Hybrid neural network models in the problem of telemetry data multiclass classification of small spacecrafts

Abstract

The paper presents solutions to the actual problem of intelligent data analysis of telemetry data of onboard equipment (OE) of small spacecraft (SS) in order to determine their technical states. Hybrid neural network models based on modern deep learning architectures have been researched and developed to solve the problem of multiclass classification of telemetry data, which make it possible to determine the normal and abnormal states of the operation of the OE SS. For computer analysis, we used telemetry data from OE SS of the AIST group of the Samara National Research University named after Academician S. P. Korolev. Computer experiments were carried out on training, validation and testing of the constructed hybrid neural network models, which showed their sufficiently high accuracy in solving the problem under consideration. A comparative analysis of the obtained hybrid neural network models with widely used deep neural network classifiers was performed, which showed the advantage of the obtained solutions both in classification accuracy and in training and validation time.