| Coke is an indispensable raw material in the process of iron making.Its quality has a vital impact on the smooth running of the blast furnace and the quality of pig iron.In order to ensure the quality of pig iron,Most of the existing coking companies use more high-quality coking coal,resulting in the rise of cost and the waste of high-quality resources.Therefore,more accurate prediction of coke quality and optimization of coal blending scheme,so as to rational use of coking coal resources,has become the key problem that coking enterprises need to solve.In view of the complex reaction of the coking process and the difficulty of modeling based on the mechanism,a coke quality index prediction model based on Multi-layer Back Propagation Neural Network(MLBPNN)was established.In order to better fit the complex functional relationship between input and output in the coking process,the number of hidden layers in the model was increased.Aiming at the problem that the number of hidden layers and the number of nodes in the model were difficult to determine the optimal value,the Genetic Algorithm(GA)was used to optimize each parameter.On this basis,with the goal of reducing coal blending costs and meeting coke quality requirement,considering the constraints of coal inventory,various quality indicators of coal blending and coke,the optimization model of coal blending in the coking process was established.Furthermore,an improved NSGA-III model was proposed to solve the Pareto optimal solution set of the coal blending optimization model.In this algorithm,a new comparison criterion is proposed to evaluate the constraint dominance relationship in the evolution process of the population,so that the individuals can be quickly non-dominated ranking,and the redundant individuals are eliminated by using the reduplication strategy,which improves the efficiency of the solution.The actual production data of a domestic iron and steel company coking plant was used for verification.The results show that the coke quality prediction model based on multi-layer BP neural network constructed in this paper has higher prediction accuracy than other existing methods,while the Pareto optimal solution set based on the improved NSGA-III coal blending optimization method has better distribution and convergence.Furthermore,based on the above model,the coke quality index prediction and coal blending optimization software system was designed and developed,and it has been practically applied in the coking plant of a large domestic steel enterprise for more than one year.The application statistical results show that the prediction accuracy of coke quality of the system completely meets the on-site production requirements,and the execution rate of the optimization scheme of coal blending ratio reaches more than 90%,which provides an effective guide for reducing the coal blending cost in the coking process. |
