Study On Heat State Forecasting Model Of Blast Furnace Based On Particle Swarm Optimization Improved BP Neural Network

The puddling of blast furnace plays an important part in iron and steel industry which is stanchion industry of country economic.The temperature of blast furnace ensures that the blast furnace could work steadilv,which is an important guideline of estimating condition.How to establish and realize the temperature prediction model has not only important theoretical value but also practical value.However,from the theory of cybernation,the process of blast furnace is non-linearity system which has much difficulty in modeling.And the expression of temperature is especially harder. In this paper,according to the actual situation of 4# blast furnace of a large domestic iron and steel enterprises, puts forward the chemical heat that the Ti content of hot metal instead of physics heat to judge changes of temperature of blast furnace.On this basis, in accordance with the development of static model analyzes the law and the quantitative relationship of smelting process of 4# blast furnace; By analyzing the major factors of the impact of heat system of hearth and the difference between the smelting heat system of general ore and the heat system of vanadium perovskite,studys the characteristics of the heat system of hearth of 4# blast furnace.This paper based on precocious phenomenon of PSO algorithm, proposes a PSO algorithm improved,which is based on adaptive adjustment of learning factor; the performance of algorithm testing shows that PSO algorithm improved converges qiuckly and has the convergence probability and search accuracy. Then proposes that using PSO algorithm improved train BP neural network, in order to improve network results of training.According to the own characteristics of network model, need control the number of input parameters, and parameters should be selected to choose a greater impact on the output parameters.In this paper,followed by the correlation analysis between process parameters and the correlation analysis between the process parameters and Ti content of hot metal,obtaineds more sensitive parameters to temperature changes, as well as the lag times of the process parameters impact on the temperature to achieve choice on input parameters of network model.Finally, establishes heat state dynamic forecasting model of blast furnace based on the PSO-BP neural network; At the same time the actual datas in the process of operation on 4# blast furnace are pretreated, then in accordance with sample datas pretreated achieves model simulation experiment. The experimental results show that the prediction accuracy of Ti content of hot metal based on BP-PSO network improved is higher than that of the standard BP Network. This accuracy is to meet the needs of actual production Completely, and Compared to standard BP network,BP-PSO network improved significantly shorten training time.