Deep Learning based Predictive Control of Future Naval Power and Energy Systems
Abstract
Motivated by challenges in real-time control of Naval power and energy systems and recent theoretical advances in the field of deeplearning, this research proposes to learn the optimal control policy defined by a complex model predictive formulation using deep neural networks. Two designs are introduced to 1) learn the dynamic model of Naval power and energy systems from data using a physics-inspired deep learning framework and utilize the obtained data-driven model for predictive control of Naval ships in real-time, 2) train a deep learning-based optimal controller that learns the solution of model predictive control in an offline process, so that the online use of the learned controller requires only the evaluation of a neural network. The proposed designs can be executed very rapidly on embedded hardware to solve the existing computational complexities of optimal control frameworks. We aim to explore the potential of the proposed approach on a hardware-in-the-loop setup of a navalpower and energy system composed of auxiliary generators, hybrid storage units, pulsed power loads, propulsion motors, and service loads. Using advances in deep learning theories, we develop a theoretical framework for analyzing the stability and optimality of proposed deep learning designs using eigen decomposition technique for each layer of deep learning and further enhance the computational efficiency of the proposed designs. Experimental validation of the proposed methods will be conducted in an existing 10kW microgrid testbed at Lehigh University with hardware-in-the-loop(HIL) tests.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 24, 2023
- Source ID
- N000142312602
Entities
People
- Javad Khazaei
Organizations
- Lehigh University
- Office of Naval Research
- United States Navy