Soft Sensor Modeling and Adaptive Model Correction Strategy for Gold Cyanide Leaching Process

Abstract

During the gold cyanidation leaching process (GCLP), due to the change of raw material properties and longterm use of equipment, production characteristics will undergo change in working conditions or slow time-varying, which will lead to degradation in model prediction accuracy and track performance. To solve this problem, this paper proposes a type process state recognition approach and an adaptive model correction strategy. Firstly,in this paper, the model residual and the rate of residual change as monitoring indicators of soft sensor model, which effectively distinguishes the process state of the leaching process from four circumstances that steady-state, working conditions, slow time-varying and outliers. Then the local moving window PLS method (L-WMPLS) and the recursive Just in Time Learning (RJITL) method were applied accordingly to correct the soft sensor model when GCLP occur slow time-varying and working conditions. Finally, the effectiveness of the proposed method is verified by simulation.