Intelligent recommendation model for reducing silicon deviation fluctuation of hot metal in BF and application
HAN Yang,HU Zhi-bin,YANG Ai-min,LI Jie,ZHANG Yu-zhu
DOI: https://doi.org/10.13228/j.boyuan.issn0449-749x.20220667
2023-01-01
Abstract:In order to ensure the smooth operation of the blast furnace, improve the quality of blast furnace thermal regulation, and deeply understand the dynamics of molten iron silicon content variation, the fluctuation mechanism of molten iron silicon content variation was studied with the No.3 blast furnace of an iron and steel complex in southern Hebei as the research object. Quantify the six decision spaces for decreasing the variation of silicon content in molten iron. Through the comparative analysis of single variable estimation method and complex correlation coefficient joint estimation method, the joint estimation method that can give consideration to the coupling and synergism between decision variables is selected to determine the response time of the variable momentum control decision variables of molten iron silicon content and eliminate the time lag between variables.Through the comparative analysis of Hopfield neural network, Boltzmann neural network and Elman neural network algorithm, the Elman neural network algorithm with memory function is selected, and the coupling nonlinear relationship between variables is determined by taking into account the time series characteristics of variables, so as to build an intelligent prediction model for the variation of molten iron silicon content in blast furnace. Based on the goal of minimizing the variation of silicon content in molten iron, an intelligent recommendation model for regulation and decision making is constructed; Through the comparative analysis of genetic algorithm（GA） and evolutionary strategy algorithm（ES）, the ES algorithm with adaptive variation degree is selected, the recommendation model is solved intelligently, and the optimal control decision to maintain the minimum variation of molten iron silicon content is quickly obtained. The research results show that the response time estimation algorithm verifies that the sample set generated by joint estimation of response time of multiple decision variables takes into account the coupling and cooperation between decision variables; The comparison of prediction algorithms verifies the superiority of Elman neural network, and the hit rate of the prediction model constructed by the target algorithm is up to 94.10%; The comparison of intelligent optimization algorithm verifies that ES algorithm is significantly superior to the comparison algorithm in solving speed and results of recommended models, and the model maintains the excellent characteristics of offline testing in industrial application practice.