Grey relational analysis and prediction on limit parameters of coal spontaneous combustion in goaf

Abstract

The dynamic change of goaf determines the complexity between limit parameters of coal spontaneous combustion and its influencing factors. In order to achieve accurate prediction of limit parameters of coal spontaneous combustion, the grey correlation analysis was introduced to quantitatively analyze the correlation between the superior intensity limit of air leakage and its influencing factors. Further, the key parameters of support vector regression(SVR) were optimized by improved particle swarm optimization (IPSO), and the prediction model on limit parameters of coal spontaneous combustion in goaf was established. Taking the superior intensity limit of air leakage as an example, the standard PSO-SVR and the multiple linear regression model (MLR) were respectively established. Moreover, the models mentioned above were used for predicting the superior intensity limit of air leakage and compared with the existing method of variable-step grid search for optimizing SVR parameters and the neural network. The results show that the grey synthetic degree between the inferior thickness limit of coal and the superior intensity limit of air leakage was the largest, the oxygen concentration and the coal temperature were the second, and the exothermic intensity and the distance of the goaf from the working face were the smallest. The nonlinear relationship between limit parameters of coal spontaneous combustion and its influencing factors is more remarkable than the linear relationship. SVR has stronger nonlinear processing ability than neural network method. The relative error between predicted values and real values of IPSO-SVR model is within 2.6%, and its prediction accuracy is better than other four models. It is helpful for the key parameters optimization by IPSO to improve the prediction precision of SVR model. The IPSO-SVR method is effective for the prediction on limit parameters of coal spontaneous combustion.

Keywords

• grey correlation analysis
• particle swarm optimization
• support vector regression
• neural network
• limit parameters