Operating condition feature representation-based Fourier graph network for civil aircraft state estimation

This paper focuses on the modeling of the spatial-temporal consistency in real-world flight series for civil aircraft state estimation under multiple operating conditions. Previous methods typically involve both graph networks and temporal networks to capture spatial and temporal dependencies for flight series, respectively. In addition, the impact of operating conditions on key performance parameters has been neglected, which leads deep models to generate indistinguishable attention for different operating conditions. To tackle these dilemmas, we propose an operating condition feature representation-based Fourier graph network (OCFGNet). Firstly, we begin with mapping the input flight series into a hypervariable graph that considers spatial-temporal dynamics in a unified way. Then the proposed Fourier graph operator with the shared weight parameter mechanism can balance the information of neighboring nodes of different propagation sequences, facilitating the collection of various spatial-temporal dependencies in the created graph. Moreover, for better perception of complex operating conditions, the Fourier graph operator is devised to produce distinguishable attention for operating conditions while achieving lower complexity. Extensive experimental results on CMAPSS, N-CMAPSS and civil aircraft bleed air system datasets demonstrate that OCFGNet significantly outperforms existing state-of the-art methods. The relevant code will be open for further study.