Studies On Algorithms Of Coke Quality Prediction And Optimization Of Blending Coal

With the rapid development of the national economy, the demand of steel is increasing year after year and the metallurgical coke for steel-making is also increasing rapidly. How to achieve desired targets of coke and make the coke's quality stable and minimize the cost using the existing coal resources is coking industries' goal.Coking coal is a complex process, but to a specific coking factory, its coking process and heating system will remain basically unchanged in a stable period of production, the state of coke oven equipment is also stable. These essentially unchanged factors have the same influence on the quality of coke or have changed little. Thereby the factors that affect the quality of coke can be reduced to the quality of matching coal, while the property of matching coal is decided by every single coal's property and ratio. Accordingly, adjusting the matching coal and radio can control the quality of coke and the optimization with matching and ratio can lower the cost.In this paper, according to some factories' production data in recent two years, after analyzing a single coal to matching coal's forecast on the basis of theory and method on matching coal at home and abroad, it is discovered that the felting exponent of matching coal and the max thickness of gelatine don't have weigh and the weighed average error is large. After the fuzzy clustering en the single coal, calculate the weighted average of the single coal of the same type, and then use historical data to predict the future data. The forecasting results showed that this method is better than simple weighted average forecast that the method is feasible. On coke quality forecasts using linear regression analysis, robust regression and non-linear Neural Network and Support Vector Machine method, the predictive analysis of the test is carried out at the same time. The results show that the prediction accuracy of SVM is higher than the previous three methods. The average errors of ash, sulfur, M40, M10, CRI, CSR are 0.0719, 0.0364, 0.5348, 0.1001, 0.6440, 1.0681 respectively. It is proved that the complex relationship of coking and coal blending can be correctly described by the Support Vector Machine based on the statistical theory, which provides a new method for further improving the scientificity and accuracy of the forecasting of coke's quality. On the basis of the above theory, I introduce the genetic algorithm, and combine the genetic algorithms with support vector machines, which provides a new approach for the calculating the ratio of single coal.