This paper proposes decision trees (DTs) and wavelet analysis for the protection of large transformers. The DTs assimilate a large number of measured and derived quantities such as wavelet decomposition and harmonic coefficients. Wavelet decomposition is performed on 32 samples from a one-cycle window. The DT input vector contains eight wavelet coefficients from one level of resolution and other variables such as RMS differential current, RMS restraining current, percent differential current, and second and fifth harmonics. The eight detail coefficients are sorted so that the DT input vector remains relatively constant for a quasi-periodic signal as the observation window moves along the time axis. DTs are trained on data extracted from a large number of simulations. Accuracies around 95% are obtained in the presence of CT saturation when using either wavelets or harmonics in addition to the differential and restraining currents.
[1]
C. Burrus,et al.
Introduction to Wavelets and Wavelet Transforms: A Primer
,
1997
.
[2]
C. W. Taylor,et al.
Decision trees using apparent resistance to detect impending loss of synchronism
,
2000
.
[3]
S. R. Shaw,et al.
Transient event detection in spectral envelope estimates for nonintrusive load monitoring
,
1995
.
[4]
D. W. Nicoletti,et al.
A wavelet-based differential transformer protection
,
1999
.
[5]
David Chan Tat Wai,et al.
A novel technique for high impedance fault identification
,
1998
.
[6]
M. Kezunovic,et al.
Digital relays improve protection of large transformers
,
1998
.