Modular Energy Management Via Distributed and Predictive Optimization for Fuel Cell-Battery Shipboard Microgrids

Abstract

Electrification of ship power systems plays a central role in the mobility transition towards sustainable transportation. The integration of a large number of components with distinct characteristics into a shipboard microgrid benefits from a modular design and standardized interfaces. Key challenges lie in the variety of component characteristics, and an evolution of parameters during the power system operation. Further, topology alterations can occur over time, requiring a reformulation of the optimal power dispatch problem. Accordingly, a modular energy management strategy must be adaptive to these changes. This work explores a distributed energy management architecture with a central coordinating agent, realized via Lagrangian dual decomposition and a gradient-based solver. This architecture ensures both local feasibility while reaching global optimality and a power balance through a consensus mechanism. Parameter changes are incorporated in local cost functions, making extensive data exchange with a central unit obsolete. Handling a variable number of power system components, this approach is resilient to component faults, topology re-designs, and component degradation. The method is applied to a fuel-cell battery hybrid harbor tug equipped with multiple parallel modules with unique ratings and state-of-health. The energy management strategy minimizes total operating costs, based on hydrogen fuel consumption and cell degradation. Extensive mission simulations show similar performance for the distributed approach and a centralized equivalent. The predictive strategy is demonstrably superior to instantaneous optimization, yielding a cost reduction of 18.3% with a 15min prediction horizon. The model predictive control (MPC) performance increases with the horizon length, reducing operation costs by an additional 6.0% at 60min. In addition, a local decision-making heuristic shows promising potential for the cell degradation via optimized timing of on- and off switching. At 15min, this reduces operation costs by 3.0% and at 60min by 12.7%. Finally, the distributed optimization is deployed on real-time target machines to showcase the applicability of the approach on actual controller and communication hardware.