【摘要】 基于人工智能中状态空间的启发式搜索技术，提出了一种全新的大地电磁测深三维反演新思路。其基本思想是把模型空间量化成有限模型集，而后按一定的搜索策略选择一个最有希望的模型进行正演计算，再把正演结果与观测数据比较并计算目标函数，如果目标函数小于给定值，则此时的模型即为反演结果，否则继续搜索、正演、比较，不断重复此过程，直到找到一个满意的解估计。作为一种通用的反演新思路，不但适用于大地电磁测深，还可以把该反演技术应用到其它各种地球物理反演问题中。更多还原

【Abstract】 We give in this paper a new train of thought for 3-D magnetotelluric (MT) sounding data inversion. Its basic idea is that the model space is quantified into finite model sets, then select a most prospective model according to a certain searching tactics to perform forward computation, compare the forward results with the observed data and compute the objective function; if the function value is less than a given value, then the model is the inversion result, and if not, continue searching, forward com-putation and comparing repeatedly until a satisfactory solution estimate is found. As a new common in-version idea, it is not only suitable to MT sounding data inversion, but also can be appliable to other geo-physical inversion problems.更多还原