A neural network meta-model for an ore crushing plant

Abstract

A data set obtained from Ite-Consultlink was used to test the capabilities and limitations of various neural networks. The different neural networks were identified and researched. The basic neural networks theory together with an overview of how these networks train was also researched. These networks were tested using a trial version of a program called Neurosolutions. The ANN’s were then compared using a matrix format. Neural networks were identified by Ite-Consultlink as a possible means for time and/or computational savings when used in conjunction with their traditional simulation modeling techniques. Output accuracy, and training times were among the requirements used to obtain suitable network performance indicators with which assess the performance of the networks. Most of the networks provided good results on the data set whilst two of the network types did not obtain a suitable accuracy in its outputs. All the networks were compared using a relative scoring system, and a recommendation was made accordingly.