Beschreibung:
Power transformer is vital to the power grid and discovering the latent faults in advance is helpful to avoid serious problems. This paper addresses the problem of forecasting and diagnosing the faults of power transformers with small dissolved gas analyses (DGA) data samples which arise from low occurrence faults rate of transformers. Firstly, an online monitor developed in our previous work is applied to obtain the DGA data. Secondly, the ensemble learning (EL) of bagging algorithm with bootstrap resample has been used to deal with small training samples. Finally, the crisscross optimized neural network (i.e. CSO-NN) has been applied to the short-term prediction of DGA data based on which the transformer status can be forecasted. The case studies show that the proposed EL-CSO-NN algorithm integrated in the monitor is capable to achieve satisfactory classification and prediction accuracy for transformer faults forecasting