Towards precision aviation emission modeling: A hybrid paradigm of convolutional neural networks and semi-empirical formulas for full flight phase gas pollutant indices.

As the aviation industry expands, the environmental and climatic impacts of aviation emissions are becoming critically important. These emissions, especially gaseous pollutants, significantly deteriorate air quality and contribute to climate change by altering atmospheric structures. Current models for predicting these emissions face high data reliance and limited accuracy. Addressing these challenges, this paper introduces a new model, AGEP-CNNSEF (Aviation Gaseous Emissions Prediction Model based on Convolutional Neural Networks and Semi-Empirical Formulas), which enhances prediction accuracy across full flight phases. This model integrates various data sources, including engine emission databases, cruise experiments, and alternative fuel studies. It employs the Boeing Fuel Flow Method 2 along with data from aircraft Quick Access Recorder data and the Open Aircraft Performance Model to enrich the dataset for climb, cruise, and descent phases, thus optimizing the use of semi-empirical formulas and reducing overfitting risks. Comparative analysis demonstrates the model's superior accuracy, with R values for NOx, CO, and HC test sets reaching 0.98, 0.93, and 0.91, respectively, significantly outperforming traditional models, especially during the cruise phase. Application of the model to a flight from Beijing to Xinjiang indicates that the highest emissions occur during cruising, with NOx, CO, and HC contributing 68.2 %, 43.8 %, and 56.9 % of total emissions, respectively. This model provides a precise tool for calculating aviation emissions, offering essential data for environmental impact assessments.