有效的煤矿应急救援工作能够极大地降低人员伤亡和财产损失，开展煤矿应急救援能力评价研究具有十分重要的科学意义和广阔的应用前景。对此研究刚刚兴起，现有方法多是凭经验的主观评价，未能充分与煤矿灾害情景紧密结合。煤矿应急救援是个复杂巨系统，SVM 适于解决复杂系统的评价问题，但标准的SVM方法不擅长处理煤矿应急救援系统中正负类样本不平衡数据，无法考虑事故危重程度不同样本间重要性差异和训练集专家赋值不确定性带来的未确知信息评价等问题。本项目拟提出一种新的欠采样方法，通过对各样本点惩罚系数的柔性缩放，利用不同C值对新样本集进行训练，在分类算法中引入样本模糊隶属度，并在此基础上构造出新的未确知机会约束规划模型和算法，建立新的评价模型。新方法将在提高整体评价性能的同时降低少数类的误判率。研究将在一定程度上丰富和完善煤矿应急救援能力评价理论体系，推动煤矿应急救援能力评价方法从理论研究走向更广阔的应用层面。

Effective coal mine emergency rescue can reduce casualties and property losses, so research on coal mine emergency rescue capability evaluation has important scientific significance and broad application prospect. The relative research is just at a primary stage, so existing methods with subjective assessment depending on experience fail to fully integrate with specific situation of coal mine disasters. Support vector machine (SVM) applies to assessment problem of complex system such as the coal mine emergency rescue system. However, when positive and negative samples are imbalance in emergency rescue system; differences between different samples of accident danger degree are not fully considered; training sets are not certainty assigned by experts, standard SVM can not deal with those evaluation problems well. This paper proposed a new undersampling method. In this method, membership function is introduced into classification algorithm and different types of SVM are used for new sample set training in order that standard SVM method is converted to flexible control each sample's punish coefficient and reduce constraints. Then a new unascertained chance constrained programming, algorithm and a whole new assessment model is constructed. The new assessment method can improve overall evaluation performance at the same time reduce misjudgment rate of minority class. This research will be to a certain extent enrich and perfect coal mine emergency rescue capability evaluation theory, and promote it from theoretical research to a broader application level.