基于MIV-GA-BP模型预测烧结矿FeO含量

工艺绿色化、装备智能化、产品高质化已成为当前钢铁行业主要发展目标。作为影响烧结矿性能的重要指标之一,FeO的含量不仅影响烧结矿还原性的高低和烧结过程的能耗,而且在一定程度上影响高炉间接还原、燃料比等指标。针对目前研究过程中存在的数据量少、工艺结合不紧密、特征选择方法针对性不强等问题,提出了基于MIV-GA-BP算法的烧结矿FeO含量预报模型。以承钢3号烧结机1年的生产数据作为研究基础,首先选取BP神经网络作为深度学习模型,然后利用遗传算法的特点解决了网络调参难等问题,成功构建了基于遗传算法优化的BP神经网络模型。在特征选取阶段将MIV算法的优越性与工艺理论相结合,选取了拥有更好解释性的参数作为模型的输入,此方法提高了模型预测准确率,成功实现了烧结矿FeO含量的预测。上线测试结果表明,误差允许范围内模型命中率达到87.9%,对现场烧结生产具有更好的指导性。

Prediction of FeO content in sinter based on MIV-GA-BP model

Green technology, intelligent equipment and high-quality products have become the main development goals of current steel industry. As one of important indicators affecting the performance of sinter, the content of FeO not only affects the reducibility of sinter and the energy consumption of sintering process, but also affects the indirect reduction of blast furnace, fuel ratio and other indicators to a certain extent. Aiming at the problems in current research process, such as the small amount of data, lack of close process integration, and the weak pertinence of feature selection method, a prediction model of FeO content in sinter based on the MIV-GA-BP algorithm was proposed. One-year production data of Chengde Iron and Steel Co., Ltd. No.3 sintering machine was taken as the research basis, firstly the BP neural network was selected as the deep learning model, and then the characteristics of genetic algorithm was used to solve the problem of difficulty in network parameter adjustment, therefore, the BP neural network model which was optimization based on the genetic algorithm was constructed successfully. The advantages of MIV algorithm and the process theory were combined in feature selection stage, and the parameters with better interpretability were selected as the input of model. The method improves the model prediction accuracy and successfully realizes the prediction of FeO content in sinter. The online test results show that the hit rate of this model within allowable error range reaches 87.9%, which has better guidance for field sintering production.