Online tracking simulation system for a 660 MW Ultra-Supercritical circulating fluidized bed boiler based on mechanistic models and artificial neural network models

Ultra-supercritical circulating fluidized bed boiler technology plays a key role in the clean utilization of coal resources and peak-load regulation. This paper investigates the construction method of an online tracking simulation system for circulating fluidized bed boilers, integrating mechanistic models with artificial neural network models. The aim is to adjust model parameters online based on real-time data and simulate the dynamic processes of the thermal system. Compared to conventional online simulation, the online tracking simulation system achieves dynamic parameter adjustments, focusing on the research of tracking mechanisms based on control algorithms. This study explores the integration of sliding mode control with backpropagation and long short-term memory artificial neural network models within the online tracking simulation system. This method solves the problem of deviation between model prediction and actual working conditions from a structural level. The online tracking simulation provides abundant and high-quality training samples for the artificial neural network models, contributing to the development of accurate data models for the system. The hybrid model derived from their integration fundamentally alleviates the deviation issue between the online simulation system and actual processes. The results demonstrate that the sliding mode control tracking mechanism effectively corrects simulation system deviations and provides accurate samples for neural network model training. The long short-term memory model shows a 58.95% reduction in average error compared to the backpropagation model, highlighting the superior performance of long short-term memory in handling time-series data and improving prediction accuracy.