Optimized Deep Learning for Reliable Atmospheric Duct Interference Prediction in Radar and Communication System

ABSTRACT Atmospheric duct interference (ADI) significantly impacts wireless long‐distance communications, radar systems, and naval activities via anomalous signal propagation in tropospheric ducting layers. The previously used predictive methods, anchored on physical models and meteorological data, proved not to be effective in capturing the complex spatiotemporal features of ADI. This paper introduces an Improved Whale Optimization Algorithm‐enhanced CNN‐GRU model (IWOA‐CNN‐GRU) that addresses these limitations through three key innovations: (1) a hybrid convolutional and recurrent network for joint spatiotemporal feature extraction, (2) an enhanced WOA with Lévy flight (where heavy‐tailed jumps mitigate local optima with 10% probability) and adaptive parameter optimization (dynamically balancing exploration/exploitation) to effectively optimize hyperparameters, and (3) robust generalization over a variety of atmospheric conditions over tropical coastal regions. Experimental results demonstrate the improved performance of the model with 98.50% validation accuracy and significant improvements in precision (0.95 ± 0.02), recall (0.96 ± 0.01), and F1‐score (0.95 ± 0.01), a 12%–15% improvement over baselines. The IWOA optimizer performed improved convergence ( p < 0.01 in Wilcoxon rank‐sum tests) and reduced premature convergence cases by 37%, with the efficiency of computation being boosted by 14.2% during training time without compromising stability (σ 2 < 0.005 for accuracy variance). These gains confirm the algorithm's suitability for real‐time prediction of ADI in operational networks, including radar and communication networks. This work establishes a new benchmark in machine learning–based ADI prediction, with significant implications for enhancing the reliability of wireless systems under changing atmospheric conditions.