ANN modeling of a steelmaking process

Abstract

In a steelmaking shop, the output 'yield' is considered as an important performance measure while producing a specific amount of steel. It represents the operational efficiency of the steelmaking shop. It is the objective of management to maintain a high percentage of the yield. The present study was performed considering the open-hearth process of a steelmaking shop. The best subset regression analysis is applied to determine the most influencing variables influencing the output of the steelmaking process. Then, the multi-layer feed forward neural network with Levenberg-Marquardt (L-M) backpropagation algorithm is presented to predict the output yield of steel. In the present investigation, 0.0001 of MSE is set as a goal of the network training. The overtraining is given cognizance during the model building. The overall average of absolute percentage error (APE) of the model is found to be 0.5145% and the predicted yield based on the neural network is found to be in good agreement with the testing data set. © 2013 Springer International Publishing.