To address the challenges of low accuracy and poor robustness of traditional single prediction models for blast furnace molten iron temperature, a hybrid model that integrates the improved complete ensemble empirical mode decomposition with adaptive noise, kernel principal component analysis, support vector regression and radial basis functional neural network is proposed for precise and stable iron temperature prediction. First, the complete ensemble empirical mode decomposition is employed to decompose the time series of iron temperature, yielding several intrinsic mode functions. Second, kernel principal component analysis is used to reduce the dimensionality of the multi-dimensional key variables from the steel production process, extracting the major features of these variables. Then, in conjunction with the K-means algorithm, support vector regression is utilized to predict the first column of the decomposed sequence, which contains the most informative content, evaluated using the Pearson correlation coefficient method and permutation entropy calculation. Finally, radial basis function neural network is applied to predict the remaining time series of iron temperature, resulting in the cumulative prediction. Results demonstrate that compared to traditional single models, the mean absolute percentage error is reduced by 54.55%, and the root mean square error is improved by 49.40%. This novel model provides a better understanding of the dynamic temperature variations in iron, and achieves a hit rate of 94.12% within a range of ±5℃. Consequently, this work offers theoretical support for real-time control of blast furnace molten iron temperature and holds practical significance for ensuring the stability of blast furnace smelting and implementing intelligent metallurgical processes.
Full Text Finder