In this simulation I implemented an RBF-NN for the zero order approximation of a nonlinear system. The simulation includes Monte Carlo simulation setup and the RBF NN code. For system estimation Gaussian kernels with fixed centers and spread are used. Whereas, the weights and the bias of the RBF-NN are optimized using the gradient descent-based adaptive learning algorithm.
Khan, S., Naseem, I., Togneri, R. et al. Circuits Syst Signal Process (2017) 36: 1639. doi:10.1007/s00034-016-0375-7
shahin darvish, you can not identify impulse response [*h* vector] using RBF. To identify impulse response you must choose a model with similar dynamics. e.g., if you know your system is linear then try to model with LMS or if it it second order non linear system then try 2nd order expansion of Volterra etc. similarly for feedback (IIR) system you can try bilinear model etc. However, with MLP/RBF or model NN based model you can not identify the impulse response but you can mimic any unknown system with reasonable accuracy.
how can I identify h vector?
Dear @hajar arjmandi, you can download my multi-state RBF code from https://kr.mathworks.com/matlabcentral/fileexchange/68415-nonlinear-system-identification-using-multi-state-rbf
I want to identify a nonlinear system with 2 state dynamic by RBF neural network, I downloaded your m file in Mathwork website which is about Nonlinear System Identification using RBF Neural Network.. Here system has 1state dynamic , how I can change when system is 2state?
@ridha rehouma, Please see  and  may be you can find a useful example.
i want to use neural network for identification harmonics currents,the active power filter simulte in mdl how i can generate the block neural netowrk in simulink