Ensemble machine learning methods for Euler angles detection in an inertial navigation system

A.N. Grekov1, 2, A.A. Kabanov2

1Institute of Natural and Technical Systems, RF, Sevastopol, Lenin St., 28

2Sevastopol State University, RF, Sevastopol, Universitetskaya St., 33

E-mail: i@angrekov.ru

DOI: 10.33075/2220-5861-2022-1-112-120

UDC 551.465 (262.5)

Abstract:

Many countries in the world are showing great interest in the ocean, which is rich in resources and energy. In this regard, active research and development of a large number of marine equipment for the study and development of the ocean is underway. Along with the classical methods of ocean exploration, autonomous platforms such as AUVs, gliders and autonomous surface miniships have been actively introduced in recent decades.

The work is focused on increasing the reliability of the navigation information of such autonomous platforms, namely: determining the Euler angles using experimental data generated at the output of an inertial navigation system built on the basis of MEMS sensors. Two ensemble methods of machine learning are considered: majority voting (voting by majority) and weighted majority voting. The ensembles are formed by combining three supervised learning methods: support vector machine (SVM), k-nearest neighbors (KNN), and decision trees.

After hyperparameter optimization, the accuracy (accuracy) of the classification of the KNN model was 0.89 ±0.01, and the decision tree model was 0.91 ±0.01. Testing on the test data set also confirms the good performance of these classifiers: KNN accuracy – 0.90, and decision tree – 0.91. Combining the above two classifiers and the SVM classifier into an ensemble with a weighted majority gives a small increase in the accuracy (accuracy) of the classification: 0.92 on the training and test data sets, which allows recommending the obtained research results for improving the quality of navigation information. In the case of the ensemble with a majority vote, no significant increase in classification accuracy (accuracy) is found in comparison with individual classifiers.

Keywords: 

To quote:

Full text in PDF(RUS)

REFERENCES