There are three ways in which you can train a neural network online in Simulink for system identification using the NARX model:
- MATLAB's System Identification Toolbox allows online training of nonlinear models, including NARX networks. You can use the Nonlinear ARX model with adaptive learning to update parameters dynamically. Refer the following documentation: https://www.mathworks.com/help/ident/nonlinear-model-identification.html
- Simulink provides adaptive learning blocks that allow online training. You can use Neural State-Space Models or Adaptive Neuro-Fuzzy Inference Systems (ANFIS) for real-time learning. You can read more about it here: https://www.mathworks.com/help/deeplearning/ug/design-time-series-narx-feedback-neural-networks.html
- Finally, you can implement an online learning algorithm using MATLAB Function blocks inside Simulink. This requires manually updating weights using backpropagation or reinforcement learning techniques.
