%0 Journal Article %T 一种基于关联熵融合与改进回声状态网络的故障预测方法 %A 王浩天 %A  %A 段修生 %A 单甘霖 %A 孙健 %A 邱锦涛 %J 振动与冲击 %D 2019 %X 针对液压泵退化特征维数高、预测效果不佳的问题,本文提出了一种基于关联熵融合与改进ESN的故障预测方法。首先,以关联熵为基础,构建特征层级融合算法,以充分利用所提取的退化特征信息,在提高特征简洁度的基础上,进一步改善预测特征性能;在此基础上,对ESN储备池结构进行优化,重新定义邻接矩阵元素取值,建立改进的ESN预测模型,以改善网络的泛化能力,提高预测精度。最后,通过对液压泵性能退化试验的应用分析,验证了该方法的有效性。</br>Abstract:To deal with the shortcomings of high dimensional features and unsatisfying fault predicting effect in conventional hydraulic pumps fault diagnosis methods,a novel prognostic method based upon the relative entropy fusion and modified echo state network(ESN) was proposed.On the basement of relative entropy,a features fusion algorithm was presented for making full use of degradation feature informations for the sake of improving the prognostic feature performance based on the simplifying of features.Furthermore,the structure of the reserve pool in ESN was modified,and neighboring matrix elements were redefined.Then,a modified ESN predicting model was established to raise the network performance and increase the predicting accuracy.The proposed method was verified by its application in hydraulic pump performance degradation experiments. %K 故障预测 %K 关联熵 %K 信息融合 %K 回声状态网络< %K /br> %U http://jvs.sjtu.edu.cn/CN/abstract/abstract8218.shtml