Research On Preparation Method Of High-precision ASF Database Based On Transfer Learning

In the ground-based long-wave navigation timing system,the signal is affected by factors such as time/space variation characteristics on the transmission path,and the propagation delay will produce hundreds of nanoseconds or even subtle changes.It is very fatal in terms of one of the propagation delays called Additional Secondary Phase Factor(ASF),which is the main factor affecting the positioning accuracy of the system.In practical applications,the method of establishing a database is usually used,and the database is loaded into the timing receiver for real-time correction.Therefore,the research on a high-precision and low-cost ASF database preparation method is of great significance.In this thesis,the spatial variation characteristics of the propagation delay of low-frequency ground waves are studied in detail.In the area with long propagation distance and large variation of topographic features,the traditional propagation delay theory has a large error between the predicted value and the measured value.A back-propagation neural network(BPNN)prediction model based on transfer learning is proposed,which effectively improves the prediction accuracy of propagation delay in complex regions,and provides a new method for the preparation of ASF database.The specific contents are as follows:(1)Based on the traditional numerical calculation method,the relationship between spatial feature variation and propagation delay is analyzed.The results show that the propagation delay of low-frequency ground waves is closely related to the propagation distance on the propagation path and the change of topography.The longer the propagation distance and the larger the terrain change in the propagation path,the larger the propagation delay value.(2)Based on the theory of deep learning,a BPNN model based on the predicted value of propagation delay theory is established.Within the signal coverage of the ground-based longwave navigation and timing station the entire working area divided into several sector ring areas.Each small area is modeled by a neural network,and the theoretical value of the propagation delay obtained by the integral equation method is used as a training set to train the BPNN model,and a theoretical value that fits well to the theoretical prediction value of the propagation delay is obtained.prediction model.(3)Using the transfer learning method,establish a BPNN numerical prediction model whose prediction results are closer to the actual value.According to the transfer learning theory,using the propagation delay value obtained from the actual measurement,the previously trained theoretical value prediction model is modified to obtain a BPNN prediction model with a more accurate prediction result of the propagation delay measurement value.And designed a control experiment to analyze the relationship between the selection of the measured sample location and the model prediction effect after transfer learning.The research results of this thesis can provide a new idea for the preparation of highprecision propagation delay correction database,and have certain reference value.