Regional zenith tropospheric delay prediction using DBO-optimized CNN-LSTM with multihead attention.

Zenith Total Delay (ZTD) is integral to applications such as atmospheric water vapor inversion and precise positioning in the Global Navigation Satellite System (GNSS). The development of high-precision regional ZTD models has emerged as a significant area of research within the GNSS domain. This study addresses the challenges associated with achieving high-precision tropospheric delay predictions under specific conditions and the limitations of CNN-LSTM models, particularly regarding suboptimal hyperparameter optimization and convergence to local optima. We propose a novel regional ZTD prediction model, the CNN-LSTM-Multihead-Attention (CLMA) model, optimized using the Dung Beetle Optimization (DBO), referred to as ZTD-DBO-CLMA. This model synergistically integrates the spatial feature extraction capabilities of Convolutional Neural Networks (CNN) with the temporal sequence modeling strengths of Long Short-Term Memory (LSTM) networks, enhanced through advanced hyperparameter optimization techniques. The model facilitates synchronized learning of CNN and LSTM components via the DBO optimization algorithm and the incorporation of a multihead attention mechanism.In our study, we utilized five consecutive months of ZTD data from 40 International GNSS Service (IGS) stations within the European region, sampled at one-hour intervals, to investigate regional ZTD prediction models. We employed the ZTD-DBO-CLMA model and compared it to the ZTD-CLMA model, which lacks DBO optimization. The results indicate that the ZTD-DBO-CLMA model significantly enhances prediction accuracy, reducing the mean absolute error (MAE) and root mean square error (RMSE) by 0.31 mm and 1.38 mm, respectively, while increasing the coefficient of determination (R²) by 39.43%. Furthermore, the DBO algorithm consistently demonstrates its optimization efficacy across diverse weather conditions, thereby improving the precision of ZTD predictions.