Multi-Scale Dilated Autoformer for UAV Energy Consumption Forecasting

Understanding power consumption conditions is necessary for optimizing UAV energy use, particularly during flight under varying weather conditions and environmental factors. Maintaining UAV energy while accounting for multiple influencing variables and vulnerability to weather conditions provides an appropriate case study for advanced predictive modeling. This study investigates UAV power consumption during hovering flight by forecasting power usage using a MDFA network to improve prediction accuracy and better adapt to rapid weather-induced variations. To capture intricate temporal dependencies and recurrent oscillatory behavior, the integrated model combines multi-scale dilated convolutions with a Fourier-enhanced mechanism. According to the experimental results, this model achieves 3% error reductions under all tested flight conditions, indicating a significant improvement in performance. Overall, the MDFA model consistently showed better performance under high power consumption conditions than under low power consumption conditions, and it produced the lowest error in heavy flight compared to low and medium flight.