论文部分内容阅读
针对岩爆现象发生的不均衡及发生机理受多因素影响的问题,在分析重取样技术的基础上,设计并实现了自适应选择近邻的混合重取样算法,并将其用于岩爆危险性预测.该方法结合过取样和欠取样方法的优势,改进了SMOTE过取样算法在产生合成样本过程中存在的盲目性及只能复制生成数值属性的问题,新算法能根据实例样本集内部分布的真实特性,自适应调整近邻选择策略,对不同属性的数据采取不同的复制方法生成新的少数类实例,控制和提高合成样本的质量;并通过对合成之后的数据集,用改进的邻域清理方法进行适当程度欠取样,去掉多数类中的冗余实例和边界上的噪音数据,减少其规模,在一定程度上达到相对均衡,从而,可有效地处理非均衡数据分类问题,提高分类器的性能.该算法在VCR采场岩爆实例上进行实验,预测的结果与实际情况完全一致,表明在工程实例岩爆危险性实例数据非均衡情况下实施混合重取样方案是可行的,预测准确率高,具有良好的工程应用前景.采用该方法可找到岩爆发生的主控因素,为深部开采工程的合理设计与安全施工提供科学依据.
In view of the unbalanced rockburst phenomenon and the multi-factor influencing mechanism, a hybrid resampling algorithm is designed and implemented based on the analysis of resampling technique. The algorithm is applied to the rock burst risk This method combines the advantages of oversampling and undersampling methods to improve the blindness of SMOTE oversampling algorithm in generating composite samples and the problem of generating only numerical attributes. The new algorithm can be based on the internal distribution of the sample set Real characteristics, adaptive adjustment of neighbor selection strategy, different replication methods for different attributes of data to generate a new minority instances, control and improve the quality of the composite samples; and through the synthesis of the data set, with improved neighborhood cleanup This method can reduce the sample size and reduce the size of the redundant data in most classes, reduce the size of the data and reduce the size of the data, so that the problem of unbalanced data classification can be effectively solved, Performance.The experiment is carried out on the case of VCR stope rockburst and the prediction result is completely consistent with the actual situation, It is feasible to carry out the mixed resampling scheme under the case of unbalanced rockburst risk data with high prediction accuracy and good prospect of engineering application.The main controlling factors of rockburst occurrence can be found by this method and is suitable for the deep mining project Design and construction to provide a scientific basis for safety.