露天矿卡车外排土场内部运距预测模型
A prediction model of the truck dumping haul distance in open-pit mine

为有效提高外排土场物料移运规划中运输功能耗模型的精度，以建立更为详细的排弃物料堆置次序优化、规划模型，针对年末排土计划中尚缺乏逐条带运距推估方法的问题展开研究，提出一种采用极限学习机算法（ELM）训练多元非线性运距曲线的预测模型，并将年末排土工程计划位置上已设计运输线路的排土条带作为训练样本，训练预测模型学习运距与影响因子间的时变特征，最终通过非线性运距表达推估待排物料块体的时变运距。为进一步增强ELM算法的预测精度，利用改进粒子群算法建立基于结构风险最小化的参数优化算法，改善了传统经验风险最小化算法的泛化能力，提高了算法预测精度。研究结果表明：采用模拟试算图解法最终确定ELM模型隐含层节点数为27；仿真测试中得出文中算法预测精度评价指标分别为均方误差0.006 8、拟合优度0.995 3、相对误差期望0.027%、绝对误差期望0.62、错估系数0.03、执行效率1.49 s；对比多组智能算法预测模型，其绝对误差分别0.116 2、0.094 7、0.139 1，其错估系数分别为0.230、0.200、0.266，算法明显具有更好的预测效果。
In order to effectively improve the accuracyof transportationfunction consumption models, open-pit designer can establish a more detailed material transportation planning model for theproblems that cannot be solved for lack ofestimation method of strip-by-strip transport distance in the annual plan. In this paper a prediction model of multivariate nonlinear haul distance curve trained by extreme learning machine was proposed. The dump strip on transport line designed for year-end dump project location was taken as the training samplesto train prediction model to learn the time varying trait of hual distance and influence factor. Finally, the nonlinear estimation of haul distance expression was used to predict block variable distance. In order to enhance the prediction accuracy of the ELM algorithm,the modified particle swarm optimization algorithm was adopted to build the model of parameters optimization aimed at structural risk minimization and realized the structural risk correction to improve the accuracy of prediction algorithm. The results show that the method of ELM model ultimately determine the number of hidden layer nodes to be 27 through the test of simulation by trial and graphic test.The evaluation indexesof algorithm prediction accuracy (mean square error, goodness of fit, relative error expectation, absolute error expectation, misestimation coefficient, execution efficiency) are 0.006 8, 0.995 3, 0.027%, 0.62,0.03 and 1.49 srespectively.Compared with other prediction model of intelligent algorithm,their absolute error are 0.116 2, 0.094 7, 0.139 1 and the coefficient of miscalculation are 0.230, 0.200, 0.266. In conclusion, the algorithm has better prediction effect obviously