Secondary Reheat Steam Temperature Prediction Based on Hybrid Deep Learning

An accurate prediction model of the secondary reheat steam temperature in the ultra-supercritical double-reheat coal-fired power plant is difficult to build due to the characteristics of strong nonlinearity, multi-variable coupling, large inertia, and various disturbances. This paper proposed a joint data-driven prediction method based on GRU-Pooling-Attention to improve the prediction accuracy and training speed. Firstly, CatBoost algorithm is taken to calculate the importance of features, and the input feature set is selected combined with expert knowledge. Then the GRU is adopted to learn the time series relationship between data, the pooling method is embedded to reduce the number of parameters and to highlight key values, and the attention mechanism is introduced to assign different weights to the output of each time step of GRU to reflect the large delay and large inertia of the system. The hybrid deep learning method is applied to the prediction of the secondary reheat steam temperature based on historical data, and the results show that it can obtain accurate and fast multi-step prediction. The trained model can be used to explore the dynamic characteristics of the processes.