基于收缩极限学习机的故障诊断鲁棒方法

为降低特征噪声对分类性能的影响,提出一种基于极限学习机(extreme learning machine,ELM)的收缩极限学习机鲁棒算法模型(CELM)。采用自编码器对输入数据进行重构,将隐层输出值关于输入的雅克比矩阵的F范数引入到目标函数中,提取出更具鲁棒性的抽象特征表示,利用提取到的新特征对常规的ELM层进行训练,提高方法的鲁棒性。对Mnist、UCI数据集、TE过程数据集以及添加不同强度的混合高斯噪声之后的Mnist数据集进行仿真,实验结果表明,提出的方法较ELM、HELM具有更高的分类精度和更好的鲁棒性。

Contractive-ELM based robust method for fault diagnosis

To reduce the influence of feature noise on classification performance,a contractive-ELM robust algorithm based on extreme learning machine(ELM)was presented.The input data were reconstructed using the autoencoder,the Frobenius norm of the Jacobian matrix of the hidden layer output about the input was introduced into the objective function,and the abstract feature representation with more robustness was extracted.The new features extracted were used to train the conventional ELM layer to improve the robustness of the method.Performance comparisons of the method were presented using Mnist dataset,UCI datasets,Tennessee Eastman process datasets and Mnist datasets with mixed Gaussian noise of different levels.Experimental results show that the proposed algorithm has higher accuracy and better robustness than ELM and HELM.