Prediction of Copper Matte Grade Based on DN-GAN Stacking Algorithm

Abstract

Aiming at the problem of insufficient data in actual production, we propose numerous process parameters for copper matte smelting, and complex relationships between process parameters and quality indicators, which make it difficult to accurately predict the copper matte grade. A method based on data neural generative adversarial network (DN-GAN) data generation and ensemble learning is proposed. First, the DN-GAN network is used to expand the data to solve the problem of insufficient data volume, where the original data are fused with the expanded data to form a new dataset. The tree-structured Parzen estimator (TPE) optimization algorithm is used to optimize the hyperparameters of the basic model in the stacking integration algorithm. Finally, the optimal hyperparameter combination is adopted to predict the grade of copper matte. The experimental results show that the prediction method proposed in this paper has a mean square error (MSE) of 0.093, a mean absolute error (MAE) of 0.236, and a goodness of fit (R²) of 0.979. Thus, the effectiveness of the generated adversarial network and stacked ensemble prediction in this paper has been verified, providing a novel approach for predicting the grade of copper matte.