Hybrid electric vehicles powered by fuel cells and batteries have attracted significant attention as they have the potential to eliminate emissions from the transport sector. However, fuel cells and batteries have several operational challenges, which require a power and energy management system (PEMS) to achieve optimal performance. Most of the existing PEMS methods are based on either predefined rules or prediction that are not adaptive to real-time driving conditions and may give solutions that are far from the actual optimal solution for a new drive cycle. Therefore, in this paper, an intelligent PEMS using reinforcement learning is presented, that can autonomously learn the optimal policy in real time through interaction with the onboard hybrid power system. This PEMS is implemented and tested on the simulation model of the onboard hybrid power system. The propulsion load is represented by the new European drive cycle. The results indicate that the PEMS algorithm is able to improve the lifetime of batteries and efficiency of the power system through minimizing the variation of the state of charge of battery.

No abstract

No abstract

The electrification of a ship power-train is growing at a fast pace to improve efficiency and reduce emissions. The implementation of new technologies requires test and validation using various modeling approaches. However, many of the existing models of the ship hybrid power system are too complicated and demand high computational requirements, which make them inappropriate for the real-time applications. The realtime simulation model offers the benefits of testing different control algorithms along with hardware-in-the-loop testing. The bond graph-based dynamic modeling of a ship hybrid power system with a DC grid is presented as applicable to realtime simulation. The overall system model is established using different component models with varying fidelity, so-called mixedmodeling approach. In this approach, the components and control functions are modeled with different complexity such that it can capture the necessary system dynamics while minimizing the computational time. Results show that the modeled system is capable of simulating different operating strategies of the hybrid power system. Moreover, the mixed-modeling approach has enabled the system to simulate in nearly 2.5 times faster than the real-time.