Abstract: 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.