Adaptive training of artificial neural network

Adaptive training of a neural network for nonstationary processes is reported within the framework of a multilayer perceptron model using the backpropagation (BP) algorithm. The error introduced by small changes in system parameters is reflected to adapt the changes in the converged weight matrix. The error is minimized using a constrained optimization method like the gradient projection method (GPM). The method is applied for harmonic prediction in voltage waveforms. The results for a sample system are discussed.© IEE