Development and Application of Prediction Model for End-point Manganese Content in Converter Based on Data from Sub-lance

摘要

Base on smelting data from converter sub-lance in a factory, the prediction models for end manganese content in converter were established by Multiple Linear Regression (MLR) and BP Neural Network (BP-NN) respectively. Prediction results showed that, MLR model was easy to set up, but could not accurately describe steelmaking process and its results were unsatisfactory, while BPNN model got more accurate prediction results for end manganese content in converter based on proper selection of model structure, adequate training using sample data and then correct determination of the weights. According to the spot tests, prediction relative error hit rate of 90.38% within ± 10% or 96.15% within ± 15%.