Satellite multi-granularity observation model selection method for satellite data-real fusion test

Abstract

The invention discloses an on-satellite multi-granularity observation model selection method for a satellite digital-real fusion test, and belongs to the field of spacecraft digitization and computer science, and the method comprises the steps: designing a digital-real fusion test process of an on-satellite multi-granularity observation model; the method comprises the steps of demand preparation before testing, digital twinborn observation model granularity division, on-satellite observation model granularity selection and data-real fusion on-satellite observation model correction. And designing a granularity selection method of the on-satellite observation model, and realizing on-satellite real-time selection of the model granularity of the observation model by judging the emergency degree of the on-satellite observation task, the calculated amount demand and the calculation error allowable range.

Description

Technical Field

[0001] The invention belongs to the field of spacecraft digitization and computer science, and specifically relates to an on-board multi-granularity observation model selection method for satellite digital-real fusion testing. Background Art

[0002] With the rapid development of space science, important space assets such as space stations and relay satellites need to be monitored for a long time to avoid losses. Observation satellites observe and track approaching targets and then respond to protect space assets. Space observation missions have become an important area of ​​concern for space activities. Observation satellites have fast response to autonomous onboard decisions. The core of autonomous decision-making of observation satellites is the onboard computing model, and its reliability and calculation accuracy directly affect the effectiveness of observation missions. In order to control the risks of space assets and increase the autonomous response capabilities of observation satellites, it is of great value to conduct research on onboard observation models.

[0003] The current research on onboard observation models has the following problems: ① The two existing mainstream observation model schemes cannot achieve a balance between decision-making speed and certainty, and there are contradictions between them. The highly programmed observation model scheme relies on a complete and formal decision-making process, emphasizing the use of recognized models to output decisions after collecting complete data. It has strong reliability but takes a long time to make decisions. The other intuitive or heuristic observation model scheme has low computational requirements and emphasizes rapid decision-making and rapid response, but it is generally recognized that such models are not reliable enough and the risk of decision-making is relatively high. However, in some situations, a small amount of decision-making deviation caused by rapid decision-making can be allowed, and too long decision-making time is also a risk in a changing emergency response situation. ② For the onboard observation model of the satellite, it is currently pre-stored in the onboard computer before launch, or injected through the ground station before a specific task, which cannot meet the needs of multiple dynamic emergency tasks and lacks model correction and processing during the task process. Summary of the invention

[0004] The technical problems to be solved by the present invention are as follows: ① The two existing mainstream observation model schemes cannot achieve a balance between decision-making speed and certainty, and there are contradictory contradictions. The highly programmed observation model scheme relies on a complete and formal decision-making process, emphasizes the use of recognized models to output decisions after collecting complete data, has strong reliability but takes a long time to make decisions, and the other intuitive or heuristic observation model scheme has low computational requirements and emphasizes rapid decision-making and rapid response, but it is generally recognized that such models are not reliable enough and the risk of decision-making is relatively large. However, in certain situations, a small amount of decision-making deviation caused by rapid decision-making can be allowed, and too long decision-making time is also a risk in a changeable emergency response situation. ② For the on-board observation model of the satellite, it is currently pre-stored in the on-board computer before launch, or injected through the ground station before a specific task, which cannot meet the needs of multiple dynamic emergency tasks and lacks model correction and processing in the task process. In response to the above problems, the present invention improves and designs the on-board observation model method with the help of digital-real fusion and multi-granularity modeling related technologies, and to a certain extent improves the autonomous response capability of the observation satellite.

[0005] The present invention solves the technical problem by adopting the following technical solution: a method for selecting a multi-granularity observation model on board for satellite digital-real fusion testing, characterized in that the specific method flow is as follows:

[0006] Step S101, building a digital twin observation scene for satellite digital-physical fusion testing and matching a digital twin satellite entity model, specifically includes:

[0007] ① Define the types and number of satellite entities included in the satellite data-real fusion test observation scenario, define the communication relationship and scenario conditions, and define the scenario celestial environment, communication model, and various physical field models;

[0008] ② Match the digital twin satellite entity model, including the digital twin satellite entity subsystem model, the digital twin satellite entity component model and the digital twin satellite entity onboard software model;

[0009] Step S102 defines the rules for dividing the granularity of satellite digital-real fusion test observation models. The on-board multi-granularity observation model library is divided into two parts: the on-board multi-granularity orbit determination model library and the on-board multi-granularity orbit change model library. Each granularity requires data of different magnitudes, including:

[0010] ① For the on-board multi-granularity orbit determination model, the number of sampling times and the calculation accuracy of the fitting algorithm are defined as the basis for dividing the model granularity of the orbit determination part of the observation model. The more sampling times and the more accurate the fitting algorithm, the longer the orbit determination time consumed on the satellite. Conversely, the fewer sampling times and the coarser the fitting algorithm, the less the orbit determination time consumed on the satellite.

[0011] ② For the on-board multi-granularity orbit change model, the fuel optimization effect is defined as the basis for dividing the granularity of the orbit change part of the observation model. The longer the optimization calculation time, the better the fuel optimization effect. Conversely, the shorter the optimization calculation time, the worse the fuel optimization effect.

[0012] ③ In each round of calculation of on-board observations in the digital-real fusion test, multiple sampling is first performed, and then fitting is performed to solve the orbital parameters of the target star. Finally, the orbital maneuver pulse approach observation is solved based on the orbital parameters of the target star. Each observation task includes multiple rounds of calculation processes. The calculation time directly affects whether the task can be successful. Based on this granularity division rule, the real-time granularity selection is performed in association with the urgency of the task;

[0013] Step S103, designing an on-board multi-granularity observation model for the satellite digital-real fusion test observation satellite, firstly establishing a high-fidelity model for digital twin satellite observation, and deriving multiple multi-granularity observation model libraries with different fidelity from the high-fidelity model based on the division basis of step S102, namely, a coarse-grained model library, a medium-grained model library and a fine-grained model library, as the on-board observation model selection library for subsequent satellite digital-real fusion tests;

[0014] Step S104: configure the initial state of the satellite data-real fusion test observation scene and inject the observation task as the starting condition for subsequent tests;

[0015] Step S105: The satellite selects the onboard observation model granularity according to the urgency of the real-time observation mission, evaluates the current situation to give the best decision-making speed and certainty trade-off, and selects the model level that is most likely to provide the best fit; in the long-distance flying stage when the mission is not urgent, select the observation model granularity that allows more calculation time but higher calculation accuracy, also known as the fine-grained model, to save fuel to the maximum extent; in the close-range game pursuit stage when the mission is more urgent, select the observation model granularity that allows less calculation time but poorer calculation accuracy, also known as the coarse-grained model, to adapt to the fast and intense game process;

[0016] Step S106: Based on the real satellite telemetry parameter data, the digital twin satellite entity model is corrected by digital-real fusion, including the digital twin satellite entity orbit determination and orbit change model, the high-fidelity model of the digital twin satellite observation, and the multi-granularity observation model, and the on-board observation model selection library of the satellite digital-real fusion test is updated.

[0017] The advantages of the present invention compared with the prior art are:

[0018] (1) Design a digital-real fusion test process for satellite multi-granularity observation models. This includes pre-test requirements preparation, digital twin observation model granularity division, onboard observation model granularity selection, and digital-real fusion onboard observation model correction. A solution for satellites to correct the model during observation missions is provided, which improves the accuracy and reliability of onboard observation models during emergency response missions to a certain extent.

[0019] (2) Define the rules for dividing the granularity of observation models in satellite data-real fusion tests, and design a method for selecting the granularity of onboard observation models. Based on the high-fidelity model, a variety of multi-granularity models with different fidelity are divided by distinguishing the data volume requirements, computing volume, model fitting accuracy, and model optimization effect, and a multi-granularity observation model library is constructed for real-time selection onboard. By judging the urgency of the onboard observation task, the computing volume requirements, and the allowable range of calculation errors, the model granularity of the observation model can be selected in real time onboard the satellite, which improves the response efficiency and decision reliability of the onboard observation task to a certain extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of the observation satellite digital-real fusion test of the present invention;

[0021] Figure 2 The present invention is a flow chart of a method for selecting a multi-granularity observation model on board for satellite digital-real fusion testing. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the protection scope of the present invention.

[0023] According to one embodiment of the present invention, a method for selecting a multi-granularity observation model on board for satellite digital-real fusion testing is provided. Figure 1 This is a flow chart of the observation satellite digital-real fusion test method of the present invention, Figure 2 The flowchart of the on-board multi-granularity observation model selection method for satellite digital-real fusion testing of the present invention is introduced in the following example using a typical three-star scene and a three-layer observation model granularity level. Figure 1 As shown, the specific implementation includes:

[0024] Step S101, building a digital twin observation scene for satellite digital-physical fusion testing and matching a digital twin satellite entity model, specifically includes:

[0025] ① Define the satellite entity types and entity quantity included in the satellite digital-real fusion test observation scene as 3, define the communication relationship and scene conditions, and define the scene celestial environment, communication model and various physical field models;

[0026] ②Configure the digital twin satellite entity model, including the digital twin satellite entity subsystem model, the digital twin satellite entity component model and the digital twin satellite entity onboard software model;

[0027] Step S102 defines the rules for dividing the granularity of satellite digital-real fusion test observation models. The on-board multi-granularity observation model library is divided into two parts: the on-board multi-granularity orbit determination model library and the on-board multi-granularity orbit change model library. Each granularity requires data of different magnitudes, including:

[0028] ① For the on-board multi-granularity orbit determination model, the number of sampling times and the calculation accuracy of the fitting algorithm are defined as the basis for dividing the model granularity of the orbit determination part of the observation model. The more sampling times and the more accurate the fitting algorithm, the longer the orbit determination time consumed on the satellite. Conversely, the fewer sampling times and the coarser the fitting algorithm, the less the orbit determination time consumed on the satellite.

[0029] ② For the on-board multi-granularity orbit change model, the fuel optimization effect is defined as the basis for dividing the granularity of the orbit change part of the observation model. The longer the optimization calculation time, the better the fuel optimization effect. Conversely, the shorter the optimization calculation time, the worse the fuel optimization effect.

[0030] ③ In each round of calculation of on-board observations in the digital-real fusion test, multiple sampling is first performed, and then fitting is performed to solve the orbital parameters of the target star. Finally, the orbital maneuver pulse approach observation is solved based on the orbital parameters of the target star. Each observation task includes multiple rounds of calculation processes. The calculation time directly affects whether the task can be successful. Based on this granularity division rule, the real-time granularity selection is performed in association with the urgency of the task;

[0031] Step S103, designing an onboard multi-granularity observation model for the satellite digital-real fusion test observation satellite, firstly establishing a high-fidelity model for digital twin satellite observation, and deriving three multi-granularity observation model libraries with different fidelity from the high-fidelity model based on the division basis of step S102, namely, a coarse-grained model library, a medium-grained model library and a fine-grained model library, as the onboard observation model selection library for subsequent satellite digital-real fusion tests;

[0032] Step S104: configure the initial state of the satellite data-real fusion test observation scene and inject the observation task as the starting condition for subsequent tests;

[0033] Step S105: The satellite selects the onboard observation model granularity according to the urgency of the real-time observation mission, evaluates the current situation to give the best decision-making speed and certainty trade-off, and selects the model level that is most likely to provide the best fit; in the long-distance flying stage when the mission is not urgent, select the observation model granularity that allows more calculation time but higher calculation accuracy, also known as the fine-grained model, to save fuel to the maximum extent; in the close-range game pursuit stage when the mission is more urgent, select the observation model granularity that allows less calculation time but poorer calculation accuracy, also known as the coarse-grained model, to adapt to the fast and intense game process;

[0034] Step S106: Based on the real satellite telemetry parameter data, the digital twin satellite entity model is corrected by digital-real fusion, including the digital twin satellite entity orbit determination and orbit change model, the high-fidelity model of the digital twin satellite observation, and the multi-granularity observation model, and the on-board observation model selection library of the satellite digital-real fusion test is updated.

[0035] According to another embodiment of the present invention, a method for selecting a multi-granularity observation model on board for satellite digital-real fusion testing is proposed. Figure 2 As shown, its specific implementation includes:

[0036] Step S1051: define The orbit determination model algorithm for each granularity layer includes Two outputs, For non-cooperative target information, Information about cooperation objectives. The track change model algorithm for each granularity layer includes Two outputs, is the pulse vector, is the fuel consumption, This is the state parameter at this moment.

[0037] ,

[0038] Step S1052: First, for each granularity selection problem in each round of satellite data-real fusion test observation task, enter the coarse-grained model layer first, first perform model consistency verification on the model at this layer, and correct the model if it fails the verification. After passing the verification, use the coarse-grained model to calculate and solve:

[0039] ① At the coarse-grained observation model layer, define the threshold , Threshold and the evaluation function The relative state domain of the observed satellite and the target orbit Divided into three areas , namely, the acceptance domain, delay domain, and rejection domain;

[0040] ②If divided into the receptive domain , then take action Accept the calculation results of this granularity layer; if divided into delay domain , then take action Get more observation information and enter the next granularity layer for detailed calculation; if it is divided into the rejection region , then take action Reject the calculation results of this granularity layer, re-obtain the observation information and keep it in the calculation of this granularity layer;

[0041] ③ For the orbit determination model, a coarse-grained model is used for calculation, and a coarse-grained orbit determination threshold is given. , when the relative distance between the observed star and the non-cooperative target does not meet the coarse-grained distance threshold condition, it is divided into the delay domain. At this time, the mission urgency is low and it is allowed to enter the next granularity level to obtain more information and delay the decision; given the coarse-grained threat threshold , when the relative distance between the measured non-cooperative target and the measured cooperative target meets the coarse-grained threat threshold, it is divided into the acceptance domain. At this time, the measured cooperative target is threatened by the measured non-cooperative target and accepts the coarse-grained calculation result; when the relative distance between the measured non-cooperative target and the measured cooperative target does not meet the coarse-grained threat threshold, it is divided into the rejection domain. At this time, the measured cooperative target is not threatened and stays at this granularity layer for re-observation and calculation; the specific formula is:

[0042] ,

[0043] ④ For the track change model, a coarse-grained model is used for calculation, and a coarse-grained track change threshold is given. , when the relative distance between the observed star and the non-cooperative target does not meet the coarse-grained distance threshold condition, it is divided into the delay domain delay decision; given the coarse-grained threat threshold , when the orbital maneuvering fuel is solved Less than the upper limit of remaining fuel of the tracking star When , it is divided into the acceptance domain and the coarse-grained calculation results are performed; Greater than When the fuel limit is exceeded, it is divided into the rejection domain and stays in the current granularity layer for re-observation and calculation. The specific formula is:

[0044] ,

[0045] ⑤ The coarse-grained layer is divided into 1 layer in total, and step S1052 enters the first layer of the coarse-grained layer;

[0046] Step S1053: For the problem of the previous layer being divided into the coarse-grained layer delay domain, enter a finer-grained model layer;

[0047] Step S1054: For the problem classified into the coarse-grained layer delay domain, enter the intermediate granularity layer:

[0048] ①Similar to the division logic of the coarse-grained layer, the threshold is replaced with the middle-grained layer , and the evaluation function Domain Divided into three areas , namely, the acceptance domain, delay domain, and rejection domain;

[0049] ②For orbit determination, there are:

[0050] ,

[0051] ③For orbital maneuvers, there are:

[0052] ,

[0053] ④ The medium granularity is divided into 1 layer in total, and step S1054 enters the first layer of the medium granularity;

[0054] Step S1055: For the problem of the delay domain of the previous layer divided into the middle granularity layer, enter the model layer with finer granularity;

[0055] Step S1056: For the problem classified into the delay domain of the intermediate granularity layer, enter the finest granularity layer:

[0056] ① The fine-grained layer is the lowest granularity layer in the on-board observation model selection library, and no longer divides the delay domain. Only relying on the threshold and the evaluation function Domain Divided into two areas .

[0057] ②For orbit determination, there are:

[0058] ,

[0059] ③For orbital maneuvers, there are:

[0060] ,

[0061] Step S1057, the initial value selection of the granularity for non-first-round calculation of the observation task. When the on-board observation task of the digital-real fusion test is single, the solution is refined layer by layer from the coarse-grained layer to the fine-grained layer. In multi-round long-term observation tasks, multiple granularity selection decisions are involved. In order to reduce the amount of calculation brought by layer-by-layer iteration, a decision-making loop that inherits prior knowledge is added, that is: in the same observation task condition, the granularity layer accepted by the previous model granularity selection is directly used as the first granularity layer entered by the next model granularity selection.

[0062] In summary, the present invention discloses an on-board multi-granularity observation model selection method for satellite digital-real fusion testing, including designing a digital-real fusion test process for the on-board multi-granularity observation model, and designing a granularity selection method for the on-board observation model. It can solve the problems existing in the current satellite observation model to a certain extent and achieve a better balance between decision-making speed and certainty in on-board observation tasks.

[0063] The contents not described in detail in the specification of the present invention belong to the prior art known to the professional and technical personnel in this field.

[0064] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for selecting a multi-granularity observation model on board for satellite digital-real fusion testing, characterized in that: The specific method flow is as follows: Step S101, building a digital twin observation scene for satellite digital-physical fusion testing and matching a digital twin satellite entity model, specifically includes: ① Define the types and number of satellite entities included in the satellite data-real fusion test observation scenario, define the communication relationship and scenario conditions, and define the scenario celestial environment, communication model, and various physical field models; ② Match the digital twin satellite entity model, including the digital twin satellite entity subsystem model, the digital twin satellite entity component model and the digital twin satellite entity onboard software model; Step S102 defines the rules for dividing the granularity of satellite digital-real fusion test observation models. The on-board multi-granularity observation model library is divided into two parts: the on-board multi-granularity orbit determination model library and the on-board multi-granularity orbit change model library. Each granularity requires data of different magnitudes, including: ① For the on-board multi-granularity orbit determination model, the number of sampling times and the calculation accuracy of the fitting algorithm are defined as the basis for dividing the model granularity of the orbit determination part of the observation model. The more sampling times and the more accurate the fitting algorithm, the longer the orbit determination time consumed on the satellite. Conversely, the fewer sampling times and the coarser the fitting algorithm, the less the orbit determination time consumed on the satellite. ② For the on-board multi-granularity orbit change model, the fuel optimization effect is defined as the basis for dividing the granularity of the orbit change part of the observation model. The longer the optimization calculation time, the better the fuel optimization effect. Conversely, the shorter the optimization calculation time, the worse the fuel optimization effect. ③ In each round of calculation of on-board observations in the digital-real fusion test, multiple sampling is first performed, and then fitting is performed to solve the orbital parameters of the target star. Finally, the orbital maneuver pulse approach observation is solved based on the orbital parameters of the target star. Each observation task includes multiple rounds of calculation processes. The calculation time directly affects whether the task can be successful. Based on this granularity division rule, the real-time granularity selection is performed in association with the urgency of the task; Step S103, designing an on-board multi-granularity observation model for the satellite digital-real fusion test observation satellite, firstly establishing a high-fidelity model for digital twin satellite observation, and deriving multiple multi-granularity observation model libraries with different fidelity from the high-fidelity model based on the division basis of step S102, namely, a coarse-grained model library, a medium-grained model library and a fine-grained model library, as the on-board observation model selection library for subsequent satellite digital-real fusion tests; Step S104: configure the initial state of the satellite data-real fusion test observation scene and inject the observation task as the starting condition for subsequent tests; Step S105: The satellite selects the onboard observation model granularity according to the urgency of the real-time observation mission, evaluates the current situation to give the best decision-making speed and certainty trade-off, and selects the model level that is most likely to provide the best fit; in the long-distance flying stage when the mission is not urgent, select the observation model granularity that allows more calculation time but higher calculation accuracy, also known as the fine-grained model, to save fuel to the maximum extent; in the close-range game pursuit stage when the mission is more urgent, select the observation model granularity that allows less calculation time but poorer calculation accuracy, also known as the coarse-grained model, to adapt to the fast and intense game process; Step S106: Based on the real satellite telemetry parameter data, the digital twin satellite entity model is corrected by digital-real fusion, including the digital twin satellite entity orbit determination and orbit change model, the high-fidelity model of the digital twin satellite observation, and the multi-granularity observation model, and the on-board observation model selection library of the satellite digital-real fusion test is updated.

2. The on-board multi-granularity observation model selection method for satellite digital-real fusion testing according to claim 1 is characterized in that: The step S105 observation model granularity selection method specifically includes: Step S1051: define The orbit determination model algorithm for each granularity layer includes Two outputs, For non-cooperative target information, For cooperation target information, The track change model algorithm for each granularity layer includes Two outputs, is the pulse vector, is the fuel consumption, The state parameters for this moment are: ， Step S1052: First, for each granularity selection problem in each round of satellite data-real fusion test observation task, enter the coarse-grained model layer first, first perform model consistency verification on the model at this layer, and correct the model if it fails the verification. After passing the verification, use the coarse-grained model to calculate and solve: ① At the coarse-grained observation model layer, define the threshold , Threshold and the evaluation function The relative state domain of the observed satellite and the target orbit Divided into three areas , namely, the acceptance domain, delay domain, and rejection domain; ②If divided into the receptive domain , then take action Accept the calculation results of this granularity layer; if divided into delay domain , then take action Get more observation information and enter the next granularity layer for detailed calculation; if it is divided into the rejection region , then take action Reject the calculation results of this granularity layer, re-obtain the observation information and keep it in the calculation of this granularity layer; ③ For the orbit determination model, a coarse-grained model is used for calculation, and a coarse-grained orbit determination threshold is given. , when the relative distance between the observed star and the non-cooperative target does not meet the coarse-grained distance threshold condition, it is divided into the delay domain. At this time, the mission urgency is low and it is allowed to enter the next granularity level to obtain more information and delay the decision; given the coarse-grained threat threshold , when the relative distance between the measured non-cooperative target and the measured cooperative target meets the coarse-grained threat threshold, it is divided into the acceptance domain. At this time, the measured cooperative target is threatened by the measured non-cooperative target and accepts the coarse-grained calculation result; when the relative distance between the measured non-cooperative target and the measured cooperative target does not meet the coarse-grained threat threshold, it is divided into the rejection domain. At this time, the measured cooperative target is not threatened and stays at this granularity layer for re-observation and calculation; the specific formula is: ， ④ For the track change model, a coarse-grained model is used for calculation, and a coarse-grained track change threshold is given. , when the relative distance between the observed star and the non-cooperative target does not meet the coarse-grained distance threshold condition, it is divided into the delay domain delay decision; given the coarse-grained threat threshold , when the orbital maneuvering fuel is solved Less than the upper limit of remaining fuel of the tracking star When , it is divided into the acceptance domain and the coarse-grained calculation results are performed; Greater than When the fuel limit is exceeded, it is divided into the rejection domain and stays in the current granularity layer for re-observation and calculation. The specific formula is: ， ⑤ The coarse-grained structure is divided into m layers in total, m ≥ 1, and step S1052 initially enters the i-th layer of the coarse-grained structure, i = 1; Step S1053: For the problem of the previous layer being divided into the coarse-grained layer delay domain, enter a finer-grained model layer, i+1; Step S1054: Repeat step S1053 until i>m and enter the intermediate granularity layer: ①Similar to the division logic of the coarse-grained layer, the threshold is replaced with the middle-grained layer , and the evaluation function Domain Divided into three areas , namely, the acceptance domain, delay domain, and rejection domain; ② For orbit determination, the intermediate granularity model is used for calculation, and there are: ， ③ For orbital maneuvers, the intermediate granularity model is used for calculation, which is: ， ④ The medium granularity is divided into n layers in total, n ≥ 1, and step S1054 initially enters the jth layer of the medium granularity. j=1； Step S1055: For the problem of the delay domain of the previous layer divided into the middle granularity layer, enter the model layer of finer granularity, j+1; Step S1056: Repeat step S1055 until j>n to enter the finest granularity layer: ① The fine-grained layer is the lowest level of granularity in the on-board observation model selection library. It no longer divides the delay domain and only relies on the threshold and the evaluation function Domain Divided into two areas ; ② For orbit determination, a fine-grained model is used for calculation, with: ， ③For orbital maneuvers, a fine-grained model is used for calculation, with: ， Step S1057, the initial value selection of the granularity for observation tasks other than the first round of calculation. When the on-board observation task of the digital-real fusion test is single, the solution is refined layer by layer from the coarse-grained layer to the fine-grained layer. In multi-round long-term observation tasks, multiple granularity selection decisions are involved. In order to reduce the amount of calculation brought by layer-by-layer iteration, a decision-making loop that inherits prior knowledge is added, that is, in the same observation task condition, the granularity layer accepted by the previous model granularity selection is directly used as the first granularity layer entered by the next model granularity selection.

Images