Development of a novel soft sensor for flotation process of copper ore with neural network and variable selection, X. Ning, Z. Guihong, Y. Hu, S. Kai, M. Shiyi, L. Xianjie, W. Junpenga, and Duan Weijie

Flotation is one of the most important processes in the copper production industry. The copper grade in the concentrate is a pivotal indicator of the flotation process performance. However, it is difficult to achieve real-time measurement through hardware sensors. Based on the production data provided by an industrial operation, a novel soft sensor based on multi-layer perceptron (MLP) is proposed for effective monitoring of this pivotal indicator. The proposed soft sensor uses the nonnegative garrote (NNG) to perform global variable selection for MLP, while a local search (LS) approach is incorporated to improve the model presented by NNG. Simulation results show that the proposed algorithm has higher prediction accuracy and better model simplicity than other algorithms. Furthermore, the variables selected are consistent with the field experience, and the developed model can provide reference of feedback control for the optimisation of the process. Keywords: Flotation process of copper ore, nonnegative garrote, local search, soft sensor, multi-layer perceptron