Forecasting Model of Flocculation Process Based on Neural Network

Abstract

A task of flocculation processes in the production of potassium fertilizers forecasting model building is shown. Traditional models has insufficient accuracy in compare with experimental data. A regression-differential model is adequate enough but inexplicable. A neural network for flocculation processes modeling is suggested. A choice of of back propagation algorithm for network training and sigmoid activation function is shown. Neural network software system based on FANN library is used. The network was learned by five experimental trends and tested on six trend with a good result. Statistical indices of the neural network of this structure are determined. A rational organization of training and testing of the neural network for modeling the flocculation process is studied. As a result, the possibility of using neural networks for modeling the flocculation process in specific equipment is shown.