Prediction of coal flotation performance using a modified deep neural network model including three input parameters from feed
Xiangning Bu, Shaoqi Zhou, January Kadenge Danstan, Muhammad Bilal, Fawad Ul Hassan, Chao Ni
Abstract
Flotation is an effective method widely used in coal preparation. However, the complexity of the flotation process and the actual production site make the quality detection of flotation products mainly depend on manual work. This makes the automation of the flotation process and the stability of product quality difficult. Furthermore, the published literature indicates commonly using only a few dozen data sets and application of traditional machine learning methods to predict the flotation performance. In this study, a total of 641 sets of valid data were collected for training and validation of the model. On this basis, a deep neural network (DNN) model was built to predict the quality of flotation products. 80% of the total data were used for training, and 20% for testing the DNN model. To improve the prediction accuracy and generalization performance of the model, a data augmentation method was applied and a new loss function was proposed. Finally, a DNN model was obtained, which is capable of maintaining high prediction accuracy by simply adjusting the input parameters while the feed properties are changing.