GA-SVM和神经网络组合模型预测充填钻孔寿命

 充填钻孔是充填料浆从地表输送到井下采场的咽喉工程，是矿山正常运转的保障，因此对矿山充填钻孔使用寿命进行预测十分重要。通过建立支持向量机（SVM）和BP神经网络组合预测模型，用训练集对模型进行训练，以验证集预测值的均方误差作为SVM适应度函数，通过遗传算法（GA）对SVM模型参数进行优化选择，应用优化得到的SVM模型进行预测，并结合BP神经网络进行残差修正，最终得到预测结果。以某矿为例，通过GA得到SVM模型最优参数：适应值（均方误差mse）=0.0111，惩罚系数C=47.0768，核函数参数σ=2.2638。通过优化的SVM模型，对预测集充填钻孔寿命进行预测，经BP神经网络残差修正，预测结果的相对误差均控制在3%左右。对比单一预测模型，组合预测模型预测结果更加理想，精度更高，在类似的预测工程中有良好的推广价值。

Prediction of Backfill Drill-hole Life Based on Combined Model of GA-SVM and Neural Network

The backfill of a drill-hole is a throat engineering process in which the filling slurry is transported to the underground stope from the surface to ensure the safety of the mine normal operation. To predict the service life of the mine backfill drill-hole, a combination forecasting model of the Support Vector Machine (SVM) and the BP neural network is established in this paper. The mean square error of the value is taken as a fitness function of the SVM. Then, the SVM model parameters are optimized through the Genetic Algorithm(GA). Then, the optimized SVM is applied to predict the prediction set. The final forecast result is obtained by means of the revision of the residual error through the BP neural network. A certain mine is taken as an example, its drill-hole life is predicted through the combination forecasting model, and the optimal parameters are obtained. The adaptive value (mean square error mse) is 0.0111; the penalty coefficient C is 47.0768; the kernel function parameter σ is 2.2638. The accuracy of the model is analyzed. The relative error of the predicted results is about 3%. Compared with the single prediction model, the combination forecasting model enjoys a higher accuracy.