dc.description.abstracten |
Parameter identification of Dynamical systems examined in the context of its value
for the DNN training process facilitation and mitigation of the data imbalance. In
the result of the theoretical analysis of the stated problem’s methodological roots,
were proposed the approach of augmenting usual NN model with part responsible
for explicit representation of required parameters. Moreover, this approach involves
the change of a learning procedure towards indirect supervision setting, where the
NN responsible for system modeling, in form of time series prediction, does not ob-
serves the dataset at all but is being taught via the intermediate step of identification
of target system parameter knowing its physical interpretation. Work considered
the dynamical process on the example of DC motor described by the system of the
ordinary nonlinear differential equations. |
uk |