深部煤矿多灾种安全评价方法与应用研究

我国煤矿将逐步转入深部开采，多灾种耦合是深部煤矿的新特征，而现有煤矿安全评价多是对浅层煤矿单一灾种的评价，开展深部煤矿多灾种安全评价研究具有十分重要的科学意义和广阔的应用前景。深部煤矿多灾种评价是个复杂巨系统，SVM适于解决复杂系统评价问题，但标准的SVM方法不擅长处理深部煤矿多灾种评价系统中复杂的多源次衍生不平衡样本数据，无法考虑不同致灾因子重要性差异和训练集样本模糊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多源次衍生不平衡样本数据处理方法，使用不同采样率对分类样本进行交叉欠采样；设计新的可变域的离差标准化公式，以提升传统数据规范化能力；在分类算法中引入模糊信息粒化方法和模糊转化函数来解决模糊不确定性问题，提出一种既能处理未确知信息又能给出绝对风险概率值的“情景-应对”型智能评价模型。新方法将丰富煤矿安全评价理论体系，推动深部煤矿多灾种评价方法从理论走向广阔的应用层面。

China's coal mines are gradually shifting to deep mining, while the coupling of multiple disasters is a new feature of deep coal mines. The existing coal mine safety evaluation is mostly the evaluation of single disaster in shallow coal mines. As a result, this study on safety evaluation of multiple disasters in deep coal mines has important scientific significance and broad application prospect in reducing mining catastrophes. The study will utilize support vector machine (SVM) to assess the complex system problems of deep coal mining safety evaluation. However, in the multi-disasters evaluation system of deep coal mines, the sample data obtained will be complex, multi-source, sub-derivative and unbalanced. In addition, uncertainty in the information evaluation caused by different disaster factors and the fuzzy uncertainty of training set samples are not fully considered which will be major system problems to analyze in standard SVM. Thus, this paper proposed a new multi-source, sub-derived and unbalanced sample data processing method, which uses different sampling rates to conduct cross-under-sampling of classified samples. To improve the normalization ability of traditional data, a new variable-field deviation standardization formula is designed. And the fuzzy information granulation method with the transformation function will be introduced into the classification algorithm to solve the fuzzy uncertainty problem. Thus, these new methods will enrich the theoretical system of coal mine safety evaluation and promote the application of multi-disaster evaluation system in deep coal mines.