聚变诊断数据分析的重要任务之一是通过反演算法从所测量数据中计算出相关物理量值。由于测量数据普遍存在系统和随机误差，这些误差会造成计算结果具有不确定性。对于聚变诊断，了解分析结果的不确定性对相关物理的研究工作具有重要意义。贝叶斯概率理论把带有不确定性的量以概率的形式表达从而具备良好的对解的不确定分析能力。贝叶斯图模型理论提供了一个针对不同聚变诊断系统以及待计算物理参数之间存在的复杂相互关系的有效分析工具。通过利用贝叶斯中的联合概率可以实现对多个不同诊断系统的集成分析，从而实现解的唯一性和一致性。该工作中的贝叶斯层析重构方法通过利用非平稳高斯过程解决等离子体芯部和边缘具有不同平滑度的问题，从而显著提高了计算能力并成功实现对等离子体芯部的磁流体力学不稳定性模的空间结构和位置的重构。

One of the main tasks of data analysis in fusion diagnostics is to infer the physical parameters of plasma from the measured data by inversion methods. The measured data are often subject to systematic and random errors which lead to the uncertainty of our results. Understanding the uncertainty is of importance to the study of relevant physics. Bayesian probability theory expresses all the quantities which bear uncertainty in the probabilistic form, thus has the strength in uncertainty analysis. Bayesian graphical model provides an effective tool to handle the complex interdependence between various diagnostic systems and the physical parameters to be inferred. A joint probability in the Bayesian theory enables an integrated analysis of many heterogeneous diagnostics to achieve the uniqueness and consistency of the solutions. For a Bayesian tomography method developed in this work, Non-stationary Gaussian Process has been adopted to take into account the different smoothness between plasma center and edge, thus can significantly improve the computational capacity as proved by the success in recovering the structure and position of MHD instabilities in plasma core.