Abstract: In order to improve the desulfurization efficiency of KR process in iron and steel smelting process, this paper proposes a prediction model of desulfurizer addition based on improved grey wolf optimization algorithm (IGWO) and deep neural network (DNN). Firstly, the prediction samples were constructed based on the actual production data of the steel plant, and the importance of features was evaluated by random forest (RF). Six process parameters significantly related to the target variable of CaO desulfurizer were selected. Secondly, a good point set strategy and a delayed nonlinear convergence factor are introduced into the standard grey wolf optimization algorithm (GWO) to balance the global exploration and local development capabilities of the algorithm. Finally, IGWO is used to optimize the initial weights and thresholds of DNN, so as to improve the convergence speed and prediction accuracy of the network. The experimental results show that the determination coefficient (R2) of the proposed IGWO-DNN model on the test set reaches 0.93, and the mean absolute error (MAE) and root mean square error (RMSE) are 105.42 and 152.18, respectively. Compared with the traditional model, this method can effectively capture the discrete and fluctuating characteristics in the desulfurization process when dealing with the coupling of complex working conditions and multi-dimensional parameters and significantly improve the adaptability and prediction accuracy of the actual production conditions.

Key words: grey wolf optimization algorithm; deep neural network; additive amount of desulfurizer; good point set strategy; convergence factor