基于GA-SVM回归的成矿有利度预测方法探讨

Discussion on Prediction Method for the Contribution Degrees to the Metallogenesis Based on GA-SVM Regression

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。

Gold mineralization is a very complex geological process and calls for close coordination of varied geological 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, parameters 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 precision, which is more suitable for nonlinear metallogenic prediction work. This provides a new idea for ore-forming prediction and has a strong practical significance.