An Online Robust Soft Sensing Method for Blast Furnace Hot Metal Quality

Abstract

The present invention provides an on-line robust soft measurement method for blast furnace molten iron quality, comprising: selecting six controllable variables with the highest correlation to blast furnace molten iron quality parameters among the controllable variables in the blast furnace smelting process as input variables; and simultaneously selecting output variables; Determine the order of the random weight neural network model; initialize the relevant parameters and variables of the random weight neural network; robust initial stage; use the random weight neural network model and the acquired blast furnace ironmaking process data to estimate the current molten iron quality parameters online; Stick online sequential learning phases. In the present invention, the online sequential random weight neural network based on the Cauchy distribution weighted M estimation is introduced, and the contribution of the sample data to the establishment of the model is determined according to the size of the residual, which solves the adverse effects of a large number of outliers on the modeling in the modeling process At the same time, it can constantly correct the model parameters according to the newly measured blast furnace ironmaking process data containing outliers, adapt to the current working conditions, eliminate the influence of outliers and accurately predict.

Description

technical field [0001] The invention belongs to the technical field of automatic control of blast furnace smelting, and in particular relates to an online robust soft-measurement method for molten iron quality in a blast furnace. Background technique [0002] Blast furnace ironmaking is a very complex nonlinear dynamic process of reducing iron from iron ore and other iron-containing compounds to smelt qualified molten iron. As the most important production index in the blast furnace ironmaking process, the molten iron quality index directly determines the quality of subsequent steel products and the energy consumption state of the blast furnace smelting process. In order to achieve the goals of high quality, low consumption, high yield and long life, it is necessary to monitor and control the blast furnace ironmaking process in real time. At present, parameters such as silicon [Si] content (chemical heat), molten iron temperature (physical heat), sulfur [S] content, and pho...

AI Technical Summary

Problems solved by technology

These methods mainly consider the soft sensing of molten iron quality parameters under ideal furnace conditions, and their robustness is poor. When the modeling data contains outliers, these methods cannot suppress the interference of outliers and predict the molten iron quality parameters more accurately.

[0007] The patent application number "201610118914.7" applied for "a multivariate molten iron quality soft-sensing method based on robust random weight neural network" can solve the above problems, but a large amount of data is continuously generated in the dynamic process of ironmaking, and has significant engineering Due to the time-varying and nonlinear dynamic characteristics of the situation, as the dimension of the data volume increases, the time and computational complexity of solving the inverse operation of the output matrix of the hidden layer increase rapidly. This method does not have the ability to handle large amounts of data and needs to be used regularly The new ironmaking process data retrains the prediction model offline, and the model parameters cannot be updated online in real time according to the new process data to adapt to the current working conditions for accurate prediction, which increases equipment costs and labor costs

To sum up, at present, there is no online robust soft-sensing method for multivariate dynamic adaptive working conditions for molten iron quality parameters (Si content, P content, S content, and molten iron temperature) in the blast furnace smelting process at home and abroad.

Images