基于GA-SVM的露天矿抛掷爆破抛掷率预测

摘要

This paper probed into the whole height bench cast blasting process and described the influence factors from 3 major perspectives:natural geological,blasting scheming and factitious ones,and selected the throwing rate which was generally accepted in the cast blasting field to assess the blasting performance.Then a novel GA-SVM model was constructed to analyze the real collected explosion data from open pit mining,and verified in a certain open-pit.Also the MIV method was employed to analyze the influence factor at each input factor.The study indicate that:① the presented GA-SVM model performs more robust and accurate than other artificial intelligence models such as BP,RBF,GRNN and GA-BP,which has a more stable prediction accuracy of 83.75%.Moreover,due to the ubiquitous paradigm of the presented approach,it provides a single,unified approach to evaluating other blasting performance factors such as the longest thrown distance and loose coefficient etc;② for this certain open pit which maintains a steady lithological character and design parameters,the bench height,explosive specific charge possess a positive correlation coefficient with the throwing rate,while line of least resistance,the slope angle and the profile width perform the opposite.%分析了高台阶抛掷爆破的机理过程,并从自然地质、爆破设计和人为操作3个角度出发,结合某矿区的实际开采情况,提取其中10个参数作为影响该矿区抛掷爆破效果的主要因素,以爆破领域中广泛接受的抛掷率作为抛掷爆破效果的评价因子,采用此矿区爆破生产中的实际数据建立了基于遗传算法优化的支持向量机模型GA-SVM。基于建立的GA-SVM模型,采用平均影响值（Mean Impact Value,MIV）作为评价标准,对各因素的影响程度进行了评定。结果表明：①GA-SVM模型能够比较快速、准确地根据此矿区的爆破设计参数预测出抛掷爆破的抛掷率,平均预测精度稳定在83.75%,与其他智能算法如BP,RBF,GRNN相比,GA-SVM具有更好的鲁棒性和更佳的预测精度。由于计算流程的统一性和预测方法的普适性,GA-SVM模型对于其他抛掷爆破参数（如最远抛距、松散系数等）也具有良好的外推性;②对于此露天矿区而言,在其自然因素（如岩性等）和爆破设计因素（如炸药类型、起爆顺序、装药结构等）已确定的情况下,台阶高度、炸药单耗与抛掷率正相关,且台阶高度比炸药单耗对抛掷率的影响更大;而最小抵抗线、坡面角和剖面宽对于抛掷率呈现负相关,其他影响因素对于此露天矿抛掷率的影响较弱。