
Robust Online Switched Model Identification
In practice, influenced by the changes of external factors, time series data may display abrupt changes in its behaviors, such as stock prices vs regulatory policies, EEG signals vs emotions, human-made systems under malfunctions or battery draining. Given such streaming data, the goal is (i) to learn a model that can characterize its behaviors and (ii) identify the change of behaviors. The challenge comes from solving this in an online fashion.

Adaptive Control for Switched Systems
In this problem, we seek to adaptively control a switched system but with unknown dynamics in a model-based way. That is, the control scheme alternates between estimating system dynamics and designing controllers with the estimated dynamics. We show how the dynamics estimate accuracy and controller optimality improve over time.

System Mode Reduction for Switched Systems
Though switched models are capable of modeling abrupt changes, system complexity issues arise if the number of changes gets large. We utilize clustering techniques to reveal underlying lumped structure and use it to construct reduced models. We show the reduced model have dynamics approximates well the original one. It can also be used as a surrogate to perform analysis or controller design with significant computation cost savings.
Publications
- Du, Zhe, Zexiang Liu, Jack Weitze, and Necmiye Ozay. “Sample complexity analysis and self-regularization in identification of over-parameterized ARX models.” 2022 61th IEEE Conference on Decision and Control (CDC) (forthcoming). IEEE, 2022. [Paper], [Slides], [Poster]
- Du, Zhe, Laura Balzano, and Necmiye Ozay. “Mode reduction for Markov jump systems.” IEEE Open Journal of Control Systems (forthcoming), 2022. [Paper]
- Du, Zhe, Necmiye Ozay, and Laura Balzano. “Clustering-based mode reduction for Markov jump systems.” Learning for Dynamics and Control Conference. PMLR, 2022, pp. 689–701. [Paper], [Poster]
- Du, Zhe, Yahya Sattar, Davoud Ataee Tarzanagh, Laura Balzano, Necmiye Ozay, and Samet Oymak. “Identification and adaptive control of Markov jump systems: sample complexity and regret bounds.” arXiv preprint arXiv:arXiv:2111.07018. [Paper]
- Du, Zhe, Yahya Sattar, Davoud Ataee Tarzanagh, Laura Balzano, Necmiye Ozay, and Samet Oymak. “Data-driven control of Markov jump systems: sample complexity and regret bounds.” 2022 American Control Conference (ACC), 2022, pp. 4901-4908. [Paper], [Slides]
- Du, Zhe, Yahya Sattar, Davoud Ataee Tarzanagh, Laura Balzano, Samet Oymak, and Necmiye Ozay. “Certainty equivalent quadratic control for Markov jump systems.” 2022 American Control Conference (ACC), 2022, pp. 2871-2878. [Paper], [Slides]
- Du, Zhe, Yahya Sattar, Davoud Ataee Tarzanagh, Laura Balzano, Necmiye Ozay, and Samet Oymak. “Identification and adaptive control of Markov jump systems: sample complexity and regret bounds.” ICML Workshop on Reinforcement Learning Theory, 2021. [Paper], [Poster]
- Du, Zhe, Necmiye Ozay, and Laura Balzano. “Mode clustering for Markov jump systems.” 2019 IEEE 8th International Workshop on Computational Advances in Multi-Sensor Adaptive Processing (CAMSAP). IEEE, 2019. (Best Paper Award) [Paper], [Slides], [Poster]
- Du, Zhe, Laura Balzano, and Necmiye Ozay. “A robust algorithm for online switched system identification.” IFAC-PapersOnLine 51.15 (2018): 293-298. [Paper], [Slides]
- Ledva, Gregory S., Zhe Du, Laura Balzano, and Johanna L. Mathieu. “Disaggregating load by type from distribution system measurements in real time.” In Energy Markets and Responsive Grids, pp. 413-437. Springer, New York, NY, 2018. [Paper]