Nirlipta Mohanty
|
Position:
Department: Lab: eMail: Homepage: |
PhD Researcher
Instrumentation and Control Embedded Systems Lab [email protected] http://www.coepembeddedlab.com/mohanty |
Biography
Nirlipta Ranjan Mohanty is a full time research scholar at Department of Instrumentation and Control, College of Engineering, Pune. He received his B.Tech degree in Applied Electronics and Instrumentation engineering and M.Tech degree in Instrumentation & Control engineering. His research interests include process control, machine learning, embedded control and model predictive control.
Personal
Research Interests
- Stochastic Model Predictive Control
- Probability & Statistics
PhD Thesis
Learning Based Model Predictive Control
Abstract:
Model predictive control (MPC) is an advanced control strategy that predicts the future behavior of the plant taking into account constraints and uncertainty. The challenges in implementation of MPC in most actual systems are its computational complexity and memory constraints as it solves an online optimization problem at each sample time. Recent advancement in the field of machine learning is an encouraging remedy for these problems. By using machine learning algorithms to approximate, MPC online optimization is completely eliminated that scale down the computational burden and memory requirement. Machining learning based plant model can also be used while implementing MPC where mathematical model of the plant is not available or it is difficult to obtain. Closed-loop performance evaluation of the learning based MPC will be verified on different embedded platform.
Abstract:
Model predictive control (MPC) is an advanced control strategy that predicts the future behavior of the plant taking into account constraints and uncertainty. The challenges in implementation of MPC in most actual systems are its computational complexity and memory constraints as it solves an online optimization problem at each sample time. Recent advancement in the field of machine learning is an encouraging remedy for these problems. By using machine learning algorithms to approximate, MPC online optimization is completely eliminated that scale down the computational burden and memory requirement. Machining learning based plant model can also be used while implementing MPC where mathematical model of the plant is not available or it is difficult to obtain. Closed-loop performance evaluation of the learning based MPC will be verified on different embedded platform.
Publications
