论文部分内容阅读
求解反演问题最常见和最灵活的方法是把它当成贝叶斯(Bayesian)推理中的问题来推导。这种方法允许我们把具体反演问题的多种不同类型的信息合并成为单一计算问题。我们可以把哪些有关模型特征是合理的,哪些是不合理的先验信息,有关观测数据及其误差的信息,以及有关数值误差和理论误差的信息等综合起来,由此预测给定模型可能产生的数据。在所有模型空间范围内,我们以所谓后验概率的标量数值函数为最终结果。在一
The most common and flexible way to solve an inversion problem is to deduce it as a problem in Bayesian reasoning. This approach allows us to combine multiple different types of information that specifically inversed problems into a single computational problem. We can synthesize which information about model features is reasonable and which are unreasonable, the information about the observed data and its errors, the information about numerical errors and theoretical errors, and so on, so as to predict that a given model may be generated The data. In all the model space, we use the so-called posterior probability scalar numerical function as the final result. In a