Development of a supervisory training rule for multilayered feedforward neural network using local linearization and analytic optimal solution

A new supervisory training rule for the multilayered feedforward neural network (FNN) using local linearization and analytic optimal solution is proposed. The cause of the nonlinearity of the neural network in the training is pinpointed and the nonlinearity is removed by a local linearization. And, the optimal solution of the linearized FNN minimizing the objective function for the training is analytically derived. The proposed training rule shows the shortest training time among the previous approaches. The superiority of the proposed approach is demonstrated by applying the proposed training rule to the modeling of the pH process.