Physics-informed machine learning modeling for predictive control using noisy data

Alhajeri, Mohammed S.; Abdullah, Fahim; Wu, Zhe; Christofides, Panagiotis D.

doi:10.1016/j.cherd.2022.07.035

Cited by 33 publications

(10 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The above equation does not hold around the neighborhood of saddle points ℬ δ x e ð Þ since Equations ( 21) and (38) are not valid there.…”

Section: Sample-and-hold Implementationmentioning

confidence: 99%

“…36 Additionally, recurrent neural networks (RNNs) have also been integrated with first-principles knowledge to build hybrid model-based MPCs 37 and the effect of noise on these model's performance has also been studied. 38 But in these works, the concept of DA of a machine learning model has not been explicitly incorporated in the MPC design. To incorporate the DHM's DA within the LMPC controller, we propose to develop and incorporate within the MPC design a Control Barrier Function (CBF) that has prominence in the field of safety.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Deep hybrid model‐based predictive control with guarantees on domain of applicability

Bangi

Kwon

2023

AIChE Journal

View full text Add to dashboard Cite

A hybrid model integrates a first‐principles model with a data‐driven model which predicts certain unknown dynamics of the process, resulting in higher accuracy than first‐principles model. Additionally, a hybrid model has better extrapolation capabilities compared with a data‐based model, which is useful for process control and optimization purposes. Nonetheless, the domain of applicability (DA) of a hybrid model is finite and should be taken into account when developing a hybrid model‐based predictive controller in order to maximize its performance. To this end, a Control Lyapunov–Barrier Function‐based model predictive controller (CLBF‐based MPC) is developed which utilizes a deep hybrid model (DHM), that is, a deep neural network (DNN) combined with a first‐principles model. Additionally, theoretical guarantees are provided on stability as well as on system states to stay within the DA of the DHM. The efficacy of the proposed control framework is demonstrated on a continuous stirred tank reactor.

show abstract

“…The above equation does not hold around the neighborhood of saddle points ℬ δ x e ð Þ since Equations ( 21) and (38) are not valid there.…”

Section: Sample-and-hold Implementationmentioning

confidence: 99%

mentioning

confidence: 99%

Deep hybrid model‐based predictive control with guarantees on domain of applicability

Bangi

Kwon

2023

AIChE Journal

View full text Add to dashboard Cite

show abstract

“…In ref process structural knowledge such as the relationships between the process input and output variables was used to design partially connected NN models to improve the generalization performance and reduce training time. The PI technique therefore effectively enforces the physical principles or the governing equations into the RNN architecture, and is uniquely advantageous in enhancing the trainability and generalizability of neural networks, especially in the small data regime, which is characterized by data scarcity or insufficiently sampled training data. − This distinctive integration of expert domain knowledge in terms of the underlying physical laws and the process dynamics encapsulated in the available data informatively steers the training process toward the learning of mechanistically and observationally consistent solutions, and is applicable for solving both the forward and inverse problems. Recently, PI technology has been applied in various engineering problems (e.g., thermodynamics and fluid mechanics).…”

Section: Introductionmentioning

confidence: 99%

Physics-Informed Online Machine Learning and Predictive Control of Nonlinear Processes with Parameter Uncertainty

Zheng

2023

Ind. Eng. Chem. Res.

Self Cite

View full text Add to dashboard Cite

In this work, we present a physics-informed recurrent neural network (PIRNN)-based modeling approach for nonlinear dynamic systems with parameter uncertainty. Physics-informed modeling approaches can improve the generalization performance of machine learning models by embedding the knowledge of physical laws in the learning process. Based on the standard PIRNN modeling approach for the nominal system without model uncertainties, we develop a novel PIRNN-enhanced modeling method that integrates online estimation of uncertain process parameters into the training process using the latest process data. The PIRNN-enhanced modeling approach is incorporated into the design of model predictive control (MPC), where an error-triggered mechanism is employed to trigger the quantification of modeling uncertainties and update of PIRNN models accordingly. Finally, a chemical process example is used to demonstrate the superiority of the proposed PIRNN-enhanced online learning mechanism in comparison to the conventional purely data-driven online update strategy for nonlinear systems subject to process parameter uncertainty.

show abstract

“…The use of neural network-based MPCs is well-established in chemical systems. − The prevalent type of neural network architecture that most chemical engineering applications use for predictive control scheme with time-series data are such as the recurrent neural network (RNN) and NARX, which is a type of RNN . In one of the recently reported studies, Alhajeri et al utilized a partially connected RNN model that is established with process-structure knowledge to incorporate it in a Lyapunov-based MPC (LMPC) .…”

Section: Introductionmentioning

confidence: 99%

Control of the Multi-Timescale Process Using Multiple Timescale Recurrent Neural Network-Based Model Predictive Control

Jian

Zabiri

Ramasamy

2023

Ind. Eng. Chem. Res.

View full text Add to dashboard Cite

This study attempts to offer an alternative to the problem of implementing model predictive controllers (MPC) in conditions where the timescale multiplicity of the process model is not accounted for when incorporated into the MPC. Modeling methods that do not account for the timescale multiplicity in system’s dynamics tend to become ill-conditioned and stiff when inversed in model-based controllers, thus requiring high computational loads to solve the equations. Therefore, this study proposes an alternative approach to the control of multi-timescale processes based on the use of multiple timescale recurrent neural network (MTRNN)-based neural network predictive controllers (NNPC). The effectiveness in handling setpoint tracking scenarios by the proposed method is evaluated using a benchmark nonexplicit two-timescale continuous stirred tank reactor (CSTR). After undergoing controller parameter optimization, the optimum configuration is found to be at 110, 37, and 0.2 for the cost horizon, control horizon, and control weighting factor, respectively. Results show that the MTRNN-based NNPC is able to track the reference trajectory with stable response and minimal error with a root mean square error of 0.0642. The optimized MTRNN-based controller is tested for its robustness under plant-model mismatch and is compared for its setpoint tracking abilities with a nonlinear autoregressive exogeneous (NARX)-based NNPC which showed that the proposed controller can satisfy the desired setpoint, resulting in an error that is 1.8 times lower than NARX-based NNPC.

show abstract

Physics-informed machine learning modeling for predictive control using noisy data

Cited by 33 publications

References 22 publications

Deep hybrid model‐based predictive control with guarantees on domain of applicability

Deep hybrid model‐based predictive control with guarantees on domain of applicability

Physics-Informed Online Machine Learning and Predictive Control of Nonlinear Processes with Parameter Uncertainty

Control of the Multi-Timescale Process Using Multiple Timescale Recurrent Neural Network-Based Model Predictive Control

Contact Info

Product

Resources

About