Abstract: Introduction of Multi-Agent parallel reasoning and independent solving mechanism can improve the accurate rate of fault diagnosis system.Feature extraction is used to reduce dimension and generate sub-tasks for Agents with corresponding abilities.However，in most cases the conclusions given by different diagnosis Agents will be different.In order to solve the problem，D-S evidence theory is applied to integrate conclusions given by multiple agents as multi-source evidence.Information credibility distribution matrix is introduced to denote Agent capacity in integration process.Finally，a diagnosis of examples is given to validate the method and simulation experiments verify the effectiveness of this method.

Key words: multiagent, feature dimension reduction, D-S evidence theory, credibility distribution matrix

摘要： 在故障诊断中引入多Agent并行推理、独立求解的机制能够提高诊断准确率。通过特征提取降低特征维数，形成子任务，并分配给具有专家能力的Agent处理。然而多数情况下不同Agent的诊断结论会存在差异，为解决该问题，引入D-S证据理论将各Agent的结论作为多源证据进行融合。在融合中引入可信度分配矩阵表征Agent能力。最后通过一个诊断实例说明了此方法，通过仿真对比实验验证了此方法的有效性。

关键词: 多Agent, 特征降维, D-S证据理论, 可信度分配矩阵