Algorithmic Resource Allocation for Spacecraft Operations

The operation of any spacecraft requires a constant trade-off between available resources onboard the spacecraft such as power, the correct thermal operating range, downlink capacity, and payload stakeholder interests. On a commercial spacecraft, cost-efficient operations pose an additional requirement with significant influence on the success of the mission. On hosted payload missions, the interface and contractual constraints between the spacecraft operator and payload operator add to the challenges. Economic success calls for automated scheduling of operations and must consider all of the above constraints. This paper presents the algorithm-based optimization of the operational schedule for the wildfire detection satellite mission FOREST-1, the concept of which can be transferred to the operation of any Low-Earth-Orbit Earth observation satellite. The state of the art of generally applicable algorithms is presented and a comparison for the adaptability to the underlying problem statement is made. Compared algorithms include sequential, forward-chronological, linear search, and evolutionary algorithms. For this application, the simplex algorithm was chosen due to its capabilities regarding depleting one pivotal resource to maximize a mathematically defined gain to the mission. The implementation of this algorithm, which now is used to build the schedules of FOREST-1 regularly is presented. It is compared against the manual scheduling approach used during the commissioning phase in terms of controllability, flex-ibility, transparency, and efficiency. When used for scheduling weekly operations, the automatic scheduler achieves a reliable resource allocation of at least 98%, with an average cloud coverage of 2.5% and the highest value at 13% compared to around 80% utilization, 16.5% and up to 79% respectively. The benchmarks for the manual scheduling approach required 90 minutes on average while one execution of the automated scheduler required around 20 minutes. The manually generated schedules consist of 96% of requested sequences and only three out of 73 targets where chosen from areas of interest whereas the scheduler allocated 81.25% to areas of interest and 18.75% to requests.