A dynamic weight fusion method for ASF prediction based on multi-dimensional observational data

To address the issue that the additional secondary factor (ASF) in Enhanced Long-Range Navigation (eLoran) systems is highly susceptible to environmental disturbances in complex maritime environments, while conventional prediction methods exhibit limited accuracy and strong dependence on external meteorological information, this paper proposes a dynamic weight fusion method for ASF prediction based on multidimensional observational information. The proposed method utilizes receiver signal parameters as the foundation and incorporates multidimensional observational data as inputs to construct a predictive framework that integrates multiple base models. It systematically investigates the correlations among different observational features and their capabilities in characterizing ASF variations. A dynamic weighting mechanism is introduced to adaptively adjust the weights of sub-models according to their prediction errors. Furthermore, an exponentially weighted moving average (EWMA) smoothing constraint is employed to enhance model stability and robustness. Experimental results based on real-world data show that machine learning and deep learning approaches outperform traditional linear models in capturing the nonlinear evolution and temporal characteristics of ASF. The proposed fusion strategy further improves both prediction accuracy and stability. The method achieves superior performance in both time-difference prediction tasks and outperforms benchmark methods in terms of error control and correlation metrics. The results indicate that the proposed approach effectively enhances ASF prediction accuracy in complex marine environments, thereby providing significant support for high-precision positioning in eLoran systems.