基于SVM的充填体强度与采场稳定性需求智能匹配研究

为解决矿山充填体强度的设计问题,提高矿山充填体的强度动态调整能力,本文通过调查国内百座矿山现场充填体强度实际数据,采用SVM方法建立充填体强度智能预测模型,对70组训练样本数据进行训练,采用BP神经网络模型与SVM模型的预测结果进行比较。结果表明：SVM预测模型的最大误差为3.52%,平均误差为2.41%。BP预测模型的最大误差为10.98%,平均误差为7.01%。SVM模型比BP模型预测精度更高,误差更小。采用SVM模型对三山岛金矿充填体强度进行预测,一步骤回采矿房充填体强度1.02MPa,推荐灰砂比1:12,二步骤回采矿房充填体强度0.86MPa,推荐灰砂比1:16。现场采场充填效果良好,未发生充填体失稳现象。基于SVM的充填体强度智能匹配模型能够在满足采场稳定性的前提下,减少充填成本,提高矿山的经济效益。

Research on Intelligent Matching of Backfill Strength and Stope Stability Demand Based on SVM

In order to solve the design problem of mine backfill strength and improve the adjustment ability of mine backfill strength, the SVM intelligent prediction model of backfill strength was established by investigating the actual data of backfill strength of 100 mines in China. The BP neural network model was compared with the predicted results of the SVM model by training the data of 70 samples. The results showed that the maximum error of the SVM prediction model was 3.52% and the average error was 2.41%, the BP prediction model was 10.98%, and the average error was 7.01%. The SVM model had higher prediction accuracy and less error than the BP model. The SVM model was used to predict the backfill strength of Sanshandao Gold Mine, the strength of the backfill in the mining room was 1.02MPa in one step, the recommended sand-cement ratio was 1:12, and the strength of the backfill in the second-step mining room was 0.86MPa, the recommended sand-cement ratio was 1: 16. The filling effect of the on-site stope was good and no backfill body instability occurred. The intelligent matching model of the backfill strength based on SVM can reduce the filling cost and improve the economic benefit of the mine under the premise of meeting the stability of the stope.