Modeling and optimization of coal blending and coking costs using coal petrography

Abstract

Coke is an important raw material in the steel industry, and its quality directly influences the smelting of iron and steel. To improve coal quality and reduce coal blending costs, we need to predict the coke quality and optimize the coal blending scheme. In this paper, we propose a modeling and optimization method based on the characteristics of the coal blending and coking process. First, we establish a model for predicting coke quality from coking petrography data, based on Gaussian functions and Xgboost-SVR. The model has two components. In the first part, we analyze the key characteristics of the coal blending and coke process, and extract features of the vitrinite reflectance distribution with Gaussian functions. In the second part, we use Xgboost to select a representative feature subset, and then use support vector regression (SVR) to create a model for predicting coal quality. Next, we formulate a multi-constraint optimization problem to describe the coal blending costs, and solve it using a modified particle swarm optimization. Finally, we demonstrate the effectiveness of our modeling and optimization method by applying it to actual process data. This shows that our proposed method can improve prediction performance and reduce the coal blending costs.