432632 Parameter Estimation and Nonlinear Model Predictive Control Using Meshless Modelling Framework

Monday, November 9, 2015: 4:31 PM
Salon F (Salt Lake Marriott Downtown at City Creek)
Vivek Dua, Department of Chemical Engineering, University College London, London, United Kingdom and Abiodun Owolaja, Chemical Engineering, University College London, London, United Kingdom

Nonlinearities and complex dynamics are inherent to chemical process systems. Safe and stable operation of these processes – while maintaining profitability – in the presence of process constraints is essential to meet economic and environmental criteria. Model based control plays an important role in realizing this objective and has contributed to improved process operations. Nonlinear Model Predictive Control (NLMPC) is particularly of great interest to chemical process engineers. This is due to the ability of this controller to explicitly handle constraints, consider multivariable process interactions and intrinsically manage nonlinearities in process systems [1]. The success of NLMPC is based on accurate modelling and representation of the process. NLMPC controllers which utilize fundamental models are attractive as they can be extrapolated for a wide range of data – unlike empirical models. In this work, a new paradigm for solving the fundamental models in NLMPC controllers is presented. Artificial Neural Networks (ANN) are utilized as a meshless modelling platform to provide solution to the Ordinary Differential Equation (ODE) models of dynamic process systems [2]. Traditional numerical methods have limited differentiability, experience a significant increase in complexity with increasing number of sampling points and may be challenging to use for higher dimension problems. These limitations can be overcome by utilizing the ANN integrator to provide a solution to the ODEs. The set of system ODEs is transformed into a set of Algebraic Equations (AE) and posed as a nonlinear programming (NLP) optimization problem. The NLMPC framework, where the ODEs are integrated using the ANN, is applied to setpoint tracking, regulatory and zone control [3] studies. In process control, adequate information about the system is vital. Operational transitions could lead to changes in system parameters; and knowledge of these parameters is crucial to the operation and control of these processes. Utilizing the ANN formulation, a parameter estimation problem is formulated and solved as an NLP [2]. This offers a solution approach in which available online measurements can be utilized to compute the process parameters while simultaneously providing a solution to the system of ODEs representing the process. Case studies for parameter estimation and NLMPC are presented to demonstrate the applicability of the proposed approach.


[1]      V. Dua, “Stability Analysis of Nonlinear Model Predictive Control : An Optimization Based Approach,” no. 1995, pp. 1–6, 2005.

[2]      V. Dua and P. Dua, “A Simultaneous Approach for Parameter Estimation of a System of Ordinary Differential Equations, Using Artificial Neural Network Approximation,” Ind. Eng. Chem. Res., vol. 51, no. 4, pp. 1809–1814, Feb. 2012.

[3]      B. Grosman, E. Dassau, H. C. Zisser, L. Jovanovic, and F. J. Doyle, “Zone model predictive control: a strategy to minimize hyper- and hypoglycemic events.,” J. Diabetes Sci. Technol., vol. 4, no. 4, pp. 961–75, Jul. 2010.

Extended Abstract: File Not Uploaded