局部加权随机森林的冲击地压危险性等级预测

Prediction of rock burst grade based on locally weighted random forest

为对煤矿冲击地压危险性等级进行预测,综合考虑煤层厚度、煤层倾角、开采深度、顶板岩性、构造情况、开采方法、有无煤柱、采煤工艺等影响因素.采用局部加权学习方法建立冲击地压危险性等级预测模型,其中分类器选择随机森林,样本间距离采用欧氏距离函数进行计算.实验选取17组冲击地压数据进行研究,其中14组数据用于建立预测模型,采用十折交叉验证法对模型进行评价,并与采用决策树和朴素贝叶斯生成的模型进行比较,预测准确率得到较大提高,最后使用该模型对其它3组测试数据进行预测,预测结果与实际类别吻合.研究结果表明:采用局部加权随机森林方法可以建立泛化性能更好的冲击地压预测模型.

In order to predict the risk grade of rock burst, thickness of coal seam, dip angle of coal seam, mining depth, roof lithology, tectonic conditions, mining method, coal pillar and mining craft were considered as the influence factors. Locally weighted learning （LWL） method was selected to build the prediction model, in which random forest was selected as the classifier, and Euclidean distance function was selected to compute the distance of the samples. 17 groups of samples were chosen and 14 in which were trained to create the prediction model. Then it was compared with those created by Decision Tree and Naive Bayes using 10-fold cross-validation. It showed that the forecast accuracy had been greatly improved.Then the rest of the samples were predicted by the model, and the classification results were according with the actual. The resutel of study shows that locally weighted random forest can build much better modle with high generalization performanace.