Online monitoring of temperature in power transformers using optimal linear combination of ANNs

Inordinate temperature rise in a power transformer due to load current is known to be the most important factor in causing rapid degradation of its insulation and decides the optimum load catering ability or the loadability of a transformer. The top oil temperature (TOT) and hottest spot temperature (HST) being natural outcome of this process, an accurate estimation of these parameters is of particular importance. IEEE/IEC among others, have proposed procedure to estimate the temperatures, however, the accuracy of the predictions are not always as good as are desired. Unacceptable temperature rise may occur due to several fault conditions other than overloading, and hence warrant an online monitoring of the transformer. This paper presents an improved model for predicting TOT and HST based on artificial neural network (ANN). A series of network architecture (p-trained network) have been proposed and trained for working out this task and are further optimally combined to give an improved accuracy.