Open Collections

Abstract

This research presents an efficient machine-learning model to predict energy consumption in Unmanned Aerial Vehicles (UAV) under varying environmental and weather conditions. Unlike conventional UAV energy models, we use Cross-Industry Standard Process for Data Mining methodologies to analyze UAV operational logs and meteorological data. The approach applies feature engineering techniques to account for the complex interaction between environmental factors and UAV energy needs. Our research includes multiple linear regression, ensemble, and neural network machine learning models. The evaluation, comprising full-range analysis and interval-specific assessments, highlights the excellent predictive accuracy of ensemble models such as Gradient Boost and eXtreme Gradient Boost. After validation and testing, the model was deployed in an experimental test flight that showed a minimal discrepancy of 0.005 Wh between predicted and actual energy consumption, an indication of the model’s robustness. This model provides UAV operators with real-time, precise energy estimates, thereby optimizing operational efficiency and safety practices. Beyond a significant advancement in UAV energy management, this work enhances UAV navigation and planning, bridging the gap between theoretical estimates and practical applications.