参数优化GA-ELM模型在露天煤矿抛掷爆破的预测

发布时间：2018-07-31 17:17

【摘要】：为有效指导露天煤矿制定正确生产计划,提高露天煤矿抛掷爆破预测的准确率,在分析露天煤矿抛掷爆破影响因素的基础上,通过"试错法"确定模型最优隐含层节点参数,进而提出一种参数优化后遗传算法(GA)和极限学习机(ELM)相结合的抛掷爆破预测模型。选取网络的输入输出相关参数,针对现有ELM输入权值矩阵和隐含层偏差,采用遗传算法对其进行优化选择;利用某露天煤矿抛掷爆破监测数据对该模型进行实例分析,并将RBF,BP,SVM,GA-BP模型预测结果与该模型进行对比分析;并引入Weibull模型,通过预测控制参数ɑ,β模拟爆堆形态。研究结果表明:(1)通过"试错法"确定GA-ELM模型最优隐含层节点数为39,有效降低系统的仿真误差,该参数下仿真误差值为0.137 7;(2)相较于传统ELM预测模型,通过遗传算法优化后,有效抛掷率,松散系数以及抛掷距离均得出更小的均方误差MSE值(0.258 0,1.748 5×10-4,3.618 4)和更高的决定系数R2值(0.986 4,0.995 3,0.970 6),改进后的GA-ELM具有更好的拟合效果和泛化能力;(3)通过与其他智能算法如BP,RBF,SVM,GA-BP相比,改进后的GA-ELM测试结果(均方误差,决定系数,仿真误差)明显优于其他预测模型,有效提高预测精度;(4)利用训练完成的GA-ELM网络预测爆堆形态时,控制参数a,β的预测误差均未超过5%,预测爆堆曲线接近真实爆堆曲线。
[Abstract]:In order to effectively guide the opencast coal mine to formulate the correct production plan and improve the accuracy of prediction of throwing blasting in opencast coal mine, on the basis of analyzing the influencing factors of throwing blasting in opencast coal mine, the optimal parameters of hidden layer node in the model are determined by "trial and error method". Furthermore, a throwing blasting prediction model based on the combination of genetic algorithm (GA) and extreme learning machine (ELM) after parameter optimization is proposed. The input and output parameters of the network are selected and the genetic algorithm is used to optimize the existing ELM input weight matrix and hidden layer deviation, and the model is analyzed by using the monitoring data of throwing blasting in an opencast coal mine. The prediction results of the model are compared with that of the model, and the Weibull model is introduced to simulate the shape of the explosion reactor by predicting the control parameters. The results show that: (1) the optimal hidden layer node number of GA-ELM model is determined by "trial and error method" to be 39, which effectively reduces the simulation error of the system, and the simulation error under this parameter is 0.137 7; (2) compared with the traditional ELM prediction model, the simulation error is optimized by genetic algorithm. Effective throwing rate, The loose coefficient and throw distance obtained smaller mean square error (MSE) value (0.258 脳 10 ~ (-4) 脳 10 ~ (-4) and higher decision coefficient (R ~ (2) (0.986 40.995 ~ 3 ~ (0.970 6). The improved GA-ELM has better fitting effect and generalization ability. (3) compared with other intelligent algorithms, such as BPP-RBFMA-SVMA-BP, the improved GA-ELM has better fitting effect and generalization ability. The improved GA-ELM test results (mean square error, decision coefficient, simulation error) are obviously superior to other prediction models, and effectively improve the prediction accuracy. (4) when using the trained GA-ELM network to predict the shape of explosion reactor, The prediction error of the control parameters a and 尾 is not more than 5, and the predicted pile curve is close to the real one.
【作者单位】：辽宁工程技术大学系统工程研究所;辽宁工程技术大学工商管理学院;
【基金】：国家自然科学基金资助项目(71371091) 辽宁省社科基金资助项目(L14BTJ004)
【分类号】：TD824.2

【相似文献】