Research On Prediction And Optimization Model For Intelligent Recommendation Of Pelletizing Ingredients

As the representative of complex process industry,the intelligent manufacturing level of iron and steel industry is more industry-leading and exemplary.With the increasing demand of steel products in China,the limited natural rich lump ore resources can not meet the demand of iron-making raw materials on a larger scale.As one of the artificial rich lump ores,pellet ore has the advantages of uniform particle size,many micropores and high mechanical strength.Based on the various excellent characteristics of pellet,it has become an indispensable part of blast furnace burden.However,pellets need to undergo external mechanical action from production to entry into blast furnace,which will cause pellets to break and affect the quality of entry into the furnace.Compressive strength of cooked pellets is an important index to characterize mechanical strength of pellets,and the factors that determine compressive strength of pellets are closely related to pellet batching and production process.Therefore,it is very important to construct the prediction and optimization model of pellet ingredients intelligent recommendation.Based on the above research background,the main research work of this project was divided into the following three aspects:1)The fundamental principles of GRNN algorithm and the design methods of network structure was analyzed.At the same time,combined with the advantages of GRNN in nonlinear fitting and flexible network structure,an improved GRNN model for predicting the compressive strength of cooked pellets was established by adding a data compression layer between the input layer and the model layer.2)Based on the performance of typical intelligent optimization algorithm,the fundamental principles of BAS algorithm and the disadvantage of easily getting into local extremum was deeply analyzed.Then,an improved BAS algorithm model suitable for pellet proportioning optimization was established by introducing Metropolis criterion of SA algorithm.3)Based on the historical production data of pellet proportioning,the prediction model of the compressive strength and the optimization model of pellet proportioning was coupled to realize the intelligent recommendation of pellet raw materials.The result of research shows that the GRNN prediction model with data compression algorithm has little error in predicting the compressive strength of pellets.The range is-5N to 5N.Based on the minimum predicted value of 1355 N and the maximum predicted error of 5N,the relative error of prediction is 0.37%.Under the optimization model of pellet proportioning,the maximum increase range of pellet compressive strength is 30.04%,the minimum improvement range is 8.24%,and the average improvement range is 16.60%.The model ran stably,which provides a new method for pellet intelligent manufacturing.Figure 32;Table 7;Reference 84...