Atmospheric Turbulence Prediction Based on Optimal Feature Extraction and Integrated Long Short-Term Memory Networks

Turbulence poses a significant threat to aviation safety due to its sudden onset and irregular nature. Traditional prediction models often struggle to capture the complex nonlinear and multi-scale characteristics of turbulence. To address this issue, this paper develops a multi-step turbulence prediction framework that integrates wavelet decomposition with deep learning. The study is based on Doppler lidar wind field observation data from the approach airspace of Kashgar Airport in 2024, aiming to predict turbulence variations over the next fifteen minutes. The model first employs two-dimensional discrete wavelet transform to extract multi-scale time-frequency features from the wind speed matrices, and then uses correlation analysis to select key indicators, thereby enhancing the physical interpretability of the features. A hybrid deep learning framework is proposed by integrating a Transformer module for capturing global periodic patterns, a deep convolutional neural network (DCNN) for extracting local disturbance features, and a bidirectional long short-term memory (BiLSTM) network for modeling bidirectional temporal dependencies. Experimental results show that the proposed fusion model consistently outperforms all the baseline models used for comparison in terms of EDR prediction accuracy, stability, and generalization capability.