Back to the program.
Application of local learning based algorithms for real time estimation of power transformer loading capability
Alfredo Vaccaro
Power System Research Laboratory;; University of Sannio;; Benevento (ITALY)
On 2002-09-26 at 15:00:00 (Brussels Time)

Abstract

An accurate real time estimation of power transformer loading capability is of critical importance in power systems since they can cause widespread power outages of the distribution power systems, especially in presence of different utilities as in to a deregulated market of electricity. At the present, the loadability of a transformer is calculated by applying transient heating equations using the transformer's specific thermal characteristics and making some simplifying assumptions. Since this method is based on simplified thermal equivalent models and requires some specific transformer data, which can vary considerably from one transformer to another, it becomes susceptible to parameter variations that affect the accuracy of the loading calculations. All this has led studies to investigate the possibility of adopting non linear regression technique jointly with soft computing methodologies for the estimation of the power transformer thermal dynamics. In agreement with these argumentation during the talk the application of a local learning based algorithm for real time estimation of power transformer loading capability is discussed. The major benefit of this approach is that it does not require a preliminary training phase since the value of the unknown mapping in a desired point is predicted by the combination of local approximators identified from the neighbouring examples of the point which are considerate relevant according to a distance measure. Moreover, since the experimental samples used in the prediction are stored in a memory, it is straightforward to update them in order to manage time varying phenomena. The experimental studies developed have evidenced as the proposed methodology appear to be very fast and effective compared with classical techniques.

Keywords

local learning, real time thermal state estimation, power components monitoring, system identification

References

  1. V. Galdi, L. Ippolito, A. Piccolo, A. Vaccaro. (2001) Application of Local Learning Techniques to power transformer thermal overload protection, IEE Proceeding - Electric Power Application, 148(2).
  2. V.Galdi, L.Ippolito, A.Piccolo, A.Vaccaro. (2000) Neural diagnostic system for transformer thermal overload protection, IEE proceeding Electric Power Application, 147(5).