Real-time system identification of an unmanned quadcopter system using fully tuned radial basis function neural networks

In this paper, we present the performance analysis of a fully tuned neural network trained with the extended minimal resource allocating network (EMRAN) algorithm for real-time identification of a quadcopter. Radial basis function network (RBF) based on system identification can be utilised as an alternative technique for quadcopter modelling. To prevent the neurons and network parameters selection dilemma during trial and error approach, RBF with EMRAN training algorithm is proposed. This automatic tuning algorithm will implement the network growing and pruning method to add or eliminate neurons in the RBF. The EMRAN’s performance is compared with the minimal resource allocating network (MRAN) training for 1000 input-output pair untrained attitude data. The findings show that the EMRAN method generates a faster mean training time of roughly 4.16 ms for neuron size of up to 88 units compared to MRAN at 5.89 ms with a slight reduction in prediction accuracy