HARDWARE AND SOFTWARE MEANS FOR DYNAMIC RECONFIGURATION OF A GROUP OF SMALL SPACE VEHICLES

Abstract

The goal of the study is to automate the control of a group of nanosatellites in conditions of its
variable number by updating its state based on sending and processing broadcast requests between
nanosatellites and using the Transformer neural network. A neural network is needed to make predi ctions
about the state of the spacecraft network. The problem of ensuring connectivity of a network of
nanosatellites is studied, which comes down to the implementation of adaptive network control with
assessment and prediction of the state of communication channels between pairs of devices based on a
neural network. Dynamic reconfiguration and machine learning of a network of devices have been developed.
Algorithmic tools have been defined for the initial training of a neural network and its subs equent
additional training, taking into account the preprocessing of the original sparse or fully connected
data sets about the network of devices. Upon completion of training on synthetic data, the created
neural network is able to predict the quality of communication, taking into account line of sight, signal
attenuation depending on distance and the state of the nanosatellite hardware platform. The developed
software system performs deterministic reconfiguration based on the current state of the nanosatellite
network and adaptive reconfiguration based on historical data by analyzing the hidden patterns of
nanosatellite functioning using the Transformer neural network. To predict the quality of communication,
a functional is used to connect the geodetic coordinates of pairs of satellites and the vectors of
their states with the elements of the matrix of the quality of communication between nanosatellites with
a given initial time, the value of the time interval, and the value of the sampling step of the measurement
process. The use of neural networks implemented on GPUs made it possible to predict possible
states of nanosatellites and carry out reconfiguration of the constellation ahead of schedule, including
removing “problematic” nanosatellites from the network.