Comparison of Prediction Models for Power Draw in Grinding and Flotation Processes in a Gold Treatment Plant

摘要

As one of the principal anticipated goals in 2015, government and scientists have been paying increasing attention to energy saving. Energy-saving potentials play an important role in economical and sustainable development in the gold industry. Through analyzing the factors that significantly influence energy consumption in the grinding and flotation processes in a gold treatment plant, three models for energy consumption prediction are established based on large amounts of actual production data. The multiple linear regression model demonstrates low prediction accuracy. In consideration of the advantages of artificial neural networks (ANNs), a back-propagation (BP) neural network model is built to provide higher prediction accuracy. Moreover, a hybrid GA-BP neural network model is established combining the typical characteristics of a genetic algorithm (GA) and a BP neural network. Subsequently, validation and comparison of the relative prediction errors, as well as the RMSE of the three models illustrate that the hybrid GA-BP neural network model presents the highest prediction accuracy. The total shift percentage of the hybrid GA-BP neural network model is 98% and 80%, when the relative prediction errors of the model are within +/- 5% and +/- 3%, respectively, and its prediction results show a minimum RMSE of 1.29. In contrast, of the three models, the hybrid GA-BP neural network model can provide the highest prediction accuracy of energy consumption, and consequently, can offer a positive reference for real production.