A Novel Geolocation-Based APU Performance Prediction Technique

To address the issue of insufficient consideration of the geographical environment impact of departure airports in the performance prediction of civil aricraft Auxiliary Power Units (APU), this paper proposes a PlaceFormer prediction framework based on geographical location mapping. This framework innovatively designs a location information embedding module that maps discrete airport location information into highdimensional vectors to implicitly encode complex environmental factors such as geography and climate. Simultaneously, the DiffAttention mechanism is introduced to focus on the performance variation differences between consecutive flight missions, thereby enabling more accurate modeling of the cumulative degradation process of the APU. Experiments conducted on a dataset constructed from real Flight Data Recorder (QAR) data show that compared to traditional Transformer, GRU, and other state-of-the-art prediction models, PlaceFormer reduces the Root Mean Square Error (RMSE) in predicting multiple APU units. These results validate its application potential in leveraging geographical location information for more accurate APU performance state prediction.