基于PCA-ABBP强分类器的矿区采空塌陷危险性预测

Prediction of Underground Goaf Collapse Risk Based on PCA-ABBP Strong Classifier

为了准确、快速地预测矿区采空塌陷危险性,针对矿区采空塌陷影响因素之间存在信息重叠以及利用单一BP神经网络进行预测时存在的局部极值等问题,提出了一种PCA-ABBP强分类器模型。以北京西山某地的24组采空塌陷数据为样本,选取了采空区空间叠置层数等7个变量作为矿区采空塌陷的影响因素,以前17组数据作为训练样本,建立基于PCA-ABBP强分类器的矿区采空塌陷危险性预测模型。利用该模型对后7组数据进行预测,预测结果与实际完全相符,而单一BP神经网络预测的平均误差为17.14%,验证了所提出模型的有效性和可靠性。

In order to forecast the underground goaf collapse risk accurately and quickly,a PCA-ABBP strong classifier model is proposed. And the proposed method is mainly focus on the problems of information overlap between underground goaf collapse influencing factors and defects of single BP neural network. Based on 24 historical collapse information of Beijing Xishan Mine,and seven variables,such as the number of overlapping layers,are chosen as the influencing factors. The PCAABBP strong classifier of underground goaf collapse risk is established using the first 17 training samples. On this basis,the last 7 samples are predicted using the established model,and the predicted results are in complete agreement with the actual results. However,the average error of single BP neural network prediction is 17. 14%,which verifies the validity and reliability of the proposed model.