Optimizing parameters of supervised learning techniques (ANN) for precise mapping of the input–output relationship in TMCP steels

Abstract

Optimization of artificial neural network architecture and training algorithm is undertaken to map the input–output relationship in thermomechanically processed high‐strength low‐alloy steels. Primarily, the model complexities are varied by varying the number of hidden layers and hidden units. A number of algorithms are tried for training the network. Also, different transfer functions are tried to find the best option. It is found that a four‐layer network with 48 hidden units can perform best in terms of attaining the lowest training error when the network uses hyperbolic tangent transfer function, and is trained with scaled conjugate gradient algorithm or Levenberg–Marquardt algorithm.