基于深度神经网络的水下导航适配区划分与预测方法

A Method for Dividing and Predicting Underwater Navigation Adaptation Zones Based on Deep Neural Networks

摘要: 本文使用神经网络等算法对水下导航适配区进行数据驱动分类预测进行深入研究。研究旨在通过处理重力异常数据来提高重力梯度辅助导航的精度，解决国产仪器在精确重力梯度数据上的缺失问题。研究建立了一个基于K-Means算法的模型，该模型通过对插值后的重力异常数据进行处理，实现了对导航适应区的精细化可视化、划分和校准。进一步地，研究还利用神经网络，根据重力异常值的均值和方差等特征指标来预测不同区域的导航适应性。在实际应用中，所开发的系统在分类和预测方面显示出了很高的准确性和泛化能力，证明了选定的特征指标能够有效反映重力异常值的特性，并且基于深度学习的适应性评估标准能够清晰地区分不同的适应性区域。然而，研究也识别出了系统可能存在的误差和局限性，这些主要是由于数据收集条件的差异，以及所选特征指标可能未能全面覆盖或者对某些特征高估。

Abstract: This article conducts in-depth research on data-driven classification prediction of underwater navigation adaptation zones using algorithms such as neural networks. The research aims to improve the accuracy of gravity gradient assisted navigation by processing gravity anomaly data, and solve the problem of missing precise gravity gradient data in domestic instruments. A model based on the K-Means algorithm was established, which processed the interpolated gravity anomaly data and achieved fine visualization, division, and calibration of the navigation adaptation area. Furthermore, the study also utilizes neural networks to predict navigation adaptability in different regions based on characteristic indicators such as the mean and variance of gravity anomalies. In practical applications, the developed system has shown high accuracy and generalization ability in classification and prediction, proving that the selected feature indicators can effectively reflect the characteristics of gravity anomalies, and the adaptability evaluation criteria based on deep learning can clearly distinguish different adaptive regions. However, the study also identified potential errors and limitations in the system, mainly due to differences in data collection conditions, as well as the fact that the selected feature indicators may not fully cover or overestimate certain features.