A New Strategy for Neural Emulator Learning Rate Tuning

In the present paper, a real time recurrent learning-based emulator is presented for nonlinear system. The neural emulator is developed with fully connected recurrent neural networks. The instantaneous neural emulator parameters adapt themselves to estimate dynamical behaviors using a starting term. An intuitive and bad choice of this term can affect the emulation performances. To overcome this problem, an adaptation of the emulator learning rate, independently of the starting term, is proposed in this work for nonlinear system. The effectiveness of the offered strategy is illustrated through numerical simulations in comparison with the method based on the starting term. The obtained results using the proposed strategy are far better than those obtained with the last one.