Prediction of manganese content at the end point of converter steelmaking based on SSA-LSTM

Manganese is an important alloying element in steel, adding the right amount of manganese can not only improve the smelting operation, but also improve the performance of iron. In order to meet the steel grade composition requirements, manganese and iron alloy additions need to be made during the steel making operation. Therefore, manganese content prediction at the end of converter steelmaking plays an important role in raw material addition and smelting cost optimization. To address the problems of noise in the prediction of nonlinear time series data and low accuracy of a single prediction method, this paper proposes a method of end-point manganese content prediction based on Singular Spectral Analysis (SSA) and Long Short-Term Memory (LSTM) network. The mean absolute error of the proposed prediction method was 1.19% and the mean squared heel error was 1.48% using the converter steel production data of Hebei Jingye Group. The results show that the method can solve the problems such as the existence of more noise and the unstable prediction effect of a single method; compared with the time series prediction method without SSA processing, the prediction errors of the time series with SSA are reduced, which fully proves the effectiveness of the method.