On-orbit spacecraft generalized intelligent control system architecture and autonomous control method

Abstract

The application discloses a kind of generalized intelligent control system architecture and autonomous control method of on-orbit spacecraft.It includes: knowledge base, for storing the theoretical knowledge of multiple known tasks and the theoretical feature vector generated based on each theoretical knowledge, each feature vector corresponds to different task situation grade;Theoretical knowledge includes the environment, experience and rules corresponding to each task;Perception unit is used to perceive task environment information and controlled object information, generates task situation assessment results based on the information perceived and each theoretical feature vector, and sends the assessment results to the decision model;The decision model is trained to obtain the execution strategy of the corresponding task as the output with multiple known tasks and the theoretical knowledge in the knowledge base as the input;Decision model is used to generate execution strategy based on assessment results and theoretical knowledge in knowledge base, and sends execution strategy to control unit;Control unit is used to complete corresponding task based on execution strategy.The spacecraft of the application can autonomously complete on-orbit task.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aerospace technology, in particular to a generalized intelligent control system architecture and autonomous control method for on-orbit spacecraft. BACKGROUND

[0002] At present, the perception and decision-making process of the launched on-orbit spacecraft need to be supported by ground personnel using high-fidelity models, that is, ground operation personnel need to analyze telemetry data, give task and planning instructions, and then upload and inject them into the on-orbit spacecraft to enable the on-orbit spacecraft to complete the corresponding tasks.

[0003] This operation mode of ground personnel participating in the loop control can complete some space assembly, orbit maneuvering and attitude orientation tasks, but when facing increasingly complex unexpected tasks or highly maneuverable confrontation tasks, due to the problems of time delay between earth and space, data distortion and space uncertainty, this mode has become a bottleneck restricting the implementation of space missions, and it is difficult to meet the growing needs of space exploration and game.

[0004] Therefore, there is an urgent need for a generalized intelligent control system architecture and autonomous control method for on-orbit spacecraft to solve the above problems. SUMMARY

[0005] The present application provides a generalized intelligent control system architecture and autonomous control method for on-orbit spacecraft, which can autonomously complete on-orbit tasks without interacting with ground equipment. The technical solution is as follows:

[0006] In a first aspect, the present application provides a generalized intelligent control system architecture for on-orbit spacecraft, which comprises:

[0007] a knowledge base for storing theoretical knowledge of multiple known tasks and theoretical feature vectors generated based on the theoretical knowledge, each feature vector corresponding to a different task situation level, for reflecting task information from multiple aspects; the theoretical knowledge includes the environment, experience and rules corresponding to each task;

[0008] a perception unit for perceiving task environment information and controlled object information, generating a task situation assessment result based on the perceived information and each theoretical feature vector, and sending the assessment result to a pre-trained decision model; the decision model is trained with multiple known tasks and the theoretical knowledge in the knowledge base as input and the execution strategy of the corresponding task as output;

[0009] the decision model for generating an execution strategy based on the assessment result and the theoretical knowledge in the knowledge base, and sending the execution strategy to a control unit;

[0010] The control unit is used for completing a corresponding task based on the execution strategy.

[0011] In a second aspect, the embodiment of the present application further provides an autonomous control method of an on-orbit spacecraft, which is applied to any possible control system architecture, and the method comprises the following steps:

[0012] The knowledge base is used to store theoretical knowledge of a plurality of known tasks and theoretical feature vectors generated based on the theoretical knowledge;

[0013] The perception unit is used to perceive task environment information and controlled object information, a task situation assessment result is generated based on the perceived information and the theoretical feature vectors, and the assessment result is sent to a pre-trained decision model;

[0014] The decision model is used to generate an execution strategy based on the assessment result and the theoretical knowledge in the knowledge base, and the execution strategy is sent to the control unit;

[0015] The control unit is used for completing a corresponding task based on the execution strategy.

[0016] The embodiment of the present application provides a generalized intelligent control system architecture of an on-orbit spacecraft. The architecture adopts an integrated control mode of "perception-decision-control". First, a knowledge base is generated based on known task cases, so that a corresponding feature vector and other information can be matched from the knowledge base based on information acquired by the perception unit. Then, a decision model is constructed based on neural network technology, and the model is trained based on known task cases and related theoretical knowledge, so that the model has the ability to generate a corresponding execution strategy according to actual task information. Finally, the control unit is used to execute the corresponding strategy to complete the corresponding task. As can be seen, when performing on-orbit tasks, the present application does not need to interact with ground equipment, nor does it need to send control instructions from the ground equipment, but can complete various tasks by using its own perception function and decision function, and has the ability to handle unexpected situations. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiment or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0018] Figure 1 is a schematic diagram of a generalized intelligent control system architecture of an on-orbit spacecraft provided by an embodiment of the present application;

[0019] Figure 2 is a schematic diagram of an outer loop of the control system architecture provided by an embodiment of the present application;

[0020] Figure 3 Fig. 1 is a schematic diagram of an inner loop of a control system architecture according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0022] The specific implementation of the above concept will be described below.

[0023] Referring to Figure 1 The embodiment of the present application provides a generalized intelligent control system architecture of an on-orbit spacecraft, and the architecture comprises:

[0024] a knowledge base, used for storing theoretical knowledge of a plurality of known tasks and theoretical feature vectors generated based on the theoretical knowledge, each feature vector corresponding to a different task situation level and used for reflecting task information from a plurality of aspects; the theoretical knowledge comprises environment, experience and rules corresponding to each task;

[0025] a perception unit, used for perceiving task environment information and controlled object information, generating a task situation assessment result based on the perceived information and the theoretical feature vectors, and sending the assessment result to a pre-trained decision model; the decision model is trained with a plurality of known tasks and the theoretical knowledge in the knowledge base as input and an execution strategy of the corresponding task as output;

[0026] a decision model, used for generating an execution strategy based on the assessment result and the theoretical knowledge in the knowledge base, and sending the execution strategy to a control unit;

[0027] the control unit, used for completing a corresponding task based on the execution strategy.

[0028] In this embodiment, first, a knowledge base is generated based on known task cases, so that corresponding feature vectors and the like information can be matched from the knowledge base based on the information obtained by the perception unit. Then, a decision model is constructed based on neural network technology, and the model is trained based on known task cases and related theoretical knowledge, so as to have the ability to generate a corresponding execution strategy according to actual task information. Finally, the control unit is used to execute the corresponding strategy to complete the corresponding task. As can be seen, when performing on-orbit tasks, the application does not need to interact with ground equipment, nor does it need to send control instructions from the ground equipment. Instead, it can complete various tasks using its own perception and decision-making functions, has the ability to handle unexpected situations, is stable, and has high control accuracy.

[0029] It should be noted that the task in the present application can be rendezvous and docking, space manipulation, or game confrontation.

[0030] The modules in the control system architecture will be described in detail as follows:

[0031] First, for the knowledge base:

[0032] The theoretical knowledge in the knowledge base mainly comes from previous on-orbit tasks such as docking, manipulation, and game confrontation under information conditions. When storing, the knowledge base is stored in a hierarchical manner according to the order of environment, experience, and rules, thereby forming a theoretical feature vector corresponding to each known task. The specific contents stored by environment, experience, and rules are as follows:

[0033] 1. The environment stores both static data and dynamic data formed during the execution of the task, mainly stores space electromagnetic, illumination environment, etc., and also includes manually input information;

[0034] 2. Experience mainly stores different decisions and control parameters that should be taken under different task states, etc. It is determined by relevant task experts through task experience accumulation, and the experience is changed accordingly through intelligent learning algorithm, which is a necessary condition for decision-making;

[0035] 3. Rules mainly store non-executable task areas, the correspondence between the types of tools used by the opponent and the types of tools that should be used by us, the weight setting of the task, etc. It can be determined by relevant experts through known task experience accumulation, or it can be changed accordingly through intelligent learning algorithm of the command and control model

[0036] It should be noted that the knowledge base is the source of knowledge for reasoning. Since it has memory and learning functions, the knowledge base is constantly updated with the increase of the number of tasks and experience.

[0037] It should be noted that the theoretical feature vector corresponding to each task is generated based on the environmental information, experience and rules corresponding to the task, and the theoretical feature vector can reflect the task information from multiple levels. Taking game confrontation as an example, the feature vector can reflect the task information from the following three levels:

[0038] The first level is the global situation, that is, determining whether the opponent is in an attack state, a defense state, a stalemate state or a retreat state, which is used to represent the final goal of the opponent.

[0039] The second level is the sub-situation, and for any state of the global situation, the corresponding sub-situation constitutes the global state, for example, the attack situation can be composed of main attack sub-situation, feint sub-situation, support sub-situation and cover sub-situation.

[0040] The third level is the game action situation, such as the judgment of the working state of the concentration and the maneuver.

[0041] Of course, the above is only a part of the example, and the user can define the corresponding feature vector according to the actual task needs. In addition, the task situation level can include very urgent, urgent, general urgent and not urgent, etc. The situation level is used to represent the degree of urgency of the task, and is used to guide the generation of the execution strategy.

[0042] Second, for the perception unit:

[0043] The perception unit includes a perception module, a generation module, a trigger module and an evaluation module connected in turn.

[0044] 1. The perception module is used to perceive the task environment information and the controlled object information, and send the perceived information to the generation module or the control unit.

[0045] 2. The generation module is used to generate the real feature vector of the task based on the received information.

[0046] The real feature vector is used to reflect the real situation of the task, and its content is the same as the above-mentioned theoretical feature vector, which will not be described here.

[0047] 3. The trigger module is used to determine whether to change the current execution strategy based on the preset trigger rule.

[0048] In this module, the preset trigger rule at least includes event triggering, and the specific rule is:

[0049] For each control period, it is judged whether the difference between the real feature vector generated in the current control period and the real feature vector in the previous control period is greater than a set threshold value; if yes, it is determined to change the current execution strategy, and if no, it is determined not to change the current execution strategy.

[0050] When the difference between the feature vectors generated by two adjacent control periods is large, it indicates that the task has a sudden situation or a large decision error. If the previous execution strategy is still used, it cannot adapt to the new task situation, so it is necessary to re-analyze the situation and make appropriate decision adjustment to meet the task needs. Otherwise, it is considered that the task situation has not changed, and the current execution strategy can be used, thereby ensuring the real-time processing of the task.

[0051] It should be further pointed out that the preset triggering rule also includes time triggering, that is, determining a decision period, and re-determining the execution strategy in each decision period.

[0052] Under this triggering rule, the execution strategy is re-made every other decision period, regardless of whether the real feature vector has changed significantly, so as to ensure that the selected execution strategy in each decision period is the optimal solution. It should be pointed out that the decision period is greater than the control period, and the decision period is an integer multiple of the control period. In addition, the specific values of the threshold, the control period and the decision period are determined according to the task needs, and the present application does not make specific limitations.

[0053] 4. The evaluation module is used to generate a task situation evaluation result based on the matching result of the real feature vector and the theoretical feature vector in the knowledge base, and send the evaluation result to the decision model.

[0054] The evaluation module uses the example reasoning technology to match the theoretical feature vector corresponding to the real feature vector from the knowledge base, and determines the task situation corresponding to the matched theoretical feature vector as the task situation evaluation result.

[0055] Third, for the decision model:

[0056] The decision model is constructed based on a neural network. The decision model is connected with the knowledge base and the evaluation module. After the decision model is constructed, it is trained offline with multiple known tasks and information stored in the knowledge base. Each known task corresponds to a feature vector, a task situation level corresponding to the feature vector, and a rule in the knowledge base, and is labeled with a corresponding execution strategy. After training, the decision model can establish a mapping relationship between new task cases and execution strategies. In addition, the known task case library can be supplemented or adjusted according to the actual situation to ensure that the obtained neural network can effectively generate decision variables.

[0057] Fourth, for the control unit:

[0058] The control unit includes multiple sub-models and controlled objects, and each sub-model corresponds to different types of tasks and corresponding controllers.

[0059] When the execution strategy needs to be changed, an outer loop formed based on the perception unit, the knowledge base, the decision model and the control unit performs the corresponding task, and the outer loop is as shown in the middle red line part. Figure 2

[0060] When the execution strategy does not need to be changed, an inner loop formed based on the perception unit and the control unit performs the corresponding task, and the inner loop is as shown in the middle red line part. Figure 3

[0061] In some embodiments, the outer loop formed based on the perception unit, the knowledge base, the decision model and the control unit performs the corresponding task, including:

[0062] The evaluation module sends the new task situation evaluation result to the decision model;

[0063] The decision model generates a new execution strategy based on the new task situation evaluation result and the corresponding rules in the knowledge base, and sends the new execution strategy to the control unit; the new execution strategy includes a new controller and a new gain scheduling result;

[0064] The control unit switches the controller to the new controller, and adjusts the controller parameters based on the new gain scheduling result, so as to output a control signal to the controlled object based on the new controller and perform the corresponding task.

[0065] This embodiment can realize stable switching between the controller and the control parameters. The space environment is complex and changeable, and multiple sets of control parameters may be needed at the controller design level to cope with different task execution modes, and multiple controllers may be needed at the decision level to effectively control all tasks. Therefore, in the process of a task action, in order to achieve the task purpose, effective, stable and rapid switching needs to be performed between multiple sets of control parameters or controllers.

[0066] In some embodiments, the inner loop formed based on the perception unit and the control unit performs the corresponding task, including:

[0067] The perception module sends the perceived task environment information and the controlled object information to the current controller; the current controller performs gain scheduling based on the current controlled object information, and adjusts the controller parameters based on the scheduling result, so as to output a control signal to the controlled object based on the controller and perform the corresponding task.

[0068] In this embodiment, gain scheduling is performed according to the state of the controlled object (such as the flight orbit of a spacecraft), appropriate controller parameters for the current task state are selected, and the execution process with high timeliness and high precision is further realized through effective switching of the controller parameters.

[0069] In some embodiments, the inner loop adopts a linear / nonlinear active disturbance rejection controller switching mode;​​

[0070] In response to the control unit receiving the corresponding execution strategy, every preset period, it is judged whether the error between the current state and the target state of the controlled object is greater than the preset range, if yes, the linear active disturbance rejection controller is used to control the controlled object, if no, the linear active disturbance rejection controller is switched to the nonlinear active disturbance rejection controller, and the nonlinear active disturbance rejection controller is used to control the controlled object.

[0071] In this embodiment, when the error is large, the linear active disturbance rejection controller is used to control the controlled object, which can realize fast action. When the error is small, the nonlinear active disturbance rejection controller is used to control the controlled object, which can realize high-precision control.

[0072] In addition, the control unit can also include a feedback module for filtering out noise in the data sent by the perception module.

[0073] The application also provides an autonomous control method of an on-orbit spacecraft, which is applied to the control system architecture in any of the above embodiments, and the method comprises the following steps:

[0074] The knowledge base is used to store theoretical knowledge of a plurality of known tasks and theoretical feature vectors generated based on the theoretical knowledge;

[0075] The perception unit is used to perceive task environment information and controlled object information, and a task situation assessment result is generated based on the perceived information and the theoretical feature vectors, and the assessment result is sent to the pre-trained decision model;

[0076] The decision model is used to generate an execution strategy based on the assessment result and the theoretical knowledge in the knowledge base, and the execution strategy is sent to the control unit;

[0077] The control unit is used to complete the corresponding task based on the execution strategy.

[0078] It should be noted that the control method provided in this embodiment and the control system architecture provided in the above embodiments are based on the same inventive concept, so they have the same beneficial effects, which will not be described here.

[0079] Finally, it needs to be pointed out that, in this document, relational terms such as first, second, third, and fourth and the like can only be used to distinguish one entity or action from another entity or action, without necessarily requiring or implying any actual relationship or order between or among such entities or actions. Moreover, the terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0080] The above description is merely preferred embodiments of the present application, and it is obvious to those skilled in the art that, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should also be considered as falling within the scope of the present application.

Claims

1. A generalized intelligent control system architecture for on-orbit spacecraft, characterized in that, The architecture includes: The knowledge base stores theoretical knowledge of various known tasks and theoretical feature vectors generated based on this knowledge. Each feature vector corresponds to a different task status level, reflecting task information from multiple levels. The theoretical knowledge includes the environment, experience, and rules corresponding to each task. The task is a game-theoretic adversarial task. The environment includes spatial electromagnetic and lighting conditions. The experience includes different decisions and control parameters to be taken under different task states. The rules include unexecutable task areas, the correspondence between the types of tools used by the opponent and the types of tools we can use to respond, and the weight of the task. The perception unit is used to perceive task environment information and controlled object information, generate task situation assessment results based on the perceived information and various theoretical feature vectors, and send the assessment results to a pre-trained decision model. The decision model is trained with multiple known tasks and theoretical knowledge in the knowledge base as inputs and the execution strategy of the corresponding task as output. The decision model is used to generate an execution strategy based on the evaluation results and the theoretical knowledge in the knowledge base, and to send the execution strategy to the control unit. The control unit is used to complete the corresponding task based on the execution strategy; the sensing unit includes a sensing module, a generation module, a triggering module and an evaluation module connected in sequence. The sensing module is used to sense task environment information and controlled object information, and send the sensed information to the generation module or the control unit. The generation module is used to generate the true feature vector of the task based on the received information; The triggering module is used to determine whether to change the current execution strategy based on preset triggering rules; The evaluation module is used to generate a task situation evaluation result based on the matching result between the real feature vector and the theoretical feature vector in the knowledge base, and send the evaluation result to the decision model; The control unit includes multiple sub-models and the controlled object, with each sub-model corresponding to different types of tasks and corresponding controllers; When it is necessary to change the execution strategy, the corresponding task is executed based on the outer loop formed by the perception unit, the knowledge base, the decision model and the control unit: the new task situation assessment result is sent to the decision model using the evaluation module; The decision model selects a new sub-model and a corresponding controller based on the new task situation assessment results and the corresponding rules in the knowledge base, generates a new execution strategy, and sends the new execution strategy to the control unit; the new execution strategy includes a new controller and a new gain scheduling result. The control unit switches the controller to the new controller and adjusts the controller parameters based on the new gain scheduling result, so as to output control signals to the controlled object based on the new controller and execute the corresponding task. When there is no need to change the execution strategy, the corresponding task is executed based on the inner loop formed by the sensing unit and the control unit: the sensing module sends the sensed task environment information and controlled object information to the current controller; the current controller performs gain scheduling based on the current controlled object information, and adjusts the controller parameters based on the scheduling result, so as to output control signals to the controlled object based on the controller and execute the corresponding task.

2. The architecture according to claim 1, characterized in that, The preset triggering rule is as follows: For each control cycle, the following is performed: determine whether the difference between the true feature vector generated in the current control cycle and the true feature vector in the previous control cycle is greater than a set threshold; if yes, determine to change the current execution strategy; if no, determine not to change the current execution strategy.

3. The architecture according to claim 2, characterized in that, The preset triggering rules also include: determining a decision cycle, and redetermining the execution strategy for each decision cycle, wherein the decision cycle is longer than the control cycle and the decision cycle is a positive integer multiple of the control cycle.

4. The architecture according to claim 1, characterized in that, The inner loop adopts a linear / nonlinear active disturbance rejection controller switching mode; In response to the control unit receiving the corresponding execution strategy, every preset period, it is determined whether the error between the current state and the target state of the controlled object is greater than a preset range. If so, the linear active disturbance rejection controller is used to control the controlled object; otherwise, the linear active disturbance rejection controller is switched to a nonlinear active disturbance rejection controller, and the nonlinear active disturbance rejection controller is used to control the controlled object.

5. The architecture according to claim 1, characterized in that, The assessment module determines the mission situation assessment results based on the following method: Based on the example reasoning method, theoretical feature vectors corresponding to real feature vectors are matched from the knowledge base, and the task situation level corresponding to the matched theoretical feature vectors is determined as the task situation assessment result.

6. An autonomous control method for an on-orbit spacecraft, characterized in that, Applied to the control system architecture as described in any one of claims 1-5, the method comprises: The knowledge base is used to store theoretical knowledge of various known tasks and theoretical feature vectors generated based on each theoretical knowledge. The sensing unit is used to sense task environment information and controlled object information. Based on the sensed information and various theoretical feature vectors, a task situation assessment result is generated and sent to a pre-trained decision model. The decision model generates an execution strategy based on the evaluation results and theoretical knowledge in the knowledge base, and sends the execution strategy to the control unit. The control unit completes the corresponding task based on the execution strategy.

Images