Neural diagnostic system for transformer thermal overload protection

Recent studies by various authors have shown that the IEEE Transformer Load ing Guide model and the more recent modified equations. proposed by the K3 Working Group of the IEEE Power System Relaying Committee, are lacking in a ccuracy in the prediction of the maximum winding hot-spot temperature of a power transformer in the presence of overload conditions. The result is a r eal winding hot-spot temperature greater than the predicted one. A novel te chnique to predict the maximum winding hot-spot temperature of a power tran sformer in the presence of overload conditions is presented. The proposed m ethod is based on a radial basis function network (RBFN) which, taking in t o account the load current; the top oil temperature rise over the ambient t emperature and other meteorological parameters, permits recognition of the hot-spot temperature pattern. Data obtained from experimental tests allows the RBFN-based algorithm to be tested to evaluate the performance of the pr oposed method in terms of accuracy.