摘要：In this paper, a serial hybrid model of a gold cyanidation leaching process is proposed. The serial hybrid model consists of mass conservation equations of gold and cyanide as well as two kernel partial least square (KPLS) models, which are used as the estimators of the unknown kinetic reaction rates without the complicated kinetic model structures considered. The proposed serial hybrid model makes full use of both the a priori process knowledge and the ability of a data‐driven model to discover the information behind data sets. Moreover, before training the KPLS models, the proposed estimation strategy based on Tikhonov regularization is used to estimate the kinetic reaction rates, which can mitigate the effect of measurement noise on the estimation results effectively. The proposed serial hybrid model has been applied to a gold cyanidation leaching plant to predict the gold leaching rate. The prediction results show that the proposed serial hybrid model can track the real leaching rate of gold closely and has the best prediction accuracy at both dynamic and steady states compared with the pure KPLS and mechanistic models, thereby laying an important foundation for the successful implementation of optimization and control of the leaching process.
