基于降噪自编码器与改进卷积神经网络的采煤机健康状态识别

Health Status Identification of Shearer Based on Denoising Autoencoder and Improved Convolutional Neural Network

针对采煤机监测参数间关联性强、冗余信息多且受强噪声干扰导致其健康状态识别困难及传统的采煤机状态识别方法在健康状态指标构建中人工参与过多导致识别准确率不高的问题,提出一种基于降噪自编码器(denoising autoencoder,DAE)与改进卷积神经网络(improved convolutional neural network,ICNN)的采煤机健康状态识别方法。首先,对原始监测数据作滑动平均降噪处理并进行归一化;其次,通过无监督训练降噪自编码器实现数据降维、特征提取,进而构建健康状态指标;然后,根据降噪后的监测数据与健康状态指标训练改进卷积神经网络模型,实现采煤机健康状态的自动识别;最后,利用采煤机仿真数据完成模型验证并与其他多种健康状态识别方法进行对比。结果表明:该方法识别准确率达98.38%,明显高于其他方法,可为后期的预知维护提供理论支持。

Given the strong correlation between the monitoring parameters of a shearer,redundant information,and the interference of strong noise,it is difficult to recognize its health status.The traditional shearer status recognition method requires several manual interventions in the construction of the health status indicator,resulting in poor recognition accuracy.A denoising autoencoder and an improved convolutional neural network-based models are proposed to identify a coal shearers’health status.First,moving average noise-reduction is performed on the original monitoring data and normalized.Next,data dimensionality reduction and feature extraction through unsupervised training of noise-reducing autoencoders is implemented,and the health indicators are constructed.Furthermore,the convolutional neural network model is improved by training the monitoring data after preprocessing.The health state index is predicted to achieve automatic recognition of the health status of the shearer.Finally,the model is verified by comparing the shearer’s simulation results with those obtained from other health status identification methods.The results show that the recognition accuracy of this method is 98.38%,which is significantly higher than other methods and can provide theoretical support for predictive maintenance.