Abstract
Space debris is a major problem in space exploration. International bodies continuously monitor a large database of orbiting objects and emit warnings in the form of conjunction data messages (CDMs). An important question for satellite operators is to estimate when fresh information will arrive so that they can react timely but sparingly with satellite maneuvers. We propose a statistical learning model of the message arrival process, allowing us to answer two important questions: (1) Will there be any new message in the next specified time interval? (2) When exactly and with what uncertainty will the next message arrive? The average prediction error of our Bayesian Poisson process model is smaller than the baseline by more than 4 hours in a test set of 50,000 close encounter events.
Introduction
As Earth's orbit becomes increasingly crowded with space debris, satellite owners and operators (O/Os) rely heavily on the Space Surveillance Network (SSN) for collision warnings. These warnings, delivered as Conjunction Data Messages (CDMs), provide critical information such as the Time of Closest Approach (TCA) and the probability of collision.
For operators, the primary challenge is timing. Maneuvers are costly, consuming fuel and interrupting service. Acting too early on outdated data wastes resources, while waiting too long for a "better" update that may never arrive risks a catastrophic collision. Standard machine learning models often struggle with this task because data for any single conjunction event is scarce, typically consisting of only a few messages. This paper addresses the need for a reliable, probabilistic estimate of when the next piece of information will arrive.
Methodology
The arrival of CDMs is modeled as a homogeneous Poisson Process (PP), assuming that the time between messages (inter-CDM time) follows an exponential distribution with a rate $\lambda$. To overcome the data scarcity of individual events, a Bayesian Estimation framework is employed:
- Prior Distribution: A large historical dataset is used to establish a global Gamma prior for the arrival rate $\lambda$, representing general messaging behavior across all events.
- Likelihood: As new messages arrive for a specific ongoing event, the model calculates the likelihood of these observed arrival times.
- Posterior (MAP): By combining the global prior with event-specific data, the model calculates a Maximum a Posteriori (MAP) estimate for that specific event's arrival rate ($\lambda_k$).
This approach allows the model to make educated predictions even with very few initial observations, refining its accuracy as more data is received for a specific event. The study utilized a dataset provided by the European Space Agency (ESA), with a test set of 50,000 independent close encounter events to ensure statistical significance.
Results and Discussion
The Bayesian Poisson Process was compared against a "Classic" Poisson model (using Maximum Likelihood Estimation) and a "Baseline" model that assumes the next interval will match the previous one.
As shown in the results table, the Bayesian model achieved a Mean Absolute Error (MAE) of approximately 3.6 hours (0.15 days). This represents a significant improvement over the baseline. Notably, the Root Mean Square Error (RMSE) improvement exceeded 4 hours, demonstrating the model's robustness against large errors.
Key insights from the study include:
- Poisson Behavior: Despite the complexity of orbital mechanics, CDM arrivals behave remarkably like a classic stochastic process.
- Robustness: The Bayesian approach "smooths" predictions using global trends, making it far more reliable than the Classic Poisson approach, which is highly sensitive to outliers in small sample sizes.
- Uncertainty Quantification: The framework provides 90% credibility intervals, offering operators a mathematical measure of confidence for tactical decision-making.
Conclusion
While the current model assumes a constant arrival rate, future work could explore non-homogeneous Poisson Processes to capture potential increases in message frequency as the TCA approaches. Additionally, integrating other CDM features, such as object size or orbit type, could further refine the prior distribution. Despite these potential areas for growth, the Bayesian Poisson Process provides a significant leap forward in helping satellite operators manage the risks of orbital conjunctions with greater precision.
Summary updated