubes. Two types of data-driven models are developed, namely artificial neural networks (ANN) and Bayesian machine learning (BML) models. Hybrid series and parallel all-nonlinear dynamic-static neural networks are leveraged as ANN. These networks have been shown to yield superior performance compared to many of the existing state-of-the-art models⁸ like long short-term memory (LSTM) or gated recurrent unit (GRU) types of recurrent networks. Developed BML algorithms offer capabilities for handling noisy data and uncertainty quantification along with estimation of interpretable model parameters. The hybrid modeling approaches are used to address several challenges in modeling of industrial superheaters including time-varying uncertainties in fouling deposit due to coal ash, non-uniform distribution of flue gas and steam that is affected by the load and state of fouling deposit, lack of measurements especially for the flue gas, and uncertainties in the boundary conditions especially for the flue gas, etc. Overall, the proposed hybrid models are found to yield very accurate prediction of the transients of the superheater when tested with the industrial data.

References:

1. Su, H.-T., Bhat, N., Minderman, P. A. & McAvoy, T. J. Integrating Neural Networks With First Principles Models for Dynamic Modeling. Dynamics and Control of Chemical Reactors, Distillation Columns and Batch Processes (IFAC, 1993). doi:10.1016/b978-0-08-041711-0.50054-4.
2. Psichogios, D. C. & Ungar, L. H. A hybrid neural network‐first principles approach to process modeling. AIChE J. 38, 1499–1511 (1992).
3. Rajulapati, L., Chinta, S., Shyamala, B. & Rengaswamy, R. Integration of machine learning and first principles models. AIChE J. 68, 1–17 (2022).
4. Raissi, M., Perdikaris, P. & Karniadakis, G. E. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J. Comput. Phys. 378, 686–707 (2019).
5. Yu, J., Petersen, N., Liu, P., Li, Z. & Wirsum, M. Hybrid modelling and simulation of thermal systems of in-service power plants for digital twin development. Energy 260, 125088 (2022).
6. Bradley, W. et al. Perspectives on the integration between first-principles and data-driven modeling. Comput. Chem. Eng. 166, 107898 (2022).
7. Hamid, A., Hasan, A. H., Azhari, S. N., Harun, Z. & Putra, Z. A. Hybrid modelling for remote process monitoring and optimisation. Digit. Chem. Eng. 4, 100044 (2022).
8. Mukherjee, A. & Bhattacharyya, D. Hybrid Series/Parallel All-Nonlinear Dynamic-Static Neural Networks: Development, Training, and Application to Chemical Processes. Ind. Eng. Chem. Res. 62, 3221–3237 (2023).