Gold recovery modeling based on interval prediction for a gold cyanidation leaching plant

Abstract

The production index of gold cyanidation leaching process has an important influence on the economic benefits of the plant-wide hydrometallurgical process. In the actual leaching production process, due to the fluctuations of the previous procedure, the leaching process is often affected by the uncertainty and process disturbance. And hence, the prediction accuracy of the traditional point prediction models (such as ANN: Artificial Neural Network) decreases seriously and cannot provide any quantified information about the uncertainty or process disturbance. To solve the above problem, the interval prediction technique based on Radial Basis Function (RBF) ANN is proposed and used to model a gold cyanidation leaching plant encountering uncertainty and process disturbances in this paper. The objective function trained in the interval prediction model is not based on the prediction error, but the comprehensive measure index, namely, the coverage width criteria, which consists of the prediction interval coverage probability and the prediction interval normalized averaged width. Compared to the traditional point prediction, when the process uncertainty and disturbances are present, the interval prediction model will provide more helpful process information to the operators or process designers for determining further optimization and control strategies. The simulation and practical application results show that most of the real values of gold recovery can be covered between the upper and lower bounds with a predefined probability, which indicates the effectiveness and reliability of the interval prediction model, thus laying an important foundation for plant-wide optimization and control.