基于多源信息融合的矿岩可爆性评价及应用

在数据挖掘的基础上, 采用粗糙集对矿岩可爆性数据进行了数据级融合, 除去冗余属性, 然后采用BP神经网络进行特征级融合, 从而建立基于多源信息融合的矿岩可爆性评价模型。对原始数据进行了离散归一化处理, 应用粗糙集对决定矿岩可爆性指数的6个因素进行了属性约简, 剔除了平均合格率, 而保留了漏斗体积、大块率、小块率、岩体声波速度和波阻抗等5个因素, 并对约简的准确性进行了验证。分别建立了矿岩可爆性评价的BP神经网络模型和粗糙集-BP神经网络模型, 前者对矿岩可爆性指数的预测值与实际值的平均偏差为8.33%, 而后者为6.75%。利用建立的粗糙集-BP神经网络模型预测某矿山井下采场的矿岩可爆性指数为78.43, 计算出采场的炸药单耗为0.65 kg/m³, 而现场试验值为0.67 kg/m³, 进一步验证了该模型的正确性。

Evaluation of Rock Mass Blastability Based on Multi-source Data Fusion and its Application

After data mining, the redundant attributes were removed for rock mass blastability using data fusion with rough set theory. Then, BP neural network was used for information fusion in terms of feature, so as to build an evaluation model for rock mass blastability based on multi-source data fusion. After attribution reduction with rough set theory for 6 deciding factors of rock mass blastability index based on the discretion and normalization processing of the raw data, the attribute of average qualified percentage was deleted with attributes of crater volume, mass ratio of big rock blocks, ratio of small rock blocks and wave impedance retained. The accuracy of such reduction process has been verified. A BP neural network model and rough set-BP neural network model were respectively established for evaluating rock mass blastability, with the average deviation rate between predictive value and actual value at 8.33% and 6.75%, respectively. By using this established rough set-BP neural network model, the blastability index of rock mass in underground stope was predicated to be 78.43, and the unit consumption of explosive for the stope was calculated to be 0.65 kg/m³, which compared to the actual value of 0.67 kg/m³ from an on-site experiment, verified the model's validity.