A comparative study of model approximation methods applied to economicMPC

Abstract

Economic model predictive control (EMPC) has attracted significant attention in recent years and is recognized as a promising advanced process control method for next‐generation smart manufacturing. It has the potential to not only improve economic performance but also significantly increase computational complexity. Model approximation has been a standard approach for reducing computational complexity in process control. In this work, we perform a study on three types of representative model approximation methods applied to EMPC, including model reduction based on available first‐principle models (e.g., proper orthogonal decomposition), system identification based on input–output data (e.g., subspace identification) that results in an explicitly expressed mathematical model, and neural networks based on input–output data. A representative algorithm from each model approximation method is considered. Two processes that are very different in dynamic nature and complexity were selected as benchmark processes for computational complexity and economic performance comparison, namely, an alkylation process and a wastewater treatment plant. The strengths and drawbacks of each method are summarized according to the simulation results, with future research direction regarding control‐oriented model approximation proposed at the end.

Model predictive control of agro‐hydrological systems based on a two‐layer neural network modeling framework

“…A multi‐level pseudorandom signal is similar to a pseudorandom binary signal, except that it has multiple levels, which helps to stimulate the nonlinear dynamics of the agro‐hydrological system. More details regarding the design of multi‐level pseudorandom inputs can be found in our previous work 22 . Figure 3 presents a segment of the training input signal and the corresponding system output for $$$ {M}_2 $$$.…”

“…TensorFlow 27 is used to train the NN models. Readers may refer to our previous work 22 for more details regarding data pre‐processing and NN training.…”

Model predictive control of agro-hydrological systems based on a two-layer neural network modeling framework

“…A multi-level pseudorandom signal is similar to a pseudorandom binary signal, except that it has multiple levels, which helps to stimulate the nonlinear dynamics of the agro-hydrological system. More details regarding the design of multilevel pseudorandom inputs can be found in our previous work [18]. Figure 3 presents a segment of the training input signal and the corresponding system output for M 2 .…”

“…TensorFlow [23] is used to train the NN models. Readers may refer to our previous work [18] for more details regarding data pre-processing and NN training.…”

“…Model approximation based on neural network (NN) is a more generalized approach compared to traditional model identification. Motivated by the accuracy and computational efficiency of the NN model reduction observed in our previous work [18] and inspired by the structure of the LPV model developed in [8], we propose a two-layer NN framework to approximate the agro-…”