Cameron Harris, Virginia Polytechnic Institute and State University; Dylan Thomas, Virginia Polytechnic Institute and State University; Jonathan Kadan, Virginia Polytechnic Institute and State University; Kevin Schroeder, Virginia Polytechnic Institute and State University; Jonathan Black, Virginia Polytechnic Institute and State University
Keywords: SDA, SSN, genetic algorithm, metaheuristic optimization
Abstract:
The rapid proliferation of objects in the space domain mandates the improvement of current space situational awareness capabilities. The existence of information-based heterogeneous sensor management strategies presents a path forward. Efforts towards improved space domain/situational awareness (SDA) can be informed by metaheuristic optimization techniques to tactically and methodically expand the existing Space Surveillance Network (SSN). This work will leverage information-based tasking strategies with multi-objective optimization techniques to determine an array of possible space-based sensor expansions that complement the existing ground-based SSN.
While the ground-based SSN is not necessarily complete, there are coverage gaps that, for several reasons, are infeasible to rectify with more ground-based sensors. Space-based sensors are emerging assets to the SSN, and they provide an actionable solution path towards mitigating coverage gaps and improving SDA beyond the capabilities of the ground-based SSN. In order to assess the effectiveness of ground- and space-based sensors working cooperatively, the SSN is evaluated using an information-based tasking approach.
An information-based tasking approach reduces workloads on a sensor network while improving state estimates on each observed RSO. Information-based sensor network tasking strategies can be applied to a distribution of ground-based sensors to estimate the capability of the network to maintain SDA. Extending this methodology to space-based sensors as well provides an idea of future space-based sensor orbit/constellation design to effectively supplement the ground-based SSN.
For the information-based sensor tasking used in this study, sensors are assigned to observe satellites at each timestep using a Munkres (Hungarian) decision algorithm. States and covariances associated with every target are estimated using an Unscented Kalman Filter. This research tasks candidate SSNs utilizing a single time-step, myopic tasker, which compiles a cost-constrained reward function using the Lyapunov exponent and Fisher Information. The tasking methodology is chosen to be representative of a methodology that would ostensibly be implemented in the future SSN. A tool has been developed to simulate a model SSN tasked against population of RSOs using this tasking methodology.
This work uses metaheuristic optimization to evaluate the optimal sensor distribution with information-based tasking capabilities. The metaheuristic optimization is accomplished by the Nondominated Sorting Genetic Algorithm II (NSGA-II). Because some input variables are discrete, the solution space is not necessarily convex, so a convex optimization technique cannot be employed. A genetic algorithm is selected because it is robust against non-convex solution spaces. A further advantage of a genetic algorithm is its stochasticity, which primes the algorithm to search throughout the solution space. A genetic algorithm is the chosen method for solving the network distribution problem because there are multiple metrics that indicate the quality, as well as frequency and quantity, of the observations made by an SSN.
Constituents of a given SSN will be predefined assets used in the genetic algorithm process. In the software, an asset is a structure with fields that define its relevant properties and other information. Properties can be selected as optimization variables, such as sensor state vectors, and will be selected from an allowable range. Other properties will be left as static to make the problem computationally feasible.
Individual constituents of the existing ground-based SSN include electro-optical, radar, and advanced radar sensors, which provide 2D, 3D, and 4D observation data respectively. Space-sensors will be restricted to operate as electro-optical sensors to match contemporary space-based SDA. Electro-optical sensors operate in the visible spectrum and are treated like optical telescopes. All sensors in the simulation tool will be subject to hardware limitations (e.g., power and frequency constraints for ground-based radar sensors) as well as direct line of sight and other physical constraints.
The variable-length genetic algorithm can vary the number of space-based sensors evaluated in a candidate solution. The genetic algorithm generates an array of space-based sensors to work cooperatively with the existing ground-based SSN; the SSN produced by the union of the ground- and space-based sensors is passed into the sensor tasking simulation tool. Each candidate solution (consisting of the entire ground- and space-based network) will have its own corresponding fitness values relative to each objective in the genetic algorithm. The preliminary objectives of the sensor network distribution optimization can be broken down into two distinct areas: quality of observation data and cost.
Given a representative set of RSOs, the sensor tasking simulation tool reports information relevant to the quality of observation data, including average RMS position error, average Mahalanobis distance, and number of observations per timespan. After each generation of the genetic algorithm, the properties of the space-based sensors are adjusted to manipulate these metrics, and adjustments that yield positive results are assigned a better fitness value.
Additionally, the cost of each set of candidate space-based sensor is considered. The specific metaheuristic optimization tool used in this work is uniquely positioned to consider cost as an objective because it is designed to model the costs of the hardware of its assets. Space-based sensors also require additional implementation cost considerations that traditional ground-based sensors do not, such as launch costs.
The sensor tasking simulation tool needs a representative set of RSOs to be evaluated against in a simulated scenario. A representative population of target satellites includes satellites with varying orbital parameters in LEO, MEO, and GEO orbital regimes. To be consistent with the current space domain, the amount of simulated observable satellites outnumbers the sensors in the simulated SSN. Due to the shortage of sensors, the tasking methodology of the sensor tasking tool is strategic in its tasking against visible satellites. If a satellite is not observed by any sensor at a given time, the tool uses two body dynamics to propagate the satellite’s state through time until it is observed.
This research is valuable in that it presents a path forward for expanding situational awareness in the space domain. The ever-growing need for augmented SDA demands that the contemporary SSN expand to accommodate current and future technology to solve traditional coverage problems. This work aims to alleviate this issue by exploring space-based sensors working with the ground-based SSN.
Date of Conference: September 14-17, 2021
Track: Dynamic Tasking