Power system security control using a neural network trained by forced termination of an iterative scheme

For analysing the security of a power system a three layer feedforward artificial neural network (ANN) has been used as a substitute for the conventional approach which is based on repeated load flow solution. The ANN based approach does not involve any rigorous system modelling and iteration in real-time. But the backpropagation (BP) algorithm commonly used to train up an ANN i.e. fix its weights and thresholds is an iterative process and though off-line it requires very large number of iterations resulting in the wastage of costly computer time. In this paper a novel technique has been proposed in which the weights and thresholds obtainable by terminating the BP training process after a certain number of iterations without waiting for the convergence are used. Extensive simulation tests have been conducted terminating the training process deliberately after various number of iterations. Results have been presented for a practical and large sized power system i.e. the 81 bus Bangladesh grid system