Modeling of the fluidized bed combustion process and NOx emissions using self-organizing maps: An application to the diagnosis of process states

Efforts to reduce harmful emissions and the increasing demands for combustion efficiency have generated a number of challenges for power plants. Changes in the operation of a combustion process, for example, can induce fluctuations that have unexpected consequences such as an increased level of emissions. Despite the importance of these changes, their impact and relevance are often ignored in analyses of industrial process data due to the complexity of these phenomena. It seems that the behavioral evolution of a process could be understood more easily by monitoring the transition of the process from one characteristic state to another. We demonstrate here that the self-organizing map (SOM) provides an efficient method for revealing the most characteristic features of input data, making it a powerful tool for discovering general phenomena and visualizing the behavior and evolution of a combustion process. In this approach the process data are analyzed using a SOM and K-means clustering to create subsets representing the separate process states in the boiler. A trajectory analysis is then performed to indicate fluctuations in the process and their implications. The results show that process fluctuations can significantly affect the levels of nitrogen oxides released. The method enables efficient diagnosis, and provides a clear illustration of the evolution of the process and an applicable means of defining the best path for achieving a more efficient process that produces less emissions.

Research highlights
► We present an efficient method for diagnosing the circulating fluidized bed process.
► We apply the method to the diagnosis of NOx emissions within process states.
► Process fluctuations can significantly affect the level of NOx released.
► The method can help in achieving a more efficient process.