摘要
矿产靶区预测是一种从统计单元集合中识别找矿靶区的非线性模式识别过程,可以利用Boltzmann机能够对外部刺激进行编码和重建的功能,实现基于Boltzmann机的矿产靶区非线性统计预测过程.鉴于此,笔者定义了面向矿产靶区预测的三层Boltzmann机模型,模型输入层神经元数目等于找矿证据数目,输出层只有一个神经元,隐藏层神经元数目由用户根据矿产靶区预测的精度要求确定;模型应用Hebbian编码和模拟退火算法相结合的随机学习算法进行训练,根据学习训练后模型输入层与隐藏层神经元之间的连接权确定找矿证据的权系数;根据证据权系数和统计单元证据组合特征计算单元成矿有利度,圈定找矿靶区.在GDAL数字图像输入输出函数库基础上,用VC++语言开发了面向栅格数据的矿产靶区预测Boltzmann机算法程序并应用于新疆阿勒泰地区的矿产靶区预测研究.结果表明,Boltzmann机模型预测的统计单元成矿有利度能够正确反映研究区已知矿床(点)的空间分布规律,因此,基于Boltzmann机的矿产靶区非线性统计预测模型是有效的.
Mineral target prediction is a nonlinear pattern recognition procedure for differentiating mineral target cells from a geological statistical cell set. This kind of nonlinear pattern recognition procedure can be implemented with Boltzmann machines due to that Boltzmann machines can encode and reconstruct external stimuli. In view of this, the authors construct a three-layered goltzmann machine model for mineral target prediction. The three layers are called input layer, hidden layer and output layer, respectively. The number of neural cells in the input layer is same as the number of evidential map patterns. The output layer has only one neural cells. The number of neural cells in the hidden layer is defined by a user according to the resolution of mineral target prediction. A stochastic learning rule integrating Hebbian encoding and simulated anealing algorithms is applied to train the model. The connecting weight coefficients between the input layer and hidden layer of a trained model are used to compute the weights of evidential map patterns. The ore-bearing favorability of a statistical cell can be determined according to the weight coefficients of map patterns and the map pattern combination in the cell. The mineral target cells can be delineated according to their ore-bearing favorabilities. A VC＋ ＋ program for raster data oriented mineral target prediction with Boltzmann machines has been developed on the basis of GDAL, a Cq- q- library for the input and output of digital image data. The model has been experimentally applied to the mineral target prediction in Ahay, northern Xinjiang. The experimental result illustrates that the predicted areas with high ore-bearing favourabilities coincide with the known mineral occurrences in the study area. Thus, the nonlinear prediction model based on Boltzmann machines is feasible for mineral target prediction.
出处
《地球物理学进展》
CSCD
北大核心
2012年第1期179-185,共7页
Progress in Geophysics
基金
国家自然科学基金项目(41072244)资助
关键词
Boltzman机
模拟退火
矿产资源
矿产勘查
靶区预测
Boltzmann machine
simulated anealing
mineral resource
mineral exploration
target prediction
