What do you mean by "approximation of dynamical system"? The names which you have mentioned are kind of learning functions. These will help you storing the memory.
in fact, I want to approximate the solution of PDE by NN, but inherently I want to know which kinds of neural network can be a better approximated model for dynamic systems.
The type of network to use depends on the type of problem to be used: continuous or discrete dynamic system, linear or non-linear. Could you send me your email to send you some information? .