Method for Loran-C Additional Secondary Factor Correction Based on Neural Network and Transfer Learning

Abstract

An additional secondary factor (ASF) correction method is proposed to improve the accuracy of Loran-C navigation and positioning based on a combination of backward propagation neural network (BPNN) and transfer learning. The BPNN is used to train a network model based on the theoretical ASF data source calculated by the integral equation method. Although the trained BPNN model fitted well with the theoretical ASF results, it is not sufficient for ASF prediction in the actual environment involving uncertain topography and geology. To compensate for this deficiency, transfer learning is employed to further finely calibrate the BPNN model against the ASF measurements. By using the advantages of BPNN and transfer learning, the proposed method achieved higher accuracy than the purely theoretical method. Measurement results are used to validate the superiority of the proposed ASF correction method to the theoretical one. Particularly, the mean absolute error of the ASF prediction results decreased from 2.02 to 0.29 <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">μ</i> s after introducing the transfer learning in the trained BPNN model.