The satellite power system is one of the core systems to ensure the normal on-orbit operation of satellite, and is also a representative of typical complex nonlinear systems. Autonomous prognostics and health management (A-PHM) is an inevitable trend in the future development of satellite, and autonomous fault diagnosis is a key part of A-PHM. Therefore, it is very necessary to carry out the research on autonomous fault diagnosis for satellite power system to improve the capability of satellite to perform its on-orbit tasks independently. Therefore, we put forward a feasible framework of soft decision-making to cope with this issue, mainly including attribute reduction, attribute weight assignment, rule extraction, and rule matching. Specifically, a neighborhood decision-theoretic rough set model (named DNDTRS) is first designed with the help of a data-driven loss function matrix for attribute reduction and weight assignment. Subsequently, the classification probability generated by DNDTRS is fed to a rule extraction model developed by the Takagi-Sugeno (T-S) fuzzy theory. Finally, a soft decision-making mechanism is proposed to execute the output after rule matching. In the experimental part, the proposed methodology is verified by the benchmark datasets and the fault data of satellite power system. The experimental results demonstrate the promised performance of our methodology.

卫星电源系统是确保卫星正常在轨运行的核心系统之一，也是典型复杂非线性系统的代表。自主预测和健康管理（A-PHM）是卫星未来发展的必然趋势，而自主故障诊断是A-PHM的关键部分。因此，有必要进行卫星动力系统自主故障诊断研究，以提高卫星独立执行在轨任务的能力。因此，针对这一问题，我们提出了一个可行的软决策框架，主要包括属性约简，属性权重分配，规则提取和规则匹配。特别，首先借助数据驱动的损失函数矩阵设计邻域决策理论粗糙集模型（称为DNDTRS），以进行属性约简和权重分配。随后，将由DNDTRS生成的分类概率输入到由Takagi-Sugeno（TS）模糊理论开发的规则提取模型中。最后，提出了一种软决策机制来执行规则匹配后的输出。在实验部分，通过基准数据集和卫星电力系统故障数据验证了所提出的方法。实验结果证明了我们方法的预期性能。DNDTRS生成的分类概率被输入到由Takagi-Sugeno（TS）模糊理论开发的规则提取模型中。最后，提出了一种软决策机制来执行规则匹配后的输出。在实验部分，通过基准数据集和卫星电力系统故障数据验证了所提出的方法。实验结果证明了我们方法的预期性能。DNDTRS生成的分类概率被输入到由Takagi-Sugeno（TS）模糊理论开发的规则提取模型中。最后，提出了一种软决策机制来执行规则匹配后的输出。在实验部分，通过基准数据集和卫星电力系统故障数据验证了所提出的方法。实验结果证明了我们方法的预期性能。