Predicting tapping temperature in electric arc furnace based on an ensemble pruning framework

The accurate prediction of tapping temperature is significant for high quality products in electric arc furnace. Although existing data models have achieved better performance than mechanism models, their accuracy still needs improvement. To this end, this paper proposes an ensemble pruning framework dedicated to the prediction of tapping temperature. In contrast to traditional ensemble models that fuse results of all base models, our framework can prune some redundant base models via the clustering technique. Though such a pruning strategy, both the diversity and the accuracy have been considered, which are two key factors of ensemble learning. A dataset from a real-world electric arc furnace is used to validate our framework, effective of which has been approved by the experimental results.