In this study, employing the idea of layer by layer feature extraction in deep learning, a deep denoising kernel mapping-based least square support vector machine long-term prediction model is proposed. The proposed model can deal with the poor long-term prediciton ability problem of the least square support vector machine with shallow kernel structure. By transforming through the deep kernel network layer by layer, the sample data are mapped to the deep feature space to improve the prediction accuracy. In addtion, the denoising algorithm is incorporated into the training process of the deep kernel network to improve the fitting performance for the data with high level noises. Furthermore, the whole network is fine-tuned to further improve the modeling ability by using the back propagation algorithm. The results of the standard dataset and the actual industrial data simulation experiments show that the proposed method can extract the feature information contained in the data, and effectively improve the prediction accuracy.