关键词: 人工神经网络, Kohonen, 反向传播（BP）, 成矿预测

Abstract: With the transformation of ore-forming prediction from qualitative to quantitative，mathematical statistical methods and techniques are introduced into this field in recent years.Traditional statistical methods are applied under the assumption that the geological phenomena are homogeneous and a geo-scale relationship is the same with the others.But those geological conditions don’t exist in the practical application.Therefore，applying the non-linear science，which is not satisfied with the nature of the principle of linear superposition，such as neural network to metallogenic prognosis becomes the development direction.This paper combines the Kohonen clustering model and the BP prediction model，performs cluster analysis of the geochemical data of 1,444 mineral points in BaoGutu，and establishes the ore-forming prediction model.The prediction accuracy is 85.2%.The result shows that the proposed method is effective and feasible，which provides a new means for solving the ore-forming prediction.

Key words: artificial neural network, Kohonen, Back Propagation（BP）, ore-forming prediction