Implementation of acid concentration model based on MSPRNN for a steel pickling process

Abstract

This paper presents a implementation of acid concentration model based on multi-step prediction recurrent neural network (MSPRNN) for a steel pickling process. The MSPRNN for predicting the values not in only one-step future but in multi-step future is applied to predict the acid concentration in the steel pickling process. The basic MSPRNN is a recursive structure predicting the multi-step future targets using the distant past inputs and the previous predicted targets. On the other hand, the proposed MSPRNN is a structure that predicts the multi-step future targets using distant past inputs and distant past targets. Even though the nonlinearity is strong because of the large time difference between the available inputs and the targets to be predicted, the proposed MSPRNN maintains robust prediction of the acid concentration in the multi-step future. © 2020 IEEE.