Self‐tuning moving horizon estimation of nonlinear systems via physics‐informed machine learning Koopman modeling

AIChE Journal, Год журнала: 2024, Номер 71(2)

Опубликована: Ноя. 22, 2024

Abstract In this article, we propose a physics‐informed learning‐based Koopman modeling approach and present Koopman‐based self‐tuning moving horizon estimation design for class of nonlinear systems. Specifically, train operators two neural networks—the state lifting network the noise characterization network—using both data available physical information. The first accounts functions model, while second characterizes system distributions. Accordingly, stochastic linear model is established in lifted space to forecast dynamic behaviors system. Based on (MHE) scheme developed. weighting matrices MHE are updated using pretrained at each sampling instant. proposed computationally efficient, as only convex optimization needs be solved during online implementation, updating does not require re‐training networks. We verify effectiveness evaluate performance method via application simulated chemical process.

Язык: Английский

---

Deep DeePC: Data‐enabled predictive control with low or no online optimization using deep learning

AIChE Journal, Год журнала: 2024, Номер 71(3)

Опубликована: Дек. 11, 2024

Abstract Data‐enabled predictive control (DeePC) is a data‐driven algorithm that utilizes data matrices to form non‐parametric representation of the underlying system, predicting future behaviors and generating optimal actions. DeePC typically requires solving an online optimization problem, complexity which heavily influenced by amount used, potentially leading expensive computation. In this article, we leverage deep learning propose highly computationally efficient approach for general nonlinear processes, referred as Deep DeePC. Specifically, neural network employed learn vector operator, essential component This trained offline using historical open‐loop input output process. With network, framework formed implementation. At each sampling instant, directly outputs eliminating need conventional The action obtained based on operator updated network. To address constrained scenarios, constraint handling scheme further proposed integrated with handle hard constraints during efficacy superiority are demonstrated two benchmark process examples.

Язык: Английский

---

Resilient distributed model predictive control of state‐delayed linear parameter varying systems with quantitative communication against denial of service attacks

Asian Journal of Control, Год журнала: 2024, Номер unknown

Опубликована: Июль 7, 2024

Abstract This work presents a resilient distributed model predictive control (MPC) method for linear parameter varying (LPV) systems with state delays and attacks in communication networks. Coordinations are required MPC (DMPC) to achieve the global performance of centralized (CMPC). However, can be severely degraded by unreliable networks, example, denial service (DoS) attacks. A framework is derived address communications DMPC. system divided into subsystems purpose. To deal uncertainties delays, “min‐max” DMPC algorithm presented buffer ensure resilience against DoS quantization scheme introduced quantize information exchanged between subsystems. An iterative interaction proposed exchange feedback laws among The stability closed‐loop under ensured using Lyapunov function method. effectiveness demonstrated through two simulation examples.

Язык: Английский

---

A modular transfer learning approach for complex chemical process network modeling

Chemical Engineering Science, Год журнала: 2024, Номер unknown, С. 121087 - 121087

Опубликована: Дек. 1, 2024

Язык: Английский

---