Towards sustainable aviation: aircraft fuel flow estimation using an improved iTransformer for pollutant emission calculations

Accurate estimation of aircraft fuel consumption and emissions is essential for sustainable aviation but is constrained by the lack of publicly available fuel flow data and the limited accuracy of existing methods. This study proposes an improved iTransformer-based approach for aircraft fuel flow estimation using ADS-B data. First, key variables influencing fuel flow are identified through Spearman correlation analysis. An LSTM network and a multi-head differential attention mechanism are then incorporated into the iTransformer architecture to improve temporal feature extraction, estimation accuracy, and model generalization. QAR and ADS-B data are preprocessed and analyzed, with the model trained using QAR data and subsequently applied to ADS-B data for fuel flow estimation. The fuel flow estimation yields an average accuracy of 93.7% based on 4,968 flights covering 11 aircraft types. Further estimation of fuel consumption and pollutant emissions was performed on a dataset of 23,756 flights with the same 11 aircraft types, and the overall average accuracy of all estimation results is higher than 90%. The resulting emission estimates are consistent with observed operational characteristics. Furthermore, the proposed method is applied to a large-scale dataset of 855,000 flight movements across mainland China in October 2024, enabling nationwide quantification of aviation fuel consumption and emissions. The results demonstrate the robustness and scalability of the proposed approach and highlight its potential to support aviation energy management, emissions monitoring, and sustainable aviation development.