Method for managing a satellite telecommunications system by optical link aided by meteorological models for continuity of service.

Abstract

A method of optical link communication between at least one satellite (10) and a plurality of transmitting and / or receiving ground stations (20) is described, comprising: obtaining cloud cover data from the ground stations, and data relating to an orbit of the satellite, determining, for each ground station (20), predictions on the possibility of establishing an uninterrupted optical link between the satellite and the ground station, for at least two determined time horizons, short term and long term, and selecting at least one of the ground stations (20) for establishing an uplink or downlink optical communication with the satellite, depending on the longest predicted duration for which it is possible to establish an uninterrupted optical link between the station and the satellite. Abstract figure: Figure 1

Description

Title of the invention: Method for managing a satellite telecommunications system by optical link aided by meteorological models for continuity of service. Technical field

[0001] The present disclosure relates to a method for establishing communications by optical link between at least one satellite and a set of ground stations. Prior art

[0002] It is known to transfer data between one or more ground stations and a satellite, with a view to retransmitting this data to receivers on the ground, via a radiofrequency link.

[0003] It may however be advantageous to replace a radiofrequency link with an optical link, which has the advantage of a much higher throughput, and also greater precision in the direction of the link, which guarantees greater security. Indeed, it is difficult to intercept an optical link without disrupting this link, and therefore making this intrusion detectable.

[0004] On the other hand, the use of an optical link between a satellite and a ground station has the disadvantage of increased risks of outages due to the passage of a cloud across the line of sight between the station and the satellite. This risk of outage is significantly higher than on radiofrequency links. Indeed, in radiofrequency links, rain can degrade the signal but does not cut it off entirely, unlike the passage of a cloud for an optical link. Since the passage of a cloud is also much more frequent, this risk of outage is significantly higher than on radiofrequency links. Consequently, to be able to ensure continuous service, it is necessary to use several ground stations in parallel and to determine at each moment the station which will be chosen to allow an optical link to be established with the satellite.

[0005] Document EP 2873172 discloses an optical communication system comprising a constellation of satellites, each satellite comprising inter-satellite optical telescopes and uplink / downlink optical telescopes for communications with terrestrial sites, and wherein the optical communication system is configured so that when the constellation of satellites passes in front of a given terrestrial site, one or more of the uplink / downlink telescopes of a satellite tracks at least two terrestrial optical telescopes of the given terrestrial site and switches between terrestrial optical telescopes so as to send data to the one having the clearest direct visibility of the given satellite. This process therefore makes it possible to take cloud cover into account when establishing communication between a terrestrial site and a satellite.

[0006] Also known from document AU2019363341 is a system for managing an optical terrestrial station comprising a device for monitoring low cloud cover, a device for determining high cloud cover from a meteorological satellite, and a device which superimposes the data on the low cloud cover and the high cloud cover and which controls the optical terrestrial station from this information. Summary

[0007] The present disclosure improves the situation. In particular, the present invention aims to improve the quality of service by choosing more efficiently which ground station will be responsible for establishing optical communication with the satellite.

[0008] In particular, an aim of the present disclosure is to enable the continuous maintenance of optical link communication between a plurality of ground stations and a satellite.

[0009] A method of communication by optical link between at least one satellite and a plurality of transmitting and / or receiving stations on the ground is proposed, the method being implemented by a computer and comprising: - obtaining updated cloud cover data specific to each ground station, - obtaining data relating to a satellite orbit, - the determination, for each ground station, taking into account data relating to the satellite orbit and by applying at least one risk model, of forecasts on the possibility of establishing an uninterrupted optical link between the satellite and the ground station, based on updated cloud cover data specific to each ground station, for at least two specific time horizons, short term and long term, said risk model being established on the basis of historical cloud cover data from ground stations, and - the selection of at least one of the ground stations for establishing an uplink or downlink optical communication with the satellite, this ground station being selected on the basis of the longest predicted duration for which it is possible to establish an uninterrupted optical link between the station and the satellite, from a given instant of establishment of the optical link, according to the forecasts on each of the two time horizons.

[0010] In embodiments, said risk model provides predictions on the possibility of establishing the uninterrupted optical link, associated with an index of confidence, the selection of said station also being carried out according to a duration of establishment of the future link and the confidence index, a more or less low weight being respectively attributed to the duration of establishment of the future link depending on whether the confidence index is respectively more or less important.

[0011] In embodiments, the orbit of the satellite is scrolling and imposes visibility time slots specific to each ground station, where communications are possible, said risk model being applied for a set of ground stations whose visibility time slots overlap.

[0012] In embodiments, the selection comprises several ground stations for establishing downlink optical communications with the satellite, these ground stations being selected on the basis of cumulative predicted durations for which it is possible to establish uninterrupted optical links between the ground stations and the satellite, from determined instants of establishment of optical links specific to each ground station and according to the forecasts, specific to each ground station, over two time horizons.

[0013] In embodiments, the forecast determination step uses two distinct risk models, of the neural network type, one of which is trained on the basis of historical data of local cloud covers of ground stations for the calculation of short-term forecasts and the other is trained on the basis of historical data of global cloud covers of ground stations for the calculation of long-term forecasts.

[0014] In embodiments, the method further comprises a step of updating the historical cloud cover data by integrating the updated cloud cover data and a step of updating said risk model.

[0015] In embodiments, the selection of at least one ground station comprises determining a communication sequence comprising a series of at least two ground stations used successively for establishing the uplink or downlink optical communication with the satellite, the sequence being determined so as to satisfy at least a first criterion relating to the absence of interruption of the optical communication during the sequence.

[0016] In embodiments, the sequence is determined to further satisfy a second criterion relating to the number of station changes during the sequence.

[0017] In embodiments, the updated cloud cover data obtained includes data acquired by ground-based cloud sensors, meteorological observation data obtained by at least one satellite, and data obtained by implementing numerical prediction models.

[0018] In embodiments, the plurality of ground stations comprises a so-called current station in optical communication with the satellite when implementing the method, and the determining step comprises: - a first determination of a possibility of establishing an optical link between the satellite and the current station for at least a determined time horizon, and; - if an optical link between the satellite and the current station is predicted to be impossible in the time horizon: • the implementation of a second determination of a possibility of establishing an optical link between the satellite and the other ground stations for at least two determined time horizons, this second determination being followed by the selection of one among the other ground stations for the establishment of communication by optical link.

[0019] In embodiments, the time of establishment of the optical link with the selected station is determined as a time prior to the time horizon for which an optical link is predicted to be impossible between the satellite and the current station.

[0020] In embodiments, at least one of the time horizons considered during the second determination ends later than the time horizon considered for the first determination, and the method comprising the implementation of a first model trained for the first determination relating to the current station, and a second model trained for the second determination relating to the other stations.

[0021] In embodiments, the method comprises: - obtaining cloudiness observation data acquired by a sensor for the current station, and implementing the first determination on the basis of said cloudiness observation data, and - obtaining meteorological observation data obtained by at least one satellite and covering a region including the other ground stations, and implementing the second determination on the basis of said meteorological observation data.

[0022] In embodiments, the data relating to cloud cover in the vicinity of a ground station comprises a time series of data relating to cloud cover in the vicinity of a ground station acquired at a determined frequency.

[0023] According to another object, a computer program product is described, comprising code instructions for implementing the method according to the preceding description, when it is implemented by a computer.

[0024] According to another object, there is described a system for communication by optical link between a satellite and a plurality of transmitting and / or receiving stations on the ground, the system comprising at least one control terminal comprising a computer and a memory, the control terminal being adapted to communicate with the stations on the ground, characterized in that it is configured to implement the method according to the preceding description.

[0025] The method described above advantageously makes it possible to select a transmitting or receiving station on the ground for establishing a communication by optical link, respectively uplink or downlink, with a satellite, the selection of the ground station being carried out on the basis of short-term and long-term estimates of probabilities of interruption of the optical link caused by cloud cover between all of the ground stations considered and the satellite.

[0026] Advantageously, it is possible not only to maintain a continuous communication link between the satellite and one of the ground stations, by efficiently selecting a station for which no cloud cover is expected, while minimizing the number of transitions from one station to another.

[0027] Another advantage of the present invention is that the calculation of a station availability for a satellite can be carried out for different time horizons, long term and short term and thus makes it possible to precisely adjust the transitions from one station to another as well as the number of transitions required. The availability calculation is here an estimation of the probability of availability at a given deadline.

[0028] In embodiments, the prediction can advantageously be carried out on the basis of local sky observation data from a ground sensor and on the basis of larger-scale data, acquired for example by satellite, which makes it possible to take into account two different observation scales and thus to obtain two long-term and short-term predictions with their respective accuracies. Alternatively, cloudiness prediction data, obtained by numerical models, can also be used to determine the possibility of establishing optical communication between the satellite and the ground stations. Brief description of the drawings

[0029] Other characteristics, details and advantages will appear on reading the detailed description below, and on analyzing the attached drawings, in which: Fig.l

[0030] [Fig.l] schematically represents an optical communication system according to an exemplary embodiment. Fig. 2

[0031] [Fig.2] schematically represents the main stages of a communication process- communication according to an example of implementation. Fig. 3

[0032] [Fig.3] schematically represents an example of selection of a sequence of ground stations for establishing optical communication. Fig. 4

[0033] [Fig.4] schematically represents an example of implementation of the selection of a sequence of ground stations using two trained models. Fig. 5

[0034] [Fig.5] represents an example of cloud cover data acquired by a cloud sensor. Fig. 6

[0035] [Fig.6] schematically represents an example of a model trained to predict the possibility of establishing an optical link between the satellite and a ground station based on cloud cover data. Fig. 7

[0036] [Fig.7] shows an example of a neural network structure that can be used to predict the possibility of establishing an uninterrupted optical link based on cloud cover data. Fig. 8

[0037] [Fig.8] illustrates results obtained by implementing the method according to an exemplary embodiment. Description of the embodiments

[0038] Reference is now made to [Fig.l], which represents a system 1 for communication by optical link between at least one satellite 10 and a plurality of ground stations 20. The ground stations 20 may be transmitting stations, the communication then being a so-called "uplink" communication in which data is sent by the ground stations to a satellite, or receiving stations, the communication then being called "downlink", in which data is sent by a satellite to the ground stations. Each ground station may be both a transmitter and a receiver, the roles of transmitter and receiver being variable depending on the communication needs.

[0039] In exemplary embodiments, the optical link communication may comprise an optical uplink communication between one or more first transmitting ground stations and the satellite, and then an optical downlink communication between the satellite and one or more second receiving ground stations. Alternatively, and as shown in [Fig.l], the communication may comprise an optical uplink communication between stations ground transmitters and the satellite, then downlink communication by another means, typically by radiofrequency communication, between the satellite and ground stations or user terminals 40. According to yet another variant, there may be uplink radiofrequency communication, and downlink communication by optical link between the satellite and the ground stations.

[0040] The ground stations 20 are advantageously in communication with each other by a terrestrial link, for example a wired communication of the optical fiber type.

[0041] The method according to the invention can be implemented for communication by optical link with one or more geostationary satellites or via moving constellations. The satellites used to establish the optical connection link can have an orbit of the GEO (geostationary), MEO (medium earth orbit) or LEO (low earth orbit) type. In the case of a non-geostationary satellite, the visibility windows of each ground station by each satellite are a function of the positions of the satellites in orbit and are precisely known and determined in advance. A moving satellite can in particular control its pointing continuously towards a ground station, during a determined period of visibility between the satellite and this ground station. The pointing of a satellite towards a ground station is for example carried out by controlling the attitude of the satellite.It is also possible to envisage that the optical communication head on board the satellite is oriented relative to the body of the satellite, to point towards a ground station. The pointing of a satellite is modified successively to point successively towards different ground stations. The optical communication head comprises for example an optical telescope. In any case, the information relating to the orbit of the satellite(s), and therefore to the relative position of the satellite(s) with respect to the ground stations, is known.

[0042] The communication system further comprises, for example, a control terminal 30, comprising at least one computer 31 and a memory 32, the control terminal 30 being adapted to communicate with the ground stations and to implement the method described below. The computer 31 may comprise one or more processors, microprocessors, microcontrollers, graphics processor, etc. The memory 32 comprises, for example, a non-volatile memory for storing code instructions that are executed by the computer for implementing the method. In addition, the memory stores, for example, at least one previously trained model for predicting the possibility of establishing an uninterrupted optical link between the satellite and at least one ground station, as described in more detail below.

[0043] In exemplary embodiments, the optical link communication system 1 further comprises at least one cloudiness sensor 50 positioned in the vicinity of at least one of the ground stations 20. A cloudiness sensor 50 is a ground sensor, oriented towards the sky, and adapted to acquire data relating to the coverage instantaneous cloudiness, that is to say at the time of observation, in the portion of the sky located in the field of vision of the sensor. For example, such a sensor may comprise an infrared sensor associated with a computer configured to determine, from the information from the infrared sensor, optical thickness information represented by the cloud cover. With reference to [Fig. 5], an example of data acquired by a cloudiness sensor is shown. The field of vision of the sensor corresponds to the circular area in the center of the figure. Each pixel contained in this circular area corresponds to a direction, defined by azimuth and elevation values, relative to the sensor. In addition, each pixel is associated with an optical thickness which depends on the thickness of the cloud layer present on the line of sight corresponding to this azimuth and this elevation. In [Fig.5], optical thicknesses are represented in shades of gray, but the information relating to the optical thickness at each pixel can take the form of a scalar value between two predetermined minimum and maximum limits.

[0044] In the case where the system 1 comprises one or more cloudiness sensors 50, the control terminal 30 is advantageously in communication with this or these sensors, for example by a wired connection, this communication being able to take place in real time or at a determined frequency.

[0045] In addition, the control terminal 30 may also receive cloud cover data acquired by a weather observation satellite 2, or cloud cover data corresponding to a weather forecast generated by a digital model (not shown).

[0046] Each ground station may also be in communication with a proximity sensor, such as a ground cloud sensor, and a dedicated local processing terminal. This local processing terminal then makes it possible to estimate the probabilities of feasibility of the optical link with the satellite at different time horizons. These probabilities are transmitted to another global processing terminal which will be responsible for deciding and selecting the ground station which will have to establish an optical link with the satellite during a future time period. This advantageously makes it possible to minimize the volumes of data transmitted between the ground stations, their proximity sensor and the station management center.In the case of a risk model to assess a short-term local probability of optical link establishment, such as those based on neural networks, the risk model of each ground station will for example be defined based on the historical data of local cloud cover specific to each ground station.

[0047] One could also consider sharing local cloud cover history data, generated by several proximity sensors to define a risk model shared between several stations, to establish short-term forecasts.

[0048] With reference to Figures 2, 3 and 4, a method of optical link communication between the satellite and the ground stations will now be described. This method makes it possible to select at least one ground station 20 from a plurality of stations to establish uninterrupted optical link communication with the satellite 10, on the basis of cloud cover data of the ground stations. In exemplary embodiments, the method makes it possible to select a sequence S of several ground stations 20 successively used for establishing the optical communication, so as to ensure continuous communication despite cloud cover of certain stations, this cover being variable over time.

[0049] The method comprises obtaining 100 data relating to the cloud cover from a set of ground stations 20. This data may for example comprise, for a given ground station, an indication of the cloud cover for a plurality of directions from the station, each direction being determined by azimuth and elevation values. In addition, this data may comprise, for a given ground station, and where appropriate for at least one determined direction from this station, a time sequence of data describing the temporal evolution of the cloud cover, the time sequence covering a determined duration and comprising data acquired at a determined frequency. This frequency may for example be between one acquisition every thirty seconds and one acquisition every ten minutes. In addition, the determined duration is for example between a few minutes and several hours.In exemplary embodiments, data relating to the cloud cover of the ground stations are continuously acquired at a determined frequency, and transmitted to the control terminal 30 in real time or at a fixed time interval.

[0050] The cloud cover data advantageously comprise image-type data acquired by cloud sensors located in the vicinity of the ground stations considered. By "in the vicinity" is meant that the distance between the cloud sensor and the ground station is sufficiently small so that the cloud cover information from the sensor is also applicable to the station. Typically, this distance is less than 100 meters, advantageously less than 50 meters, very advantageously between 0 and 10 meters.

[0051] As indicated previously, such a sensor 50 provides, for a plurality of directions relative to the sensor, an optical thickness value. The optical thickness provided by the sensor makes it possible to directly determine the attenuation of an optical signal caused by clouds by the following formula:

[0052] attenuation = 101og] {)(exp( - OD))

[0053] Where OD is the optical thickness and where “attenuation” is the attenuation evaluated in decibels. Optical communication is typically considered possible when cloud attenuation is below a threshold of 3 dB. When attenuation is higher, optical communication can be considered impossible. However, increasing the transmission power could compensate for a greater loss. Furthermore, there are other disturbances to consider, such as turbulence, unrelated to optical thickness, which could justify taking more margin.

[0054] In addition, or alternatively, the data relating to cloud cover comprise data, of the image type, of cloud cover observation acquired by at least one satellite 2 observing a region of the Earth integrating one or more ground stations of the set considered. The data obtained by a meteorological observation satellite 2 may also comprise data of the infrared image type which are processed to calculate an optical thickness represented by the cloud cover, which is as previously converted into an attenuation value caused by the clouds.

[0055] An observation satellite has, for example, a resolution of the order of 0.05 degrees or even lower, which is equivalent to pixels of size 5km or even of the order of a kilometer, at the equator, the pixels representing larger areas as one moves away from the equator.

[0056] In exemplary embodiments, the data obtained include both data acquired by cloud cover sensors 50 near the ground stations and data obtained by a weather observation satellite 2. Indeed, the cloud cover sensors are able to provide local data, particularly relevant in the short term, while satellite data are regional data making it possible to determine longer-term trends, which are thus complementary. In addition, the cloud cover sensors providing local data allow a more precise evaluation of the coverage on the line of sight between the ground station and the satellite.

[0057] In addition, or alternatively, the cloud cover data may also include cloud cover forecast data for a given geographic area, these data being obtained by applying numerical simulation models, such as the models used for weather forecasting.

[0058] Furthermore, the method also comprises obtaining 200, by the control terminal 30, data relating to an orbit of the satellite in question, this data making it possible to determine, taking into account the geographical locations of the ground stations, a subset of ground stations with which the satellite is likely to establish communication by optical link in the absence of cloud cover. For example, the satellite stores data accumulated over several different orbits corresponding to several orbital periods. For example, in the case of a LEO satellite, the satellite regularly shifts relative to the Earth from one orbit to another. For example, the data stored on board the satellite can be emptied once or twice a day, which amounts to establishing an optical link approximately every 15 orbits for a satellite at an altitude of approximately 600 km, considering, for example, around ten candidate ground stations during that day. This subset of stations can be variable over time in the case of a moving satellite, i.e. non-geostationary, but in this case, since the orbit of the satellite is known, the composition of the subset of stations with which communication by optical link is possible, in the absence of clouds, is known for the current position of the satellite as well as for future positions.The challenge is then, among all the candidate ground stations, to select enough ground stations with the choice of the associated communication slot during the satellite pass, in order to guarantee a total communication duration on the horizon, for example to guarantee the emptying from the satellite of a pre-established volume of data. The planner then ensures that the stations are available, and reserves enough communication slots where the feasibility prediction is good to guarantee the emptying. Overbooking can be considered to compensate for cases where the prediction would be erroneous and where the planned communication would not take place due to a link blocked by clouds.Thus, ground stations can be selected on the basis of cumulative predicted durations for which it is possible to establish uninterrupted optical links between ground stations and a satellite, from determined instants of establishment of optical links as a function of visibility slots and as a function of cloud cover forecasts, taking into account two time horizons.

[0059] For example, the data relating to the orbit of the satellite can be stored in the memory 32 and the computer 31 can access it in step 200 to determine, from among the set of stations for which data relating to the cloud cover are obtained in step 100, a subset of stations with which the satellite can establish communication by optical link during a determined time interval. The time interval is for example a time interval of a determined duration from the moment at which the method is implemented.

[0060] The computer 31 then determines during a step 300, from the cloud cover data obtained, forecasts relating to the possibility of establishing an uninterrupted optical link between the satellite and at least one ground station, for at least two determined time horizons, respectively short-term and long-term. The terms “short-term” and “long-term” are interpreted in a relative manner, that is to say that the first time horizon is shorter than the second. As by way of non-limiting example, the first time horizon may be between 0 and 10 minutes, for example between 30 seconds and 5 minutes. The second time horizon, long term, may be between 30 minutes and 3 hours, or even between 30 and 90 minutes. In exemplary embodiments, this step is implemented for each ground station of the subset of stations determined from the information obtained on the position of the satellite.

[0061] In the following, the term “time horizon” is used to describe a time interval between a current instant, typically corresponding to the instant of implementation of the method, and a future instant, the time interval being of determined duration. It is therefore understood that the “long-term” time horizon preferably strictly includes the “short-term” time horizon.

[0062] The computer 31 is therefore configured to predict the risk that there is at least one interruption of the optical link between the satellite 10 and the ground station 20 considered during this time interval. In exemplary embodiments, the computer 31 is configured to determine a probability that there is at least one interruption during the time horizon considered, and apply a thresholding to the probability to obtain a binary indication on the possibility, or not, of establishing the optical link continuously during the time horizon.

[0063] This step 300 of determining the forecasts relating to the possibility of establishing an uninterrupted optical link is implemented by applying, to the cloud cover data obtained in step 100, at least one risk model M, for example by a neural network previously trained on historical cloud cover data for a set of ground stations. In exemplary embodiments, the historical cloud cover data used for training the model comprises historical cloud cover data from all the ground stations for which current cloud cover data is obtained in step 100. Alternatively, this historical data may relate to ground stations different from those for which the data relating to cloud cover is obtained in step 100.

[0064] The historical cloud cover data used for training the risk model M advantageously comprise, for a ground station in question, time sequences of data comprising an indication of cloud cover or attenuation determined, at a determined sampling frequency, for example between a few seconds and a few minutes, for example of the order of 30 seconds to 2 minutes. In exemplary embodiments, these time sequences of data comprise indications of cloud cover or attenuation, determined for a plurality of directions from the ground station in question. Furthermore, as indicated previously with regard to step 100 obtaining cloud cover data, these historical cloud cover data may be data obtained by ground-based cloud sensors near the stations considered, and / or meteorological observation data obtained by satellite.

[0065] In exemplary embodiments, the step 300 of determining the forecasts relating to the possibility of establishing an uninterrupted optical link is implemented by applying two distinct models, for example by neural networks, in which a first model is configured for the calculation of short-term forecasts, i.e. at the first time horizon, and the second model is configured for the calculation of long-term forecasts, i.e. for at least a second time horizon subsequent to the first. Thus, the horizons of each of the two models advantageously allow two specific uses. The first short-term model makes it possible to determine the risk of outage, for example on a link currently in use, and provides greater precision, while the second, long-term model aims, for example, to make a first selection of a new station, when a significant risk of outage appears on the current station.The network, which uses short-term information and long-term information, advantageously allows minimizing the number of changes of current ground stations. These risk models can be established from the same historical cloud cover data. Alternatively, the risk model configured for the calculation of short-term forecasts can be established from local historical data, i.e. obtained by ground-based cloud sensors, while the risk model configured for the calculation of long-term forecasts can be established from regional historical data, i.e. satellite observation data.

[0066] With reference to [Fig. 3], step 300 of predicting the possibility of establishing an uninterrupted optical link between the satellite and ground stations makes it possible to obtain, for each ground station, represented by the terms OGS1.. .OGSn, an indication relating to the possibility of establishing a link with the satellite in an uninterrupted manner for the at least two time horizons considered. In [Fig. 3], a number n of ground stations and a set of time horizons Hl,...Hj are schematically represented.

[0067] In Figures 3 and 4, the moments when communication by optical link is possible between the satellite 10 and the station 20 in question are shown in white, and the moments when this communication is impossible are shown in black.

[0068] Step 300 of determining the forecasts relating to the possibility of establishing an uninterrupted optical link therefore makes it possible to determine, for example: - that an uninterrupted optical link is possible for a short-term time horizon H1 for the first station, - that this is not the case for the second station, - that an uninterrupted optical connection is also possible for a horizon long-term H2 time for the last station.

[0069] The method then comprises a step 400 of selecting at least one ground station 20 for establishing a communication by optical link with the satellite 10, from the forecast information relating to the possibility of establishing an uninterrupted optical link obtained in step 300 for each of the time horizons considered. The station is selected on the basis of the longest duration forecast, considering the long-term forecasts, for which it is possible to establish an uninterrupted optical link between the station and the satellite, from a determined instant of establishment of the optical link. This instant of establishment of the optical link may be a current instant, in particular in the case where the station considered is a so-called “current” station, for which a communication by optical link is already established with the satellite and for which the risk of outage is close, according to the short-term forecasts.Alternatively, this establishment time may be a later time, which lies between the current time and a time horizon for which a risk of outage is established for the current station or, in other words, at which the optical link is predicted to be impossible, according to short-term forecasts.

[0070] Thus, in exemplary embodiments, the step 400 of selecting a ground station 20 for establishing a communication by optical link with the satellite 10 comprises the determination of a communication sequence S comprising a series of at least two ground stations 20 used successively for establishing a communication by optical link with the satellite 10, the sequence S being determined so as to satisfy a criterion relating to the absence of a break in the communication during the sequence. Thus, the sequence may for example comprise the determination of a first ground station with which the communication is established, for which no break is planned during a first time horizon, then a second station for which no break is planned between the end of the first time horizon and the end of a second time horizon, subsequent to the first.

[0071] Furthermore, and in order to limit the transitions from one station to another, the sequence S can also be determined so as to respect a second criterion relating to the number of station changes during the sequence, this second criterion being for example to minimize the number of station changes, or to respect a maximum number of station changes, during the sequence. In this case, the first criterion relating to the absence of cutoff during the sequence has priority over the second criterion.

[0072] For example, still with reference to [Fig.3], a sequence S, represented by arrows, is determined in which: - the first ground station 0GS1 is used for communication until a time corresponding to a risk of communication being cut off, and - from this moment, the communication is switched to the second OGS2 station for which no risk of interruption is identified from this moment and until a later time horizon.

[0073] It is noted that the third OGSn station which is represented is not selected because, although no risk of outage is predicted for this station at the time when the risk of outage is established for the first station, the time horizon for which the link is uninterrupted is shorter than for the second station, which implies more changes of stations.

[0074] With reference to [Fig. 4], in an exemplary embodiment, the step 300 of determining predictions on the possibility of establishing an uninterrupted optical link between the satellite 10 and each of a plurality of ground stations 20 comprises a first determination 310 of the possibility of establishing an uninterrupted optical link between the satellite and a single ground station, which is the current station with which an optical link is already established, for at least one determined time horizon.

[0075] In [Fig.4], this first determination has been represented by the arrows “p”, which is implemented iteratively at a determined frequency (represented by the delta interval in [Fig.4]), as long as the optical link with the current station is predicted as possible over the time horizon considered. Preferably, the frequency at which this first determination is implemented corresponds to a time interval between two iterations which is less than the time horizon Hp considered for the prediction. Thus in [Fig.4], we observe that the fourth iteration of this determination for the current station is the last, because this iteration determines a risk of cutting off the optical link in the time horizon considered.

[0076] In this case, when a continuous optical link is predicted as impossible for the time horizon considered, the determination step comprises a second determination 320 of a possibility of establishing a communication by optical link between the satellite and the other ground stations for at least two time horizons considered, comprising a short-term time horizon and a long-term time horizon. We then return to step 400 of selecting, from among said other stations, at least one station 20 for establishing the optical link, the switch between the current station and the selected station taking place before the end of the time horizon Hp considered for the first determination and preferably as soon as the second station has been selected. Also in this case, step 400 of selecting a ground station 20 for establishing a communication by optical link with the satellite 10 can understand the selection of a sequence S of stations satisfying the criterion(s) mentioned above.

[0077] Returning to the example of [Fig.4], the implementations of the second determination for the stations other than the initially current station have been indicated by arrows "dm". It can be seen that stations No. 2 and n represented both have a short-term time horizon corresponding to a possible optical link at the time of this second determination, but that station No. 2 also has a long-term horizon corresponding to a possible optical link, unlike station n. Consequently, it is the second station which is selected.

[0078] This is not limited to a station in optical connection with a satellite. The method can be applied to N stations in service among M potential stations (with N < M). The invention thus makes it possible, more generally, to choose the set of stations for which the service will continue, and the ground stations used for which it is necessary to find, for each, an available ground station which will take over the associated optical communications.

[0079] Furthermore, once this station is selected, the method returns to the first determination 310 for the current station, which is iterated again until a risk of outage is established for the station considered, as represented by the repetition of the arrows "p" for the second station OGS2. Although the steps 100 of obtaining data relating to cloud cover and 200 of obtaining data relating to the orbit of the satellite are not shown in FIGS. 3 and 4, it is understood that the set of ground stations considered for the potential establishment of an optical link is updated over time, as a function of the data relating to the position of the satellite obtained in step 200, and that the cloud cover data for these stations is also updated by repetitions of the step 100 of obtaining data relating to cloud cover.

[0080] In an exemplary embodiment, the first determination 310 relating to the current station is implemented by applying a first model, this first model having been previously trained on historical cloud cover data from a set of stations, to predict the possibility of establishing an uninterrupted optical link between the satellite and the same given station during a determined time horizon. This time horizon may be short-term, and being shorter than the longest time horizon considered during the second determination 320 relating to the other stations, the historical cloud cover data used for training the model, then the cloud cover data used for inference of the model, may be local data obtained by a cloud sensor close to the station in question.This example does not, however, exclude that the training data and the data used for inference include . also satellite observation data and / or data obtained by numerical simulation.

[0081] In this case, as the first determination 310 is implemented for a current ground station for which communication by optical link is already established with the satellite, the learning database can exclusively comprise data sequences covering a duration greater than or equal to the time horizon considered, and comprising data acquired at a determined frequency during this duration, the first data of the sequence of which correspond to an absence of cloud cover, i.e. to a possible optical link between the satellite and the station.

[0082] Furthermore, the second determination 320 relating to the other stations and the selection of a second station to continue the communication can be implemented by applying a second model, trained to perform these tasks. Since the second determination is implemented by considering at least two time horizons, at least one of which preferably ends later than the time horizon considered for the first determination, the historical cloud cover data used for training the model, then the cloud cover data used for inference of the model, can be regional data obtained by a meteorological observation satellite. These data can be supplemented by data acquired by cloud sensors and / or data obtained by numerical simulation.In the case of local sensors, such as cloud sensors, the volume of data generated is proportional to the number of sensors for the stations concerned, unlike the first model which uses satellite data. We thus exploit the data generated at the ground station to anticipate its switchover and the data generated at all the candidate free stations to serve as relays. Alternatively, the historical cloud cover data and the cloud cover data used respectively for model training and inference can be cloud sensor data, which allows the same data to be used for both models.

[0083] Alternatively, the first determination 310 relating to the current station, the second determination 320 relating to the other stations and the selection 400 of a ground station 20 for establishing a communication by optical link can be implemented by the same model which may have been the subject of a first training for example for the second determination 320 and the selection 400 of a ground station 20 for establishing a communication by optical link, then of reinforcement learning to implement the first determination 310 with precision. Reinforcement learning allows an agent to learn how to operate the system. Thus, such an agent could decide, at each time step, to stay or change stations, such an agent also being able to choose the next station. Such learning would include in particular the risk of cut-off which would correspond to a very strong penalty case for the agent in question. In this case, the same agent would translate a risk model over both time horizons, short term and long term.

[0084] The method may also include updating the historical cloud cover data as new data is obtained, as well as updating the model(s) presented above, based on this data and the deviation between the model predictions and the reality of the cloud cover data observed after implementing the method.

[0085] With reference to figures 6 and 7, we will now describe an example of implementation of models used for the first determination 310 relating to the current station on the one hand, and for the second determination and the selection of a station on the other hand.

[0086] The data used as input to each model are data sequences, for example of the image type, acquired at an acquisition frequency of 1, for example every 30 seconds, by cloudiness sensors located near respective ground stations. The data sequences cover, for example, a period of 5 minutes. The pixels of the images comprise, for example, colors representative of the observed cloudiness. Each pixel represents, for example, a pair (azimuth, elevation). Each pixel corresponds, for example, to an optical thickness (a scalar between 0 and 5) which depends on the thickness of the cloud present on each pixel.

[0087] The data is for example preprocessed to remove unnecessary areas from the images, i.e. at least the edges of the image outside the circular area visible in the example of [Fig.5].

[0088] Preprocessing can also be carried out, for example, to extract from the data sequence a succession of samples corresponding, for at least one given direction relative to the sensor, to the time series of data acquired by the sensor for this direction. This preprocessing can, for example, be implemented by using the TFRecord data format of the TensorFlow software suite, allowing the training samples to be stored in an optimized manner.

[0089] Earth observation satellite data are, for example, of the image type in the visible or invisible spectrum, such as infrared rays. Satellite images are processed by known methods for generating satellite images. The satellite images provided are, for example, centered around the point of interest, which is the station. The satellite images used are, for example, in the form of a matrix.

[0090] Data, for example of the raster image type, used for training neural networks, for updating neural networks and for predictions made with neural networks, are for example of the same type. Updated data is for example used to calculate an inference of a network and make a prediction. Archived data is used to calculate inferences in the past, in order to verify predictions. Gradient descent and associated algorithms are for example used for the training phase. To do this, TensorFlow implements for example automatic differentiation, which uses calculation to determine gradients, in general, to calculate the gradient of the error. The error calculated at the output is for example back-propagated in the neural network, during the training phase or during an update of the neural network.

[0091] The two models used can be implemented, for example, by neural networks, for example of the convolutional type, receiving as input the preprocessed image sequences mentioned above, referenced “I”, presented for example in the form of a matrix and producing as output a matrix of scalars between 0 and 1, where each element of the matrix corresponds to the same pixel position of the input images, therefore for example to the same direction relative to the cloudiness sensor, and determining whether the link is feasible or not over time intervals of [t, t+k] minutes with k G {1.5, 10, 15, 20, 30, 60, 90} minutes. Each time interval therefore corresponds to a time horizon considered by the model.

[0092] An example of a neural network structure that can be used is shown in [Fig.7]. This network comprises a succession of convolutional layers (the number of input channels "ch" and the dimension of the convolution kernel "kernel" are indicated under each layer) followed by an activation layer involving, for example, the ReLu function. The number of channels at the network output indicated by "frame_out" corresponds to the number of intervals that we wish to predict at the network output. The network is therefore configured so that there remain "frame_out" layers (or "charnels" in English) at the network output and thus have the desired output format. For example, if we have the intervals [0.5min], [0.10min] and [0.15min] at the output, the frame_out will be 3. The network can also include a cropping layer to reduce the output to pixels corresponding to a direction greater than a given elevation threshold.The output of the network corresponds to the output of a sigmoid function. In this example, the network inputs therefore include 11 matrices of real numbers, representing the optical thickness on each pixel of an image acquired by a cloud sensor, of size 460x640. The 11 matrices correspond to the images acquired every 30 seconds for five minutes between the last five minutes and the current time when the forecast is made. The network outputs include a number of matrices equal to the number of time horizons. considered, for example 8 according to the example above, each matrix comprising scalar elements between 0 and 1 estimating the probability that the line of sight in the direction corresponding to the element of the matrix is continuously cloud-free over the time horizon considered.

[0093] As indicated above, the learning database for training the first model could for example be reduced to data for which the optical link is initially possible.

[0094] Furthermore, for the first model which predicts a risk that an optical link is impossible over at least one time horizon considered, a threshold is for example applied to the scalar output to raise an alert, which triggers the call to the second model.

[0095] On the other hand, for the second model, an execution of this model is made for each station other than the current station, and the station chosen can be the one minimizing this scalar output, which is therefore not subject to thresholding.

[0096] With reference to [Fig.8], the results obtained by the example presented above are shown, the curves representing the probability P that a communication by optical link is interrupted during a simulation lasting 3 years, with the implementation of the method described above (curve A) to choose a sequence of ground stations, compared to a simple persistent model (curve B) and to a sequence obtained with perfect knowledge of the future cloudiness data (curve C). The curves are shown as a function of the number Nb of ground stations. For this simulation, dates were randomly drawn and checked whether the dm caused a cut-off or not, the dates drawn being generated over 3 years of real data.

[0097] It is noted that the performances obtained are better than those of the persistent model and improve with the number of stations, to be quite close to the optimal theoretical performances corresponding to curve C.

Claims

Claims

1. Method of communication by optical link between at least one satellite (10) and a plurality of transmitting and / or receiving stations on the ground (20), the method being implemented by a computer and comprising: - obtaining (100) updated cloud cover data specific to each ground station, - obtaining (200) data relating to an orbit of the satellite, - determining (300), for each ground station (20), taking into account the data relating to the orbit of the satellite and by applying at least one risk model, forecasts on the possibility of establishing an uninterrupted optical link between the satellite and the ground station, as a function of the updated cloud cover data specific to each ground station, for at least two determined time horizons, short term and long term, said risk model being established on the basis of historical cloud cover data of the ground stations,and - selecting (400) at least one of the ground stations for establishing an uplink or downlink optical communication with the satellite, this ground station being selected on the basis of the longest predicted duration for which it is possible to establish an uninterrupted optical link between the station and the satellite, from a determined instant of establishment of the optical link, according to the forecasts on each of the two time horizons.,

2. Method according to claim 1, in which said risk model provides predictions on the possibility of establishing the uninterrupted optical link, associated with a confidence index, the selection of said station being carried out also as a function of a duration of establishment of the future link and of the confidence index, a more or less low weight being respectively attributed to the duration of establishment of the future link depending on whether the confidence index is respectively more or less important.

3. A method according to claim 1 or 2, wherein the orbit of the satellite is scrolling and imposes visibility time slots specific to each ground station, where communications are possible, said risk model being applied for a set of ground stations whose visibility time slots overlap.

4. A method according to claim 3, wherein the selection comprises a plurality of ground stations (20) for establishing downlink optical communications with the satellite (10), said ground stations being selected on the basis of cumulative predicted durations for which uninterrupted optical links can be established between the ground stations and the satellite, from determined instants of establishment of optical links specific to each ground station and according to the forecasts, specific to each ground station, over two time horizons.

5. Method according to one of the preceding claims, in which the step of determining (300) forecasts uses two distinct risk models, of the neural network type, one of which is trained on the basis of historical data of local cloud cover of the ground stations for the calculation of short-term forecasts and the other is trained on the basis of historical data of global cloud cover of the ground stations for the calculation of long-term forecasts.

6. Method according to one of the preceding claims, further comprising a step of updating the historical cloud cover data by integrating the updated cloud cover data and a step of updating said risk model.

7. Method according to one of the preceding claims, in which the selection (400) of at least one ground station comprises the determination of a communication sequence (S) comprising a series of at least two ground stations (20) used successively for establishing the uplink or downlink optical communication with the satellite (10), the sequence being determined so as to satisfy at least a first criterion relating to the absence of interruption of the optical communication during the sequence.

8. Method according to the preceding claim, in which the sequence (S) is determined so as to further satisfy a second criterion relating to the number of station changes during the sequence.

9. A method according to any preceding claim, wherein the obtained updated cloud cover data comprises data acquired by ground-based cloud sensors (50), data meteorological observation data obtained by at least one satellite, and data obtained by the implementation of digital prediction models.

10. Method according to one of the preceding claims, in which the plurality of ground stations (20) comprises a so-called current station in communication by optical link with the satellite during the implementation of the method, and the determination step comprises: - a first determination (310) of a possibility of establishing an optical link between the satellite and the current station during at least one determined time horizon, and; - if an optical link between the satellite and the current station is predicted as impossible at the time horizon: • the implementation of a second determination (320) of a possibility of establishing an optical link between the satellite and the other ground stations during at least two determined time horizons, this second determination being followed by the selection of one among the other ground stations for the establishment of the communication by optical link.

11. Method according to the preceding claim, in which the instant of establishment of the optical link with the selected station is determined as an instant prior to the time horizon for which an optical link is predicted to be impossible between the satellite and the current station.

12. Method according to claim 10 or 11, in which at least one of the time horizons considered during the second determination ends later than the time horizon considered for the first determination, and the method comprising the implementation of a first model trained for the first determination relating to the current station, and of a second model trained for the second determination relating to the other stations.

13. A method according to one of claims 10 to 12, comprising - obtaining cloudiness observation data acquired by a sensor for the current station, and implementing the first determination (310) on the basis of said cloudiness observation data, and - obtaining meteorological observation data obtained by at least one satellite and covering a region including the other ground stations, and implementing (320) the second determination on the basis of said meteorological observation data.

14. A method according to one of the preceding claims, wherein the data relating to cloud cover in the vicinity of a ground station comprises a time series of data relating to cloud cover in the vicinity of a ground station acquired at a determined frequency.

15. Computer program product, comprising code instructions for implementing the method according to one of the preceding claims, when implemented by a computer (31).

16. Optical link communication system between a satellite (10) and a plurality of transmitting and / or receiving stations on the ground (20), the system comprising at least one control terminal (30) comprising a computer (31) and a memory (32), the control terminal being adapted to communicate with the ground stations, characterized in that it is configured to implement the method according to one of claims 1 to 14.

Images