Постановка наукового завдання удосконалення методів прогнозування погодних умов для покращення управління польотом БПЛА

Abstract

This research provides a comprehensive and formalized statement of the scientific task aimed at bridging the critical operational gap between macro-scale atmospheric modeling and the high-precision requirements of autonomous Unmanned Aerial Vehicle (UAV) flight control loops. The author identifies that contemporary Numerical Weather Prediction (NWP) models, while effective for general aviation, lack the spatial and temporal resolution necessary to detect micro-scale phenomena – such as urban wind tunnels, localized turbulence, and rapid icing – which directly jeopardize the stability of light and medium-range UAVs. This discrepancy creates a significant scientific challenge: the need for a localized, real-time predictive framework capable of accounting for atmospheric stochasticity. The study proposes a paradigm shift toward an intelligent predictive ecosystem integrated directly into the Flight Control System (FCS). The core of this scientific task involves the development of a predictive functional 𝑓(𝑀𝑡, 𝑆𝑡, 𝑃𝑡) designed to minimize the prediction error ϵ by leveraging advanced Deep Learning architectures, specifically Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) models. These AI-driven tools are tasked with the continuous assimilation of heterogeneous data streams from onboard sensors, ground-based meteorological stations, and satellite feeds.Furthermore, the paper explores the mathematical integration of these high-fidelity forecasts into the autonomous control law 𝑢 = ϕ(𝑍𝑡, 𝑀𝑡+1 ̂). This integration enables the transition from legacy reactive compensation to proactive trajectory management, allowing the autopilot to anticipate and neutralize atmospheric disturbances before they affect flight integrity. The research substantiates that "weather-aware" autonomy is essential for optimizing battery energy consumption, enhancing mission reliability in austere environments, and ensuring the safe, systematic integration of autonomous systems into the U-space framework. Ultimately, the article concludes that the proposed multidisciplinary approach, combining aviation meteorology with reinforcement learning, is fundamental for achieving the airworthiness standards required for future digital and highly congested airspaces regulated by ICAO.