点击次数：

论文类型：会议论文

发表时间：2017-01-01

收录刊物：CPCI-S

页面范围：2422-2427

关键字：deeplearning; feature extraction; restricted Boltzmann machine; time series; long-term prediction

摘要：In this study, a deep denoising recurrent temporal restricted Boltzmann machine network is proposed for long-term prediction of time series. The network is a deep dynamic network model which is stacked by multiple denoising recurrent temporal restricted Boltzmann machines with strong modeling ability for complex high noise time series data. To better deal with high noise data, a random noise is added to original data to form a new input sample and the model is forced to fit the original data without the noise to improve the denoising ability of the model. To train and predict real data directly, the Gaussian distribution is introduced into the energy function of the model to establish a Gaussian network model. After training the network layer by layer, this paper proposes to use the back-propagation algorithm to fine-tune the whole network and improve the modeling performance. Through the actual data simulation in the production process of the iron and steel enterprise, the comparative analysis shows that the proposed method can effectively improve the prediction accuracy and achieve satisfactory results.