针对人为选定参数造成神经网络故障诊断性能不稳定问题与麻雀搜索算法（sparrow search algorithm，SSA）种群初始化时由于随机性造成寻优范围缩小和算法容易陷入局部最优等问题，采用反向学习（opposition􀆼based learning，OBL）对SSA算法中麻雀种群初始化过程进行优化，扩大搜索范围，并结合随机游走策略（random walk，RW）对寻优过程中的最优麻雀施加扰动，提高算法的局部搜索能力，降低算法陷入局部最优的风险。在此基础上，采用基于反向学习和随机游走策略的麻雀搜索算法（sparrow search algorithm based on opposition-based learning and random walk，BRWSSA）优化门控循环单元（gate recurrent unit，GRU）的隐含层节点个数，设计了一种基于BRWSSA-GRU的发动机滑油系统故障诊断模型。为了验证所设计的故障诊断模型的有效性，还设计了GRU和SSA-GRU两种故障诊断模型。最后，采用相同的滑油系统数据集对GRU、SSA-GRU和BRWSSA-GRU3种不同的故障诊断模型的有效性进行了对比试验验证。结果表明，提出的BRWSSA-GRU故障诊断模型的诊断准确率明显优于GRU和SSA-GRU方法，BRWSSA-GRU故障诊断模型的有效性得到验证。

In response to the problem of unstable fault diagnosis performance of neural network caused by artificially selected parameters，as well as the problems of narrowing the optimization range and falling into the local optima caused by the randomness of sparrow search algorithm （SSA） population initialization，opposition-based learning （OBL） was used to optimize the initialization process of sparrow population in SSA algorithm and expand the search range. Combined with the random walk strategy （random walk，RW），the optimal sparrow in the optimization process was disturbed to improve the local search ability of the algorithm and reduce the risk of the algorithm falling into local optimum.On this basis，an improved BRWSSA algorithm was used to optimize the number of hidden layer nodes of gate recurrent unit （GRU），and a fault diagnosis model of engine oil system based on BRWSSA-GRU was designed. In order to verify the effectiveness of the fault diagnosis model，two fault diagnosis models，GRU and SSA-GRU，were also designed. Finally，comparative experiments were conducted to validate three different fault diagnosis models，GRU，SSA-GRU，and BRWSSA-GRU using the same lubricating oil system dataset. The results show that the diagnostic accuracy of the proposed BRWSSA-GRU fault diagnosis model is obviously better than that of GRU and SSA-GRU methods，which verifies the effectiveness of the designed BRWSSA-GRU fault diagnosis model.