(204a) Data-Driven Distributed Moving Horizon Estimation for Large-Scale Industrial Processes

Modern industrial chemical processes have witnessed an increase in the complexity, scale, and level of integration. As a result, the conventional decision-making framework is not sufficient to meet the industrial demands for safe, flexible, reliable, and sustainable operations. This trend necessitates the development of partition-based distributed frameworks for large-scale chemical processes with highly interconnected physical units [1][2][3]. Recently, distributed state estimation has attracted a lot of attention due to its ability to provide full-state information for control, monitoring, and fault diagnosis. Within the partition-based distributed framework, a large-scale system is divided into several smaller subsystems, each of which serves as the basis for its corresponding local estimator design. Among various distributed state estimation approaches, distributed moving horizon estimation (MHE) is more favorable due to its ability to address the nonlinearity of underlying systems and handle the state constraints explicitly. However, most existing distributed MHE approaches rely on accurate first-principles models, which are hard to obtain [4][5]. Since industrial data becomes increasingly accessible, it is promising to develop data-driven distributed MHE to provide scalable and flexible state estimation for industrial chemical processes.

There are two promising data-driven frameworks for state estimation. The first framework leverages Koopman operator theory to build a linear representation for the nonlinear process directly from data. In [6], a data-driven Koopman model was developed for the nonlinear process, based on which a linear MHE design was formulated. This approach was further extended in [7] to a distributed framework to account for the large scale of nonlinear systems. Specifically, in [7], the Koopman model is identified for each subsystem in parallel, and a distributed MHE approach was formulated by leveraging the identified subsystem models. An alternative framework is to integrate MHE with Willems’ fundamental lemma, which represents system trajectories using historical system data. Different from the Koopman-based approach, this framework eliminates the need for system identification and is thus considered in our work. In [8], a robust data-driven MHE based on Willems’ fundamental lemma was proposed for linear systems for cases with noise-free offline data. This method was further extended in [9] to account for the measurement noises during the offline data collection. However, to the best of our knowledge, the data-driven distributed MHE with utilization of Willems’ fundamental lemma remains unexplored.

In this work, we focus on the state estimation of linear systems in a distributed framework. Specifically, a data-driven distributed MHE approach based on Willems’ fundamental lemma is proposed. First, we extend Willems’ fundamental lemma to a distributed framework, based on which a data-driven distributed MHE method is proposed. Subsequently, we establish the stability conditions for the proposed approach. Finally, the proposed distributed MHE is applied to some chemical processes to demonstrate its effectiveness.

References:
[1] P. D. Christofides, R. Scattolini, D. M. de la Peña, and J. Liu. Distributed model predictive control: A tutorial review and future research directions. Computers and Chemical Engineering, 51:21–41, 2013.
[2] P. Daoutidis, M. Zachar, and S. S. Jogwar. Sustainability and process control: A survey and perspective. Journal of Process Control, 44:184–206, 2016.
[3] P. Daoutidis, J. H. Lee, I. Harjunkoski, S. Skogestad, M. Baldea, and C. Georgakis. Integrating operations and control: A perspective and roadmap for future research. Computers & Chemical Engineering, 115:179–184, 2018.
[4] R. Schneider, R. Hannemann-Tamás, and W. Marquardt. An iterative partition-based moving horizon estimator with coupled inequality constraints. Automatica, 61:302–307, 2015.
[5] R. Schneider and W. Marquardt. Convergence and stability of a constrained partition-based moving horizon estimator. IEEE Transactions on Automatic Control, 61(5):1316–1321, 2015.
[6] X. Yin, Y. Qin, J. Liu, and B. Huang. Data-driven moving horizon state estimation of nonlinear processes using Koopman operator. Chemical Engineering Research and Design, 200:481–492, 2023.
[7] X. Li, S. Bo, X. Zhang, Y. Qin, and X. Yin. Data‐driven parallel Koopman subsystem modeling and distributed moving horizon state estimation for large‐scale nonlinear processes. AIChE Journal, 70(3):e18326, 2024.
[8] T. M. Wolff, V. G. Lopez, and M. A. Müller. Data-based moving horizon estimation for linear discrete-time systems. European Control Conference, 1778–1783, 2022.
[9] T. M. Wolff, V. G. Lopez, and M. A. Müller. Robust data-driven moving horizon estimation for linear discrete-time systems. IEEE Transactions on Automatic Control, 69(8):5598–5604, 2024.