Prediction model for hot rolling force in electrical steel involving phase transformation using SSA-LSTM network and mathematical model

Abstract

To enhance the rolling force prediction accuracy of non-oriented electrical steel involving phase transformation, a high-precision RFPM (rolling force prediction model) that combines the RFOM (rolling force optimisation model) with the SSA-LSTM (sparrow search algorithm and long short-term memory) network was established. The RFOM was developed by optimising the important influencing parameters, including the deformation resistance model in the phase separation region and stress state coefficient model, of rolling force. The SSA-LSTM network was proposed, which utilises the SSA to optimise the hyperparameters of the double-layer LSTM network, it enables learning the complex spatial data correlation of the rolling force and improves its prediction accuracy. The superiority of the proposed high-precision RFPM was verified, which adopts multiple evaluation indicators, by comparing the RFOM, the BP, the LSTM, and the SSA-LSTM models. The results manifest that the high-precision RFPM has better prediction performance, which has reached 94.40%, and it has better rolling force calculation accuracy, which the error ratio of δ≤5%, δ≤8%, and δ≤10% is 80.35%, 93.21%, and 97.14%, respectively. In addition, the proposed high-precision RFPM prediction time can meet the requirements of online real-time rolling force prediction. The rolling force accuracy of wide strips, especially non-oriented electrical steel involving phase transformation, can be significantly improved by optimising the factors affecting the rolling force of phase transformation electrical steel. The industrial data test shows that the high-precision RFPM based on the theoretical model and data-driven model can meet the requirements of on-site accuracy and online real-time control during the actual production in the large 1450 mm 4-high hot strip mills.