A Hybrid Model for Predicting the End-Point Phosphorus Content of Electric Arc Furnace

A hybrid model based on the combination of k-means method, BP neural network and decision tree algorithm is proposed for predicting the end-point phosphorus content of electric arc furnace. The industrial data from electric arc furnace is filtered firstly by the box-plot method, and the processed data of end-point phosphorus content is classified into three clusters by k-means analysis method. Then, three BP neural networks with different parameters for each cluster are established to deal with the data overlapping problem and increase the model accuracy. In order to obtain the optimum prediction result, a new method combined with the posterior knowledge of dephosphorization ratio and the decision tree algorithm is employed. With this method, the results predicted, respectively, by the three different BP neural networks are merged according to the merging rule set established by decision tree algorithm and the identified result is taken as the final end-point phosphorus content. In comparison with the traditional BP neural network and deep layer neural network, the hybrid model increases the prediction accuracy of end-point P content to 97.8% with ±0.006% error range, and meanwhile for the error ranges of ±0.005% and ±0.004%, the prediction accuracy is 94.2% and 83.0%, respectively.