矿井突水信息处理的SVM-RS模型

Processing Predictors of Water Inrush in Coal Mines Using a SVM-RS Model

提出了数据处理的支持向量机-粗集(SVM-RS)模型.根据原始突水样本构造SVM预测模型,对该模型依次约简部分属性做重复测试,当预测精度降低,表示该属性重要,予以保留,否则,予以约简,以此优化突水预测的属性集;利用SVM对连续的属性值进行离散化处理,以线性SVM的分类超平面确定属性值的断点位置;利用RS分析突水决策表,提取预测规则.该模型综合了SVM泛化性能优与RS分析数据、提取规则能力强的优势,在实际应用中表现良好.

A Support Vector Machine using a Reduced Set （SVM-RS） is presented as a model for predicting water inrush in coal mines. At first the model was built using the raw training samples. The trained model was retested, leaving out each attribute of the training samples in order to determine which attributes improved model accuracy and, thus, should be included in the final model. In this way a more satisfactory feature space was selected for predicting water inrush. Continuous valued attributes were discretized by a linear SVM, the hyperplanes being used to locate the discrete points in attribute space. Preprocessed data were analyzed by RS and prediction rules extracted. The method is novel in integrating the advantages of SVM and RS, thereby offsetting their individual deficiencies, and is practicable and useful.