Transformer-based Drum-level Prediction in a Boiler Plant with Delayed Relations among Multivariates

Abstract--The steam drum water level is a critical parameter that directly impacts the safety and efficiency of power plant operations. There is usually physical information behind the complex interrelations: for I. Proper control of drum level is essential for B's outlet are shared via a same parent pipe; thus, their ensuring continuous steam production while adhering to safety outlet pressures are always in the same trend. However, achieving precise control of drum level to understand all the variables causal interrelation is critical to is a challenging task due to various process disturbances and predict the final drum level. Traditionally, control Moreover, given the long reaction chain, for example, strategies for drum level regulation have relied on feedback increasing pump A's inlet flow will take a long delay to be and feedforward control techniques, often employing Proportional reflected in the increasing drum level (usually the delay will Integral Derivative (PID [1]) controllers in conjunction be around 100 seconds and it varies to different variables and with rule-based feedforward controllers. While these strategies different boiler plant), so when predicting, pump A's inlet flow have been effective to some extent, they often struggle to at time step t should be used to predict the drum level at time adapt to changing operating conditions and fail to capture the step t + t In recent years, there has been a growing interest in In this study, we aim to develop a data-driven model leveraging advanced deep-learning techniques to predict the based on Transformer architecture for predicting drum level future drum level, which enhances the PID via feedforward variations in steam boilers.