【摘要】 针对DBN(deep belief network)结构参数对特征提取性能影响很大而其影响特性不明导致最优参数难以选取的问题,以双转子系统不对中故障振动特征提取为例,通过构建不同DBN迭代次数、隐含层数、学习率及动量的重构误差曲线,分析DBN结构参数对双转子不对中故障特征提取性能影响特性。综合重构误差均方根误差RMSE(root mean square error)及计算时间,选取优化的结构参数。分析结果表明,最优参数为:迭代次数200次、隐含层节点数1 000～1 500、学习率0.01～0.1,动量项0.1～0.4。为DBN在双转子系统不对中故障识别中的应用提供结构参数选取依据。更多还原

【Abstract】 The structural parameters of deep belief network(DBN) have great influence on the feature extraction performance and the influence characteristic is not clear, so it is difficult to select optimal parameters. Aiming at the problem, the misalignment vibration feature extraction of dual-rotor system was taken as an example, the reconstruction error curves of different DBN iteration times, hidden layers, learning rate and momentum were constructed, and the performance of DBN structure parameters on the feature extraction of dual-rotor misalignment were analyzed. According to the root mean square error(RMSE) of reconstruction error and calculation time, the optimized structural parameters were selected. The results show that the optimal parameters are as follows: the number of iterations is set as 200, the number of hidden layer nodes is set as 1 000-1 500, the learning rate is set as 0.01-0.1, and the momentum is set as 0.1-0.4. This work provides a reference for the selection of structural parameters for the application of DBN in misalignment fault identification of a dual-rotor system.更多还原