Radial Basis Function Neural Network based on Growing Neural Gas Network applied for evaluation of oil agglomeration process efficiency

Abstract

In this study, the neural model for modeling of oil agglomeration of dolomite in the presence of anionic and cationic surfactants (sodium oleate and dodecylammonium hydrochloride) was implemented. The effect of surfactants concentration, oil dosage, time of mixing, pH, and mixing speed of the impeller in the process recovery were investigated using Radial Basis Function Neural Network (RBFNN). A significant problem in this modeling, was the selection of the structure of the neural network. In algorithms based on the RBFNN, the issue mentioned relates to the number of nodes in the determination of the hidden layer. Also, the distribution of functions in data space is significant. In the proposed solution, at this stage of the neural model design, the Growing Neural Gas Network (GNGN) was implemented. Such a procedure introduced automation of the calculation process. The centers were obtained from the GNGN and the structure (number of radial neurons) can be approximated based on a simple searching algorithm. The idea of the data calculations was implemented as an original algorithm that can be easily transferred to Matlab, Python, or Octave software. The values predicted from the neural networks model were in good agreement with the experimental data. Thus, the RBFNN-GNGN model used in this study, can be employed as a reliable and accurate method to predict, and in the future to optimize the performance of oil agglomeration process.