LAUSR

Online pH estimation is essential in the operational optimization of a leaching process. However, delayed and low-rate artificial testing is adopted instead of online pH meter, due to the hostile production environment and heavy maintenance workload. This article proposes an online pH estimation approach based on a mechanism-based semisupervised long short-term memory (MSSLSTM) network. A kinetic reaction model and an accumulative acid-material ratio model incorporate the reaction mechanism in the machine learning framework. It is built to describe the reaction state qualitatively and quantitatively. The artificial testing result of the last sampling instant is held and imported into the network to realize semisupervised learning, which is consistent with the multirate condition in practice. The verification results demonstrate the accuracy and stability of the estimation model, which systematically learns from the reaction state features, low-rate artificial pH testing results, and high-rate online measured basic variables.