Multi-scale local LSSVM based spatiotemporal modeling and optimal control for the goethite process

• Published in: Neurocomputing (Amsterdam) Vol. 385; pp. 88 - 99
• Main Authors: Dai, Jiayang, Chen, Ning, Luo, Biao, Gui, Weihua, Yang, Chunhua
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 14.04.2020
• Subjects: Distributed parameter system; Iron removal process; Least squares; Multi-scale kernel learning; Spatiotemporal modeling; Support vector machine; Multi-scale kernel learning; Support vector machine; Spatiotemporal modeling; Distributed parameter system; Iron removal process; Least squares
• ISSN: 0925-2312; 1872-8286
• DOI: 10.1016/j.neucom.2019.12.008

The iron removal process by goethite is an important part of zinc hydrometallurgy. In existing works, the goethite process is often modeled as a lumped parameter system, where the spatial distribution information of reactants is not involved. In this paper, the spatiotemporal modeling of the goethite process and its optimal control problem are studied. To make the infinite-dimensional distributed parameter system easier to solve, space–time separation is adopted to transform it into a finite-dimensional system. Then, a multi-scale local least squares support vector machine is proposed to establish the temporal model. This method uses multi-scale kernel learning to deal with different trends of the process and establish a local model to track the state change of the system. Through space–time synthesis, the established spatiotemporal model can approximate the distributed parameter system of the goethite process. Moreover, an optimal control strategy based on the spatiotemporal model is designed to reduce the cost of oxygen and zinc oxide consumed in the process. Finally, simulation experiments on the goethite process demonstrate the effectiveness of the proposed modeling method and optimal control strategy.