Skip to content

You are here: Home / Results / Deliverable 1.2 (June 2019)

Deliverable 1.2 (June 2019)

Refined Identification methods applicable to fit electrical models

Different filtering strategies were applied to the experimental signals in order to improve the performance of the parameter identification algorithms. Such filtering algorithms include: resizing, moving average and piece-wise linear function, low-pass/band-pass filter and linear regression.

White-box identification approach is based on solving the differential equations governing the behavior of the analyzed devices. In case of the DC/DC converters, the parameters are identified based on steady state operation and a load change (transient operation). It has been shown that a very fast identification of the parameters is possible, since the identification time is less than 1 second.

Another white-box approach is also presented. It consists on converting the problem into an optimization problem. The NLS (nonlinear least squares) optimization algorithm has been used to identify the parameters of the analyzed devices. In case of the DC/DC converters and the three-phase rectifier, the parameters are identified based on steady state operation and a load change (transient operation). In the case of the EMI filter, an input square wave voltage jointly with a capacitive load allow determining the parameters. This approach does not require to solve the differential equations governing the behavior of the analyzed devices, however it requires a higher computation load, which can be between several minutes and some hours do determine the values of the parameters.

Furthermore, this deliverable has developed a black-box based approach to identify the parameters of a Buck DC/DC converter based on simulated data. It has been shown that without any knowledge of the model structure, it is possible to replicate the behavior of a buck converter by means of a black-box approach, when a rich set of data is available. The parameter identification strategy is based on a NARX neural network.

A setup for automatically acquiring the experimental input/output voltages and currents of the analyzed devices has been developed by the MCIA-UPC team.