investigation of the performance of arx, armax and dynamic neural network methods in determining the effective parameters on alumina recovery efficiency from bauxite ( jajarm alumina company)

In this study, inputs included: analysis of al2o3, sio2, fe2o3 and tio2 in bauxite, the amount of input bauxite (ton/day), lime to bauxite mass ratio, al2o3, na2ou concentration and alpha (α) ratio in the input liquor to digestion step and percentage of bauxite particles larger than 90 µm and bauxite module were used to model the jajarm alumina extraction efficiency of bauxite# in this project, arx and armax models were first developed, then another model was developed using dynamic neural networks, and the results of different models were compared# the arx model was optimized by instrumental variable (iv) method# in armax model, the optimization method extended least squares (els) was used to calculate the parameter vector# dynamic multilayer perceptron neural network (dnn) method and levenberg-marquardt (lm) learning functions were also used# the results showed that armax-els method with the lowest mse error and the highest r value compared to the other methods, is the best method and arx method had the highest error rate and the lowest r value and the least accuracy in modeling of this system#