Innovative model-optimized machine learning for high-accuracy predicting and exploring nitrogen transformation in biomass pyrolysis.

To enhance the environmental benefits of producing "zero-carbon" fuels from waste biomass pyrolysis, it is essential to suppress the formation of nitrogen oxides (NOx) at the source. The key lies in precisely regulating the migration and transformation behavior of nitrogen during the pyrolysis process. Addressing the limitations of traditional experimental methods in analyzing this complex process, this study proposes for the first time the Pyro-SPIN (Source Parameter-based Integrated Nitrogen Migration) model, which enables the synergistic prediction and optimization of nitrogen migration pathways in biomass pyrolysis. The results indicate that temperature, nitrogen content, and oxygen content are the three key factors governing nitrogen migration. Based on the synergistic analysis of model predictions and experimental data, it was found that under the co-regulation of multi-parameters composed of particle size (<200 μm), low temperature (<500°C), pyrolysis duration (<60 min), and slow heating, nitrogen migration pathways are directed toward enrichment in the solid char. Through model optimization strategies such as chain modeling and multi-output joint training, the predictive results maintain mass conservation (the sum of nitrogen in the three phases equals 100%) while significantly improving prediction consistency. This study developed a machine learning approach to predict and regulate nitrogen migration during biomass pyrolysis, offering a new way to understand its mechanisms and optimize the process. Furthermore, the predictive model and collaborative optimization strategy established in this research can also serve as a technical reference for predicting the migration behavior of other key elements during biomass thermochemical conversion processes.