A novel technique for current prediction in 33 kV substation

Current prediction is a vital and an important aspect of power metering and control systems. Not only does it help avoid overloading shutdown situations but can also decide the rating of certain switchgear. In this paper both normal and fault condition current prediction is done using Artificial Neural Network (ANN). Performance of the ANN largely depends on how well its weights are trained. Learning algorithms used for this purpose should be robust and have the lowest possible margin of error between desired and actual outputs. We have done a comparison of two different learning algorithms - Back Propagation (BP) and particle swarm optimization (PSO) for both normal and fault current prediction in 33 kV feeders at the BSES Yamuna Power Ltd. substation (New Delhi) connected to the Northern grid. The performance index in both cases is analyzed and then compared. The results obtained show that PSO, being a group based learning algorithm is the better of the two.