基于BP神经网络的矿石非计划贫化预测

为了消除等效线性超挖（Equivalent Linear Overbreak Slough，ELOS）经验图表法估算矿石非计划贫化的局限性，采用BP神经网络算法，以采场稳定指数、水力半径、钻孔平均偏斜量和炸药单耗为输入变量，以量化矿石非计划贫化的等效线性超挖深度为输出变量，建立了隐含层神经元节点数为6的3层BP神经网络预测模型。经过120组样本数据模型训练和样本测试，BP神经网络预测模型的拟合度为0.987 42、均方误差为9×10-5，预测的相对误差约6%，形成了矿石非计划贫化预测方法。应用BP神经网络非计划贫化模型对三道桥铅锌矿试验采场进行了矿石非计划贫化计算。结果表明：基于BP神经网络的矿石非计划贫化计算值为0.717 m，与现场实测值（0.7 m）相比，其相对误差为2.4%，优于经验图表法和数值模拟分析法的计算结果（0.80 m和0.55 m），可用于实际矿山的矿石非计划贫化预测。

Prediction of Unplanned Ore Dilution Based on BP Neural Network

In order to eliminate the limitation of equivalent linear overbreak slough （ELOS） empirical graph method in estimate unplanned ore dilution，BP neural network algorithm was applied to establish the three-layers BP neural network prediction model with 6 hidden layer neurons by considering the modified stable number of stope， hydraulic radius， average deviation of borehole and powder factor as input variables and ELOS of quantified unplanned ore dilution as output variables. The training and testing of model was performed by 120 group samples， the fitting degree of the prediction model was 0.987 42， and the mean square error was 9×10-5， the relative error of the forecast was about 6%.The prediction method of unplanned ore dilution was proposed.The calculation of unplanned ore dilution was applied in the test stope of Sandaoqiao Pb-Zn Mine， the calculation results showed that the calculated value of the unplanned ore dilution based on BP neural network was 0.717 m， and the relative error between the calculated results and the measured results 0.7 m was 2.4%， which was better than the results of empirical graph method and numerical simulation analysis （0.80 m and 0.55 m）， BP neural network model could be applied to predict the unplanned ore dilution in actual mines.