Prediction of superheated steam temperature for thermal power plants using a novel integrated method based on the hybrid model and attention mechanism

The stability of superheated steam temperature (SST) is severely challenged by the adjustment of thermal power plants under a wide-load range. Accurate and efficient prediction of SST plays an important role in the control of superheat system. To this end, an SST prediction model based on a multi-mode integrated method is proposed in this paper. Firstly, conservation of energy, as an equality constraint, is introduced into the loss function of the data-driven model based on Long Short-Term Memory (LSTM) architecture. Subsequently, the physical relationship between SST and the spray water flow, as an inequality constraint, is introduced into the above loss function. Finally, an individual hybrid model for each operating mode is developed and integrated with multi-mode switching strategies based on attention mechanism. Operating data with a wide-load range is sampled from a supervisory information system (SIS) to validate the superiority of the proposed method. The comparison results demonstrated that the predictive effect of the hybrid model with physics-based loss function is not only generalizable but also scientifically consistent with the dynamic characteristics of the step response. Furthermore, the results of the multi-mode integrated model prove the effectiveness of the mode switching strategy based on the attention mechanism.