Artificial neural networks for analysis of process states in fluidized bed combustion

There are several challenges confronting energy production nowadays, such as increasing the efficiency of combustion processes and at the same time reducing harmful emissions. The latter, however, often necessitates process improvement, which requires knowledge of the behavior of the process. It is therefore important to develop and implement novel methods for process diagnostics that can respond to the challenges of modern-day energy plants. In this study the formation of nitrogen oxides (NOx) in a circulating fluidized bed (CFB) boiler is modeled by using artificial neural networks (ANN). In the approach used, the process data are first arranged using self-organizing maps (SOM) and k-means clustering to create subsets representing the separate process states in the boiler, including load increase and load decrease situations and conditions of high or low boiler load. After the determination of these process states, variable selection based on multilayer perceptrons (MLP) is performed to obtain information on the factors affecting the formation of NOx in those states. The results show that this approach provides a useful way of monitoring a combustion process.

Research highlights
► An efficient non-linear method is presented for diagnosing and monitoring the formation of nitrogen oxides in circulating fluidized bed boilers.
► The method enables analysis of the process using self-organizing maps, determination of process states and selection of variables using multilayer perceptrons.
► The method is especially advantageous when dealing with a large number of samples and the grouping of data into clusters provides extra accuracy in the emission models.