Machine learning for forecasting factory concentrations of nitrogen oxides from univariate data exploiting trend attributes

Abstract

The development of post-processing technology for spent nuclear fuel is essential to ensuring the sustainable growth of nuclear energy. However, post-processing facilities release copious amounts of emissions with high concentrations of nitrogen oxides (NOx), making the accurate measurement of their concentrations in radioactive settings greatly challenging. The application of machine learning strategies to predict NOx emissions offers a promising approach for improving the measurement and management of NOx in post-processing facilities, owing to their potential for cost reduction and operational expediency compared to conventional methods. Therefore, this study presents the outcomes of predictive activities for NOx emissions using machine learning. We employed a vector autoregression (VAR) model that considers the influence of other pollutants on NOx emissions. The results confirm that the VAR model sufficiently predicts NOx emissions. Furthermore, this study reveals the intricate interplay and feedback loops among various pollutants, thereby providing guidance for formulating comprehensive pollution control strategies. Finally, a lightweight and precise NOx forecasting model was developed by extracting the primary features affecting NOx predictions. This model has substantial significance for elevating the precision of pollutant emission forecasts and offers substantive support for the development and sustainable growth of the nuclear chemical industry.