The main task of the impurity removal process is to control the oxidation reduction potential (ORP) within the range of the optimized set value. The impurity removal process is essentially an oxidation-reduction process. Oxidation reduction potential (ORP) is an external reflection of reaction state inside the impurity removal reactor. However, actual industry is time-varying, nonlinear and multimode. It is difficult to determine the appropriate dosage of impurity remover in practice. This will lead to large fluctuations in the operation mode, affecting the safety and stability of the process and the final product quality. To solve these problems, an intelligent dynamic setting control framework (IDSCF) for the multimode impurity removal process is proposed in this paper. It includes a preset module of the dosage of impurity remover based on impurity remover utilization (IRU) estimation, an operation mode detection module based on autoencoder, a normal mode adjustment module based on fuzzy logic, and an unsteady mode adjustment module based on case-based reasoning (CBR). The framework can determine the reasonable preset dosage of impurity remover and adjust the dosage according to the current operation mode of the impurity removal process. Because the operation mode is related to the residual dosage of impurity remover added over a period of time, that is, the accumulative effect of the large-scale metallurgical reactor. When calculating the preset dosage of impurity remover, the reactant accumulation ratio (RAR) is calculated, which makes the calculation of the preset value more reasonable. In addition, it can detect the unsteady modes causing large fluctuations in the process and adjust them in time. Experiments are carried out in accordance with the data of an actual cobalt removal process. The results show that this method can effectively improve the stability of the impurity removal process, control the ORP within the set range and cope with complex mode changes.