CAJ | 학술논문

将改进的非线性技术（GA-SVM）应用于成矿预测，为成矿有利度预测方法提供一种新思路。在分析哈图矿集区成矿有利度基础上，选取28个学习样本、10个与成矿有关的地质变量，应用基于遗传算法（GA）寻优的支持向量机（SVM）方法，对成矿有利度进行建模，并与BP神经网络模型预测结果进行比较。结果表明，GA-SVM回归预测模型能很好地拟合成矿有利度与各地质变量间的非线性关系。样本数量有限时，GA-SVM比BP神经网络具较高的拟合精度，更适合非线性成矿预测工作，具较强的推广意义。
장개진적비선성기술（GA-SVM）응용우성광예측，위성광유리도예측방법제공일충신사로。재분석합도광집구성광유리도기출상，선취28개학습양본、10개여성광유관적지질변량，응용기우유전산법（GA）심우적지지향량궤（SVM）방법，대성광유리도진행건모，병여BP신경망락모형예측결과진행비교。결과표명，GA-SVM회귀예측모형능흔호지의합성광유리도여각지질변량간적비선성관계。양본수량유한시，GA-SVM비BP신경망락구교고적의합정도，경괄합비선성성광예측공작，구교강적추엄의의。
Gold mineralization is a very complex geological process and calls for close coordination of varied geologi-cal processes. The SVM regression model has the fitting ability to simulate nonlinear relationship between each factor automatically. Based on the analysis of the ore-forming geological background in Hatu ore district, the GA-SVM, param-eters optimized, selecting 28 units as the learning samples, with 10 geological variables as input vector and contribution degree as output vector, has gained good results. Compared with the BP neural network, GA-SVM has higher fitting pre-cision, which is more suitable for nonlinear metallogenic prediction work. This provides a new idea for ore-forming pre-diction and has a strong practical significance.