二维大地电磁资料的差分蚁群法反演

Two-dimensional magnetotelluric inversion using differential ant-stigmergy algorithm

差分蚁群算法是一种有效的多维函数优化算法。它将每一维变量对应的连续取值的步长离散成有限多个差分步长;然后将这些离散的差分步长和图顶点建立起对应关系;进一步利用柯西分布在图顶点上分配信息素,这样蚂蚁就可以按照信息素的浓度随机选择步长。将此算法应用到二维大地电磁资料的反演中,理论和数值实验结果表明,差分蚁群算法不受初始模型的限制,不仅能够很好地处理无噪声下的反演问题,用于实测资料的处理也取得了较好的效果。

The differential ant-stigmergy algorithm is a kind of efficient algorithm to deal with multi-dimension function optimization.Each of the one dimensional variables with continuous values is discretized first into limited differential steps.And then the corresponding relationships of these discretized differential steps to these vertices of graph are setup.Further pheromones can be distributed on the vertices based on the Cauchy distribution.Therefore ants can randomly select steps according to the pheromone concentration.This algorithm is applied to two-dimensional magnetotelluric data inversion for testing various models.Numerical experiments have shown that the differential ant-stigmergy algorithm is less affected by initial model,and obtain reasonable inversion results.Similar results are obtained on real data test.