Construction and Application of Intelligent Forecasting Model of Metallurgical Performance Based on CNN-BIGRU-Attention Algorithm

As important metallurgical performance indicators for evaluating the quality of sinter ore, the low-temperature reduction pulverization index, drop temperature, maximum differential pressure, and total characteristic value of sinter ore, their level directly affects the stability of the blast furnace production. Aiming at the problems of serious lag in sinter pyrometallurgical property detection and inaccurate prediction by single model, a sinter metallurgical performance index prediction model (CNN-BIGRU-Attention) integrating convolutional neural network, bidirectional gated recurrent unit, and Attention mechanism is proposed based on data-driven ideas. Firstly, data preprocessing is carried out on the 21 important parameters collected that affect the target variables, secondly, the parameters that have a greater impact on the target variables are screened by XGBoost and cross-validation algorithms, while the ablation and comparison experiments are carried out afterward. The experimental results show that compared with the traditional single-algorithm model, the CNN-BIGRU-Attention model has the best integrated prediction effect, and the hit rate (Acc) of the four metallurgical performance indicators is excellent, which is as high as 93.52%, 90.54%, 92.31%, and 88.26%, respectively. Finally, the CNN-BIGRU-Attention model was used for univariate analysis of raw material parameters, which intuitively revealed the influence law of different ratios of ordinary iron ore concentrates on the metallurgical performance indexes of vanadium and titanium sintered ores, and provided theoretical basis for the decision-making of sintering rationing, improvement of the metallurgical performance of vanadium and titanium sintered ores, and guaranteeing the smooth running of the blast furnace.