基于PSO改进的BP网络在爆破大块率优化中的应用  

Application of BP Network based on PSO Improved in Optimization of Blasting Boulder Yield

作　　者：赵国彦[1] 孙贵东[1] 戴兵[1] 陈英[1] 

机构地区：[1]中南大学资源与安全工程学院,长沙410083

出　　处：《爆破》2017年第2期15-19,39,共6页Blasting

基　　金：国家自然科学基金项目(51374244)

摘　　要：为解决地下矿山爆破开采采场大块率较高的问题,将PSO算法应用于BP网络中,生成PSO-BP模型对影响大块产生的主要参数进行优化。以参数孔底距、排距、一次炸药单耗、起爆位置为输入因子,大块率为输出因子建立PSO-BP模型,采用现场实测数据初步训练模型,通过控制变量法对模型参数的选取分别进行敏感性分析,得出最佳的大块率PSO-BP评价模型。增加模型各输入因子水平数,按L16(34)正交表组成优选样本,经评价模型的计算预测,搜索出最优的大块率影响参数值。研究结果表明:以东际金矿为例,采用孔底起爆方式,得出最佳大块率预测值9.98%,最优参数值是排距为1.6 m,孔底距为1.8 m,一次炸药单耗为0.350 kg/m^3。In order to solve the problem of high blasting boulder yield in underground mine stope, the PSO-BP model was produced to optimize the influence parameters of boulder yield by using PSO algorithm in the BP neural network. The parameters, such as hole-bottom spacing, row spacing, specific charge and detonation position, were used as the input data and the boulder yield was set to be the output data in model building. The best PSO-BP evaluation model of boulder yield can be obtained by using field test data to train model preliminarily and using the controlling variable method to analyze the sensitivity of model parameters. According to calculation and prediction of evaluation model, the best influence parameter values of boulder yield were obtained by increasing more levels of the input data to form the forecasted and optimized samples in the way of L16 （ 34 ） orthogonal array. Take Dongji Gold Mine as an example, the results show the best predictive value of boulder yield as 9.98%. In addition, by detonating at hole bot- tom, the best influence parameter values were obtained as row spacing 1.6 m, hole-bottom spacing 1.8 m and specific charge O. 350 kg/m3.

关 键 词：大块率 优化 PSO-BP模型 预测值 影响参数 

分 类 号：TD235[矿业工程—矿井建设]