基于过采样支持向量机的煤与瓦斯突出预测

Prediction of Coal and Gas Outburst Based on Over-sampling Support Vector Machine

基于机器学习的煤与瓦斯分类预测方法中,各突出案例的数量不平衡会导致预测准确率降低。为了提升煤与瓦斯突出预测模型的准确率及稳定性,构建了过采样算法和支持向量机(support vector machine,SVM)组合的分类预测模型。首先,通过聚类分析将突出样本分成多个簇,在每个簇中对可能的噪声点按概率去除;然后通过过采样算法合成新样本,以减少样本数量不均衡对模型训练的影响;最后,用支持向量机模型结合粒子群算法对新数据集进行训练调优。实验结果表明:提出的模型在G-mean、曲线下面积(area under curve,AUC)值上均高于传统的分类模型,具有更强的算法鲁棒性,并且随着突出样本数量的减少,其优势更加明显。

The imbalance of the number of outstanding cases lead to a decrease in the accuracy of prediction in the coal and gas classification prediction method based on machine learning.A classification prediction model combined with an oversampling algorithm and support vector machine(SVM)was constructed to improve the accuracy and stability of the coal and gas outburst prediction model.Firstly,the prominent samples were divided into multiple clusters through cluster analysis,and possible noise points were removed according to the probability in each cluster.Then the new samples through an oversampling algorithm to reduce the impact of imbalanced sample numbers on model training were synthesized.Finally,the support vector machine model combined with the particle swarm algorithm was used to train and tune the new data set.The experimental results show that the proposed model is higher than the traditional classification model in G-mean and area under curre(AUC)values,and has stronger algorithm robustness,and its advantages become more obvious as the number of prominent samples decreases.