| Sintering is a complex physical and chemical process.It is the primary link in the steel production process.The quality of sinter directly affects other subsequent production such as blast furnace ironmaking.Sintering ore and sinter quality prediction are two important components in controlling the quality of sinter in sintering.At present,in both aspects,it is still in the stage of relying on experience value and adjusting process parameters,but it has problems such as wasting resources,high production cost and low production efficiency.Through the historical production data accumulated by steel companies for many years,from the perspective of industrial data,the establishment of the model is relatively simple and easy to achieve,and can achieve the purpose of meeting production requirements,reducing production costs,and improving production quality.Therefore,this paper will study the sintering ore optimization method and the sinter quality prediction method through the industrial data in the sintering production process accumulated by the iron and steel enterprises for many years.The main research work is as follows:(1)A particle-distribution D-PSO-based sintering ore blending algorithm is proposed,and a blending model with cost optimization is established.Based on the particle velocity update of the standard PSO algorithm,the distance between the particle and the current optimal solution is integrated into the particle velocity update,so that the particle has the ability to judge independently according to the distance,thus solving the problem that the standard PSO is easy to fall into the local optimum.The actual production data of sintering ore blending is used as the experimental data,and the artificial proportioning scheme is compared with the D-PSO algorithm,the standard PSO algorithm,the genetic algorithm and the ant colony algorithm.The experimental results show that the D-PSO algorithm is Under the premise of meeting the requirements of sintering ore production,the production cost of the enterprise is reduced,and the iron content in the sinter is increased,and the content of harmful components such as sulfur and phosphorus is lowered.(2)A multi-objective particle swarm sintering blending optimization algorithm based on MOEA /D(MOEA/D-DPSO)is proposed,and a multi-target blending model with the lowestsulfur content and cost index is established.This algorithm combines the D-PSO algorithm with improved global convergence ability with MOEA / D,and introduces a reverse learning strategy to balance the convergence and distribution,thereby balancing the global convergence ability and local search ability of the algorithm during the evolution process.Experimental results show that compared with the artificial ore blending scheme,the algorithm can effectively reduce the sulfur content and cost on the premise that it meets the sintering blending production requirements.In addition,compared with the improved MOEA/D and multi-strategy mechanisms of MOEA/D algorithm,MOEA/D-DPSO effectively improves the algorithm's global convergence ability and prevents it from falling into a local optimum.(3)A sinter quality prediction algorithm based on attention mechanism-LSTM is proposed to establish a sinter quality prediction model.On the basis of the LSTM network,using the attention mechanism idea,through the redistribution of weights,paying more attention to data other than missing data,and suppressing redundant data in the data that does not contribute more to training.From another level,the consideration of missing data is strengthened,and the calculation of the loss function in LSTM is improved.The proportion of missing values in the total input data is taken into account,and the loss function is reduced during the training process.Make model predictions more accurate.The experimental results show that the prediction accuracy of the LSTM prediction method with increased attention mechanism can reach more than 92.7%,which is 1.9% higher than the original LSTM. |
PDF Full Text Request