An Improved UKFNN Based on Iterative Renewal for Intelligent Modeling of Aluminum Electrolysis

摘要

Establishing an intelligent production model for aluminum electrolysis has always been a technical problem due to the existence of more parameters and multiple chemical and physical reactions in the manufacturing system of aluminum electrolysis. An iterative renewal approach tagged with Iterated Unscented Kalman Neural Networks (IUKFNN) is developed to approximate the intelligent modelling of direct current consumption from the electrolytic aluminum industry in this study. In the propose model, iterative idea and UKFNN are integrated. And the new idea of iterative updating is adopted to improve estimator precision of the model. To be more precise, after estimated states are obtained in the UKFNN algorithm, the estimated value is returned to the measurement updating to re-sample, so that the estimated value is more accurate. The experimental results indicate that the model can significantly predict the direct current consumption in the manufacturing system of aluminum electrolysis, and the precision of the model IUKFNN is higher than UKFNN. Consequently, this paper presents a better intelligent modeling method for further approaching the real energy consumption model in electrolytic aluminum manufacturing process.