Online identification of complex multi-input-multi-output system based on generic evolving neuro-fuzzy inference system

Nowadays, unmanned aerial vehicles (UAV) play a noteworthy role in miscellaneous defence and civilian operation. A major facet in the UAV control system is an identification phase feeding the valid and up-to-date information of the system dynamic in order to generate proper adaptive control action to handle various UAV maneuvers. UAV, however, constitutes a complex system possessing a highly non-linear property. Conversely, the learning environment in modeling UAV's dynamic varies overtime and demands online learning scheme encouraging a fully adaptive and evolving algorithm with a mild computational load to settle the task. In contrast, contemporaneous literatures scrutinizing the identification of UAV dynamic yet rely on offline or batched learning procedures. Evolving neuro-fuzzy system (ENFS) where the landmarks are flexible rule base and usable in the time-critical applications offers a promising impetus in the UAV research field, and in particular its identification standpoint. The principle cornerstone is ENFS can commence its learning mechanism from scratch with an empty rule base and very limited expert knowledge. Nonetheless, it can perform automatic knowledge building from streaming data without catastrophic forgetting previous valid knowledge which is alike autonomous mental development of human brain. This paper elaborates the identification of rotary wing UAV based on our incipient ENFS algorithm termed generic evolving neuro-fuzzy system (GENEFIS). In summary, our algorithm can not only trace footprint of the UAV dynamic but also ameliorate the performance of existing ENFS in terms of predictive quality and resultant rule base burden.