论文部分内容阅读
本文介绍了一项初步试验工作，探讨神经网络方法的预测能力，考察用它作为储量估算工具的潜在可能性。设计试验方法时，要求能够与普通的克立格法和条件模拟法所得到的结果相互比较。本试验所选定的神经网络模型是四层前向债入-后向传递（four-layerfeed-forwardwithbackpropagation）模型，文中作了简要阐述。本研究所采用的数据集是从某生产矿山的一个水平中取得的样本值。神经网络方法的三次试验的输出结果列在文中，并和实际数据莫及作者之一在以前的研究工作中、用普通克立格法与条件模拟法得到的结果进行了比较。前期预试验结果表明：神经网络估值的整体精度随着输入向县的规模增大而增高。
This article presents a preliminary pilot study to explore the predictive power of neural network methods and investigate the potential of using it as a tool for estimating reserves. The design of the test method requires that the results be compared with those obtained by ordinary Kriging and conditional simulations. The neural network model selected in this experiment is a four-layerfeed-forwardwithbackpropagation model, which is briefly described in this paper. The dataset used in this study is the sample value obtained from a mine at a production level. The output of the three experiments of the neural network method is listed in the text and compared with the results obtained by the conditional simulation method in one of the authors who did not agree with the actual data in the previous research work. Pre-test results show that: the overall accuracy of neural network estimation increases with the size of input to the county.