(196b) Keynote Talk: Cyber-Secure Machine Learning Modeling and Predictive Control of Nonlinear Chemical Process Network Using Federated Learning

The security of process control systems has become crucially important since control systems are vulnerable to cyber-attacks, which are a series of computer actions employed by the attacker to compromise the security of control systems. As machine learning (ML) approaches have been widely used to develop data-driven models for nonlinear systems [1], cybersecurity and data privacy have garnered increasing research interests with the use of big data. For example, developing an ML-based detector for large-scale distributed systems requires a tremendous volume of data to be collected from all subsystems through various mediums of communication such as Internet and wireless networks, and then processed in a central server or cloud for training. However, as communication mediums are vulnerable to attackers, the ML-based detector developed on a local server or cloud could be misguided and unable to detect target cyber-attacks in the presence of data tampering or data manipulation [2, 3]. Therefore, further advances in techniques and frameworks for promoting data privacy are needed to provide tractable solutions for the implementation of ML modeling methods.

To alleviate the security concerns, a federated-learning-based model predictive control (FL-based MPC) method for nonlinear chemical process network is developed in this work. By taking advantage of the idea of federated learning that distributes a pre-trained model to all subsystems and allows each subsystem to develop and update its own model locally without sharing the raw data with the central server, we first develop an FL framework with fully-connected and partially-connected neural network structures to model the entire process network with enhanced data privacy while accounting for the heterogeneity of nonlinear systems with multiple subsystems (note that the heterogeneity of the subsystems will result in non-independent and identically distributed training data [5]). Subsequently, we develop a theoretical generalization error bound and the estimation of privacy leakage for the FL models using relative entropy and duality theory. The closed-loop stability of distributed systems under the FL-based MPC approach is further developed. Finally, a chemical reactor is used as an example to demonstrate the effectiveness of the proposed FL modeling and FL-based MPC approach.

[1] Wu, Z., Tran, A., Rincon, D., & Christofides, P. D. (2019). Machine learning‐based predictive control of nonlinear processes. Part I: theory. AIChE Journal, 65(11), e16729.

[2] Tan, A. Z., Yu, H., Cui L., & Yang, Q. (2022). Towards Personalized Federated Learning. IEEE Transactions on Neural Networks and Learning Systems, doi: 10.1109/TNNLS.2022.3160699.

[3] Chen, M., Yang, Z., Saad, W., Yin, C., Poor, H. V., Cui, S. (2021). A Joint Learning and Communications Framework for Federated Learning Over Wireless Networks. IEEE Transactions on Wireless Communications, 20(1), 269-283.

[4] Christofides, P. D., Scattolini, R., Pena, D. M., & Liu, J. (2013). Distributed model predictive control: A tutorial review and future research directions. Computers and Chemical Engineering, (51), 21-41.

[5] Sattler, F., Wiedemann, S., Müller, K. R., & Samek, W. (2020). Robust and Communication-Efficient Federated Learning From Non-i.i.d. Data. IEEE Transactions on Neural Networks and Learning Systems, 31(9), 3400-3413.