Multi-agent intelligence with privacy awareness for robust marine robotic electric drives

Abstract

This paper introduces a privacy-preserving multi-agent intelligent model tailored towards improving the movement control and energy management of robotic marine electric motors. In particular, the research incorporates the use of a fractional-order neural network algorithm (FONNA) combined with federated learning, allowing for decentralization without compromising raw data security. The modeling process entails the application of a seven-phase voltage source inverter (VSI) to offer high-performance motor behavior and exceptional fault resilience. Some of the significant discoveries include the delivery of 96.96% peak efficiency at 55 kHz and decreasing dependency on high-bandwidth communication by 75% via gradient updates. Energy resilience remains an essential prerequisite for collaborative marine robotics in harsh offshore conditions where the electricity supply can be subjected to disruption. However, even with all these developments, robotic fleets for maritime purposes are faced with some major constraints when working together. For instance, the centralized approach is easily affected by system failures since all operations will be affected if one fails. The marine area poses harsh conditions in which communication bandwidth and electrical power may be difficult to provide.