Molten steel temperature prediction model based on bootstrap Feature Subsets Ensemble Regression Trees

• Large-scale and noise data impose strong restrictions on building temperature models.
• To solve these two issues, the BFSE-RTs method is proposed in this paper.
• First, feature subsets are constructed based on multivariate fuzzy Taylor theorem.
• Second, smaller-scale and lower-dimensional bootstrap replications are used.
• Third, considering its simplicity, an RT is built on replications of each feature subset.

Molten steel temperature prediction is important in Ladle Furnace (LF). Most of the existing temperature models have been built on small-scale data. The accuracy and the generalization of these models cannot satisfy industrial production. Now, the large-scale data with more useful information are accumulated from the production process. However, the data are with noise. Large-scale and noise data impose strong restrictions on building a temperature model. To solve these two issues, the Bootstrap Feature Subsets Ensemble Regression Trees (BFSE-RTs) method is proposed in this paper. Firstly, low-dimensional feature subsets are constructed based on the multivariate fuzzy Taylor theorem, which saves more memory space in computers and indicates ``smaller-scale'' data sets are used. Secondly, to eliminate the noise, the bootstrap sampling approach of the independent identically distributed data is applied to the feature subsets. Bootstrap replications consist of smaller-scale and lower-dimensional samples. Thirdly, considering its simplicity, a Regression Tree (RT) is built on each bootstrap replication. Lastly, the BFSE-RTs method is used to establish a temperature model by analyzing the metallurgic process of LF. Experiments demonstrate that the BFSE-RTs outperforms other estimators, improves the accuracy and the generalization, and meets the requirements of the RMSE and the maximum error on the temperature prediction.