Nonlinear fitting by using a neural net algorithm

A novel transfer function which is very suitable for normalized data set and a modified conjugate gradient algorithm which converges much faster we proposed to improve the performance of the neural network training procedure. The overfitting problem is discussed in detail. The optimal fitting model can be obtained by adjusting the number of hidden nodes. A data set of furnace lining durability was used as an example to demonstrate the method. The predictive results were better then that of principal component regression and partial least square regression.