论文部分内容阅读
本文阐述了用于矿床品位估算的GEMNet（品位估计用映射网络）人工智能系统的开发过程。开展这一工作的主要目的之一是探讨把神经网络作为估算品位及其它矿床空间变异特征的工具。神经网络所具有的一些特征表明它适于用来估算品位。神经网络具有学习实例并加以归纳推广的能力（根据有限个钻孔样，预测整个矿床的品位值）。神经网络也能向先例学习，得到训练以解决某一问题。它们并不需要明确地程序化（也就是说，系统的准确性极大地依赖于从矿床中获得的先例的有效性，而不依赖于建立矿床模型时所作出的假设）。神经网络具有逼近定义域之间的复杂映射关系的能力，仅需要以映射的先例为基础（在本文中是指矿床中空间位置与品位值之间的映射或关系）。本文阐述了GEMNet系统的主要组成部分及其运行方法，同时给出以某铁矿床为基础应用于小规模问题的实例。
This article describes the development of GEMNet (Mapping Network for Grade Estimation) artificial intelligence systems for ore grade estimation. One of the main purposes of this work is to explore the use of neural networks as a tool for estimating the grade and other spatial variations of deposits. Some features of neural networks show that it is suitable for estimating grade. The neural network has the ability to learn examples and generalize them (based on a limited number of boreholes, predict the grade of the entire deposit). Neural networks can also learn from precedents and get trained to solve a problem. They do not need to be explicitly programmed (that is, the accuracy of the system relies heavily on the validity of the precedent obtained from the deposit without relying on the assumptions made in establishing the deposit model). Neural networks have the ability to approximate complex mapping relationships between domains and need only be based on mapping precedents (in this context, the mapping or relationship between spatial and grade values in the deposit). This paper describes the main components of the GEMNet system and how it works, and gives examples of the application of a certain iron deposit to small-scale problems.