Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Toward Autonomous Cooperation in Heterogeneous Nanosatellite Constellations Using Dynamic Graph Neural Networks

Guillem Casadesus-Vila, Joan-Adria Ruiz-de-Azua, Eduard Alarcon

TL;DR

This paper tackles autonomous contact-plan design (CPD) for large, heterogeneous nanosatellite constellations modeled as dynamic graphs. It introduces a dynamic graph neural network (DGNN) based on the EvolveGCN framework to learn the CP objective, the average best delivery time $F$, and couples it with a simulated annealing (SA) optimizer to generate CPs. The DGNN, trained on synthetic data, achieves a mean absolute error of $3.6$ minutes in predicting $F$ and enables CP improvements of about $29.1\%$ while reducing objective evaluations to roughly $20\times$ faster than traditional CGR-based calculations. The approach demonstrates the feasibility of onboard autonomous CP optimization for dynamic, heterogeneous networks, potentially enhancing autonomy, scalability, and operational efficiency in future Earth observation missions.

Abstract

The upcoming landscape of Earth Observation missions will defined by networked heterogeneous nanosatellite constellations required to meet strict mission requirements, such as revisit times and spatial resolution. However, scheduling satellite communications in these satellite networks through efficiently creating a global satellite Contact Plan (CP) is a complex task, with current solutions requiring ground-based coordination or being limited by onboard computational resources. The paper proposes a novel approach to overcome these challenges by modeling the constellations and CP as dynamic networks and employing graph-based techniques. The proposed method utilizes a state-of-the-art dynamic graph neural network to evaluate the performance of a given CP and update it using a heuristic algorithm based on simulated annealing. The trained neural network can predict the network delay with a mean absolute error of 3.6 minutes. Simulation results show that the proposed method can successfully design a contact plan for large satellite networks, improving the delay by 29.1%, similar to a traditional approach, while performing the objective evaluations 20x faster.

Toward Autonomous Cooperation in Heterogeneous Nanosatellite Constellations Using Dynamic Graph Neural Networks

TL;DR

This paper tackles autonomous contact-plan design (CPD) for large, heterogeneous nanosatellite constellations modeled as dynamic graphs. It introduces a dynamic graph neural network (DGNN) based on the EvolveGCN framework to learn the CP objective, the average best delivery time , and couples it with a simulated annealing (SA) optimizer to generate CPs. The DGNN, trained on synthetic data, achieves a mean absolute error of minutes in predicting and enables CP improvements of about while reducing objective evaluations to roughly faster than traditional CGR-based calculations. The approach demonstrates the feasibility of onboard autonomous CP optimization for dynamic, heterogeneous networks, potentially enhancing autonomy, scalability, and operational efficiency in future Earth observation missions.

Abstract

The upcoming landscape of Earth Observation missions will defined by networked heterogeneous nanosatellite constellations required to meet strict mission requirements, such as revisit times and spatial resolution. However, scheduling satellite communications in these satellite networks through efficiently creating a global satellite Contact Plan (CP) is a complex task, with current solutions requiring ground-based coordination or being limited by onboard computational resources. The paper proposes a novel approach to overcome these challenges by modeling the constellations and CP as dynamic networks and employing graph-based techniques. The proposed method utilizes a state-of-the-art dynamic graph neural network to evaluate the performance of a given CP and update it using a heuristic algorithm based on simulated annealing. The trained neural network can predict the network delay with a mean absolute error of 3.6 minutes. Simulation results show that the proposed method can successfully design a contact plan for large satellite networks, improving the delay by 29.1%, similar to a traditional approach, while performing the objective evaluations 20x faster.
Paper Structure (14 sections, 3 equations, 5 figures, 1 table, 2 algorithms)

This paper contains 14 sections, 3 equations, 5 figures, 1 table, 2 algorithms.

Table of Contents

  1. Introduction
  2. Motivation
  3. Literature Review
  4. Paper Objective
  5. Overview
  6. Problem Statement
  7. System model
  8. Methodology
  9. Simulated Annealing
  10. Dynamic Graph Neural Networks
  11. Results
  12. DGNN Model
  13. Contact Plan Design Results
  14. Conclusions

Figures (5)

  • Figure 1: Dynamic graph neural network architecture for predicting latency in satellite constellation networks.
  • Figure 2: Training and evaluation loss. The model is trained for 16 hours using synthetic data corresponding to 30 satellites and 20 ground stations. Hyperparameters are selected using a grid search, including different activation and loss functions, as well as the number of layers and sizes.
  • Figure 3: Predicted and true normalized BDT for 100 different contact plans. The model successfully identifies contact plans with worse objective values and achieves lower accuracies on contact plans with lower objectives. It predicts the BDT of a contact plan with a mean absolute error of 3.6 minutes.
  • Figure 4: Possible routes from satellite 9 to satellite 28. The best route is colored in blue, passing through satellites 9, 0, 23, 10, 14, and 28, with a Best Delivery Time (BDT) of 8 minutes. The proposed DGNN can predict the average BDT of the network with a mean absolute error of 3.6 minutes.
  • Figure 5: Objective improvement using the DGNN to evaluate the objective function. The predicted objective values using the trained model are shown in blue, and the values computed a posteriori using the CGR algorithm are shown in orange. The shaded area represents the standard deviation over 10 runs.