The Research On The Application Of Rough Set And Support Vector Machine In The Prediction Of Spontaneous Combustion In Caving Zone

As one of the natural disasters, spontaneous combustion in caving zone is a seriousthreat to mine safety and production.The loss it caused is huge.And it is a serious threatto the lives of the lives of the mine workers and the sustainable development of coalindustry. So it is meaningful to study the prediction of spontaneous combustion incaving zone. It can provide the foundation of the forecast information for coalmine.Effectively reduces the losses and casualties caused by spontaneous combustion, toachieve the purpose of fire disaster reduction. In the study of coal spontaneouscombustion, people gradually found there are various factors should be considered inpredictive index system selection. And it has a certain relationship with the spontaneouscombustion, this relationship is uncertain, is a kind of Nonlinear mapping. It broughtcertain limitations and difficulties in prediction of spontaneous combustion in cavingzone. Therefore, in the face of a large number of goaf spontaneous index measurementdata, how to identify the key indicators information, how to build a stronggeneralization capability of the model to simulate nonlinear relationship between theindex information and goaf spontaneous. These are some problems that need urgentsolution still exist in the study on prediction of fire in goaf areas.This paper use a approach to predict spontaneous combustion in caving zone byusing Rough Set and Support Vector Machine (RS-SVM) was proposed. The algorithmcombing rough set theory and OPTICS algorithm is proposed by discretization ofcontinuous attributes. It used in spontaneous combustion of coal sample the originalcontinuous data discretization. And RS is used to acquire spontaneous combustion incaving zone information from the discretized decision table. It means that using RS toremove redundant information and reduce the dimensionality of data in the premise ofmaintaining data integrity. Preprocessed sample data were used as training sample dataof SVM. With choosing an appropriate kernel function, the SVM parameters wereoptimized by using k-fold cross-validation and grid searching algorithm. Finally, acomparison of the performance of RS-SVM with SVM and neural networks was carriedout. The experimental results show that the prediction based on RS-SVM can improvethe recision of classification, short training time and good generalization performance. This study can provide a reasonable reference for coal mine monitoring workers to avoidand reduce coal spontaneous combustion fire accident.