Temperature and pressure time series prediction model based on autoencoder-GRU and phase space reconstruction

The temperature and pressure prediction of main and reheat pipelines is vital to the safety work of the thermal power units. In this paper, a temperature and pressure prediction algorithm based on the phase space reconstruction (PSR) and deep learning network is proposed. The deep learning algorithms, which are named as TppNet, are based on the autoencoder structure and Gate Recurrent Unit (GRU)and adopt six months' data to establish the dataset. The results proved that this method can ensure the temperature and pressure relative errors of main and reheat pipelines are less than 5% in 50 steps, and that value will not exceed 10% in 15 steps under the shutdown and startup conditions.