Search

Les satellites d’observation de la Terre sont des senseurs qui acquièrent des données, les compressent et les mémorisent à bord, puis les vident vers le sol. Des incertitudes rendent la planification des activités au sol de plus en plus discutable car la planification est alors pessimiste et les plans largement sous-optimaux. Cette thèse détaille la conception d'une planification mixte qui permet de profiter de la réalisation des paramètres incertains à bord tout en préservant la prévisibilité de l'exécution pour les opérateurs au sol. Notre première contribution concerne le problème de planification des vidages. Un mécanisme de planification flexible a été conçu dans lequel seules les acquisitions de haute priorité sont planifiées de manière pessimiste. A bord, un algorithme adapte le plan en fonction des volumes réels, en s'assurant que le vidage des acquisitions de haute priorité est toujours garanti, et insère des nouveaux vidages si possible. Notre deuxième contribution concerne le problème de planification des acquisitions. Au sol, des contraintes contribuent à éliminer du plan de nombreuses acquisitions qui auraient pu être réalisées car les niveaux de ressources à bord sont souvent plus hauts que ceux prévus par ces contraintes. Dans un nouveau mécanisme de décision, le sol produit des plans conditionnels dans lesquels la réalisation des acquisitions de basse priorité est conditionnée par des niveaux d'énergie requis. Comparées à d'autres mécanismes de planification, ces deux approches flexibles permettent d'éviter le gaspillage des ressources et de réaliser plus d'acquisitions et de vidages tout en conservant de la prévisibilité. Earth-observation satellites are space sensors which acquire data, compress and record it on board, and then download it to the ground. Some uncertainties make planning and scheduling satellite activities offline on the ground more and more arguable as worst-case assumptions are made about uncertain parameters and plans are suboptimal. This dissertation details our efforts at designing a flexible decision-making scheme that allows to profit from the realization of uncertain parameters on board while keeping a fair level of predictability on the ground. Our first contribution concerns the data download problem. A flexible decision-making mechanism has been designed where only high-priority acquisition downloads are scheduled with worst-case assumptions. Other acquisition downloads are scheduled with expected parameters and conditioned by resource availability. The plan is then adapted on board. Our second contribution concerns the acquisition planning problem. A lot of acquisitions that could have been done are eliminated when planning because of worst-case assumptions. In a new decision-making scheme, these high-level constraints are removed for low-priority acquisitions. Observation plans produced on the ground are conditional plans involving conditions for triggering low-priority acquisitions. Compared with pure ground and pure onboard methods, these two approaches avoid wastage of resource and allow more acquisitions to be executed and downloaded to the ground while keeping a fair level of predictability on the ground.

Earth-observation satellites are space sensors which acquire data, compress and record it on board, and then download it to the ground. They evolve in a dynamic environment which implies several uncertain parameters such as cloud cover or available energy on board. These uncertainties make planning and scheduling satellite activities offline on the ground more and more arguable. Until now, plans are often built on the ground and are not to be modified on board. It means that in face of uncertainties, current plans need to be robust. It implies that on one hand, all the decisions are made offline on the ground and the satellite is a simple executive which neither makes, nor changes any decision and on the other hand, worst-case assumptions are made about uncertain parameters. is makes planning very pessimistic and plans suboptimal. This dissertation details our efforts at designing a exible decision-making scheme that allows to profit from the realization of uncertain parameters on board while keeping a fair level of predictability on the ground. Our first contribution concerns the data download problem. Sophisticated onboard algorithms compress low-interest zones of acquired images such as cloudy zones. The amount of data resulting from an acquisition is then unpredictable on the ground. Until now, all decisions regarding downloads are made on the ground, and data volumes are assumed to be maximum to make the plan robust. is makes plans highly suboptimal because real volumes are almost always lower than maximum and download windows are then underused. A flexible decision-making mechanism has been designed where only high-priority acquisitions are scheduled with worst-case assumptions. Other acquisition downloads are scheduled with expected volumes and conditioned by resource availability. The plan is then adapted on board: if acquisitions have a lower volume than predicted, acquisition downloads are added to the plan. If a low-priority acquisition has a higher volume than predicted, a look-ahead is performed to prevent future high-priority downloads to be endangered. This decision-making mechanism has been evaluated against pure ground and pure onboard approaches and showed great properties, ensuring predictability for the high-priority acquisitions and efficiency for the low-priority acquisitions while being computationally light on board. Our second contribution concerns the acquisition planning problem. When an acquisition plan is computed on the ground, resource availability such as memory or energy must be checked to ensure that the execution of the plan will not in endanger the satellite. From a technical point of view, because simulating the evolution of onboard energy and temperature is computationally expensive, it cannot be done finely during the planning process. It is replaced by high-level constraints with safety margins. The negative side of such a strategy is that the real energy profile is always better than predicted, and a lot of acquisitions that could have been done are eliminated when planning because of these high-level constraints. In a new decision-making scheme, these high-level constraints are removed for low-priority acquisitions. Observation plans produced on the ground are conditional plans involving conditions for triggering low-priority acquisitions. Compared with the current approach, this approach avoids wastage of resource and allows more acquisitions to be executed.

Earth-observation satellites are space sensors which acquire data, compress and record it on board, and then download it to the ground. They evolve in a dynamic environment which implies several uncertain parameters such as cloud cover or available energy on board. These uncertainties make planning and scheduling satellite activities offline on the ground more and more arguable. Until now, plans are often built on the ground and are not to be modified on board. It means that in face of uncertainties, current plans need to be robust. It implies that on one hand, all the decisions are made offline on the ground and the satellite is a simple executive which neither makes, nor changes any decision and on the other hand, worst-case assumptions are made about uncertain parameters. is makes planning very pessimistic and plans suboptimal. This dissertation details our efforts at designing a exible decision-making scheme that allows to profit from the realization of uncertain parameters on board while keeping a fair level of predictability on the ground. Our first contribution concerns the data download problem. Sophisticated onboard algorithms compress low-interest zones of acquired images such as cloudy zones. The amount of data resulting from an acquisition is then unpredictable on the ground. Until now, all decisions regarding downloads are made on the ground, and data volumes are assumed to be maximum to make the plan robust. is makes plans highly suboptimal because real volumes are almost always lower than maximum and download windows are then underused. A flexible decision-making mechanism has been designed where only high-priority acquisitions are scheduled with worst-case assumptions. Other acquisition downloads are scheduled with expected volumes and conditioned by resource availability. The plan is then adapted on board: if acquisitions have a lower volume than predicted, acquisition downloads are added to the plan. If a low-priority acquisition has a higher volume than predicted, a look-ahead is performed to prevent future high-priority downloads to be endangered. This decision-making mechanism has been evaluated against pure ground and pure onboard approaches and showed great properties, ensuring predictability for the high-priority acquisitions and efficiency for the low-priority acquisitions while being computationally light on board. Our second contribution concerns the acquisition planning problem. When an acquisition plan is computed on the ground, resource availability such as memory or energy must be checked to ensure that the execution of the plan will not in endanger the satellite. From a technical point of view, because simulating the evolution of onboard energy and temperature is computationally expensive, it cannot be done finely during the planning process. It is replaced by high-level constraints with safety margins. The negative side of such a strategy is that the real energy profile is always better than predicted, and a lot of acquisitions that could have been done are eliminated when planning because of these high-level constraints. In a new decision-making scheme, these high-level constraints are removed for low-priority acquisitions. Observation plans produced on the ground are conditional plans involving conditions for triggering low-priority acquisitions. Compared with the current approach, this approach avoids wastage of resource and allows more acquisitions to be executed.