针对深度Q学习中单个Agent对行星齿轮箱进行故障诊断时出现的识别率低问题，利用多个Agent进行策略学习，结合模糊积分对多Agent的决策结果进行融合，提出了基于深度Q学习和模糊积分的行星齿轮箱故障诊断方法。对振动信号进行连续小波变换(continuous wavelet transform，CWT)、S变换(S transform, ST)得到相应的时频特征矩阵，随后利用原始时域数据和得到的时频特征矩阵构建多域环境状态空间，以与多个Agent交互；交互过程中将环境返回的原始时域特征、CWT时频特征，ST时频特征分别作为相应Agent的状态，通过深度Q学习算法最大化每个Agent的Q函数值，得到最优策略；采用模糊积分对多Agent的决策结果进行融合得到最终诊断结果。通过行星齿轮箱故障数据进行验证分析，经过模糊积分进行决策融合的诊断结果优于单个Agent，能够有效提高诊断精度。

Aiming at the problem of low recognition rate of single-agent in deep Q-learning for fault diagnosis of planetary gearbox, using multi-agent for strategy learning and fuzzy integral for fusing decision results of multiple-agent, a planetary gearbox fault diagnosis method based on multi-agent deep Q-learning and fuzzy integral was proposed.Firstly, the continuous wavelet transform (CWT) and S transform (ST) were performed for vibration signals to obtain the corresponding time-frequency feature matrix, and then using the original time domain data and the obtained time-frequency feature matrix, a multi-domain environmental state space was constructed to interact with multi-agent.Secondly, the original time-domain features, CWT time-frequency features and ST features returned by the environment space were taken as the corresponding agent state in interaction process.The Q function value of each agent was maximized with the deep Q-learning algorithm to obtain the optimal strategy.Finally, fuzzy integral was used to fuse the decision results of multi-agent, and obtain the final diagnosis results.The fault data of planetary gearbox were used to do verification and analysis.It was shown that the proposed planetary gearbox fault diagnosis method based on multi-agent deep Q-learning and fuzzy integral can effectively improve the diagnosis accuracy; its diagnosis results are better than those of the method based on single-agent deep Q-learning.