Sinter composition forecasting model based on big data and deep learning

Abstract

The invention discloses a sinter composition forecasting method based on big data and deep learning, and belongs to the field of sintering process control. According to the method, parameters and massive historical data related to sinter component changes in sintering production need to be taken as the basis; detecting abnormal data by adopting a box type graph method and an isolated forest algorithm, and replacing abnormal values by using a sliding window method; combining a Pearson correlation coefficient method and a key feature selection method to obtain input parameters of the model; establishing an online component prediction model based on DNN, and monitoring and predicting components such as TFe, FeO, V2O5 and CaO / SiO2 of the sinter in real time by utilizing the model according toonline detection data of a sintering machine; the model prediction result and the on-site actual detection value are good in fitting degree, and on-site operators can be assisted in timely and accurately judging the components of the sinter and the change trend of the sinter.

Description

technical field [0001] The invention relates to a method and application for establishing a sinter composition pre-model based on big data and deep learning, and belongs to the field of sintering process control. Background technique [0002] Sinter composition is one of the important quality indicators, timely and accurate grasp of the current composition of sinter plays a significant role in guiding blast furnace production. In view of the current situation that my country's iron ore resources have a high proportion of poor iron ore, before blast furnace smelting, it is necessary to mix various iron ore powders in a certain proportion, and through the processes of mixing, granulation, ignition and sintering, the powder that meets the production requirements of the blast furnace is obtained. Sinter. Due to the wide source of sintering raw materials, many varieties, large fluctuations in composition, and a large number of physical and chemical reactions involved in the sinte...

AI Technical Summary

Problems solved by technology

The BP neural network model has problems such as slow convergence speed and difficulty in online weight correction when the model involves many parameters; the model based on the SVM algorithm has shortcomings such as difficult selection of kernel functions and is not suitable for processing large-scale data samples.

Moreover, the algorithms mentioned above are all shallow learning algorithms, which are difficult to obtain complex nonlinear relationships when given a limited number of samples, resulting in limited model generalization capabilities, which in turn affect the prediction results of the model for the chemical composition of sinter

[0005] To sum up, the existing methods for judging the composition of sinter ore either have a time lag or cannot make accurate predictions. These methods can no longer meet the needs of sinter production

Images