Spacecraft potential prediction and adaptive control method and system based on artificial intelligence

Abstract

The invention provides a spacecraft potential prediction and adaptive control method and system based on artificial intelligence, and the method employs an AI prediction module to predict a spacecraft potential change prediction value in a future period of time based on the real-time potential of a spacecraft and the real-time data of the plasma environment of the space around the spacecraft; the reinforcement learning decision engine constructs a state space by using the real-time potential of the spacecraft and a spacecraft potential change predicted value, takes a control instruction sent to an actuator as an action space, takes a target of maintaining the potential to be stable in a safety interval as an award, and learns an optimal control strategy through continuous interaction with the environment; generating an action instruction a based on the optimal control strategy; the action instruction a is converted into a control instruction to drive a charged particle emission assembly to work, and charged particles are emitted to the space so as to neutralize charges on the surface of the spacecraft. The intelligent potential of the spacecraft potential can be automatically adjusted and controlled in advance according to the dynamic evolution of the space environment.

Description

Technical Field

[0001] This invention relates to the field of spacecraft space environment effect protection technology, and in particular to a spacecraft potential prediction and adaptive control method and system based on artificial intelligence. Background Technology

[0002] During the operation of a spacecraft in orbit, its surface materials continuously interact with the surrounding space plasma environment, leading to the continuous accumulation of electrostatic charges and the formation of overall or local potentials within the spacecraft. When the potential exceeds a safe threshold, it will induce various on-orbit faults and performance degradation, mainly including: 1) electrostatic discharge, which may break down the insulation of components, interfere with onboard electronic systems, and cause equipment reset or permanent damage; 2) interference with space science exploration missions, affecting the measurement accuracy of sensitive payloads such as plasma probes and electric field meters, resulting in severe distortion of detection data; 3) accelerated degradation of surface materials and thermal control coatings, affecting the long-term reliable operation of the spacecraft.

[0003] Currently, the potential control technologies used in engineering mainly include electron emission devices (such as hollow cathodes), plasma contactors, and ion emission devices. Their control strategies largely rely on fixed-threshold switching control or classical PID control. These traditional methods have significant limitations in the dynamic and ever-changing space plasma environment: First, the control response is lag-dependent, only initiating compensation after a potential anomaly occurs, lacking pre-emptive intervention capabilities; second, environmental adaptability is insufficient, making it difficult to effectively cope with drastic changes in plasma parameters caused by spacecraft entering or leaving Earth's shadow, encountering geomagnetic storms, and solar flares; third, control accuracy and stability are limited, with fixed parameters difficult to achieve optimal matching under all mission conditions, easily leading to "over-control" or "under-control" phenomena. This not only causes unnecessary consumption of propellant or launch propellant but also fails to meet the high-precision, high-stability control requirements of emerging operational missions such as constellation coordination for spacecraft potential.

[0004] Therefore, existing technologies lack intelligent potential control methods that can anticipate the dynamic evolution of the space environment and adjust control strategies accordingly. There is an urgent need to develop an active potential control system with environmental perception, intelligent decision-making, and online learning capabilities to improve the adaptability and mission reliability of spacecraft in complex space environments. Summary of the Invention

[0005] In view of this, the present invention provides a spacecraft potential prediction and adaptive control method and system based on artificial intelligence, which can autonomously and pre-adjust the intelligent potential of the spacecraft according to the dynamic evolution of the space environment.

[0006] To solve the above-mentioned technical problems, the present invention is implemented as follows.

[0007] An artificial intelligence-based spacecraft potential prediction and adaptive control method includes: Step 1: Collect time-series data, including the spacecraft's real-time potential and the space plasma environment data around the spacecraft; Step 2: Based on the time series data, the AI ​​prediction model generates predicted values ​​for spacecraft potential changes over a future period. Step 3: The reinforcement learning decision engine uses the spacecraft's real-time potential and the predicted value of the spacecraft's potential change to form a state space, issuing control commands to the actuators as the action space, and maintaining the potential stability within the safe range as the reward. It learns the optimal control strategy through continuous interaction with the environment; and generates action command a based on the optimal control strategy. Step 4: Convert the action command 'a' generated by the reinforcement learning decision engine into a control command to drive the charged particle emission component to emit charged particles into space to neutralize the charge on the spacecraft surface.

[0008] Preferably, in step 2, the AI ​​prediction model further generates prediction data of the space plasma environment around the spacecraft for a future period based on the time series data; In step 3, the state space of the reinforcement learning decision engine is further enriched with real-time acquired and predicted data of the space plasma environment around the spacecraft.

[0009] Preferably, the charged particle emission assembly is one or more combinations of an electron emitter, an ion emission device, or a plasma contactor.

[0010] Preferably, step 1 further includes filtering and normalizing the time-series data.

[0011] Preferably, the AI ​​prediction model is a trained machine learning model; the machine learning model is a long short-term memory network, a Transformer temporal model, or a hybrid model of convolutional neural networks and recurrent neural networks.

[0012] Preferably, the action command 'a' output by the reinforcement learning decision engine includes the magnitude of the emission current, the emission timing, and the magnitude of the emission direction adjustment.

[0013] Preferably, the method further includes: after the charged particle emission component performs an action, the collected time-series data changes, generating new states and rewards, which are fed back to the reinforcement learning decision engine for online fine-tuning of the optimal control strategy, enabling it to continuously adapt to new environments that the spacecraft may encounter during its on-orbit lifespan.

[0014] Preferably, step 1 uses a spacecraft potential probe and a plasma probe to collect the timing data; The spacecraft potential probes are installed on the solar-facing side, the solar-avoiding side, and the sidewalls of the spacecraft. Plasma probes collect the density and temperature of the space plasma environment surrounding the spacecraft.

[0015] Preferably, the AI ​​prediction model and reinforcement learning decision engine are deployed in the AI ​​chip of the onboard computer; when running in orbit, it runs in master mode, while a shadow mode is maintained to run in parallel and continuously collect data, and the model is fine-tuned and updated periodically.

[0016] The present invention also provides an artificial intelligence-based spacecraft potential prediction and adaptive control system, comprising a perception layer, a decision layer and an execution layer; The perception layer includes a potential and environmental state measurement module and a preprocessing module, which are used to collect time-series data, including the spacecraft's real-time potential and the space plasma environment data around the spacecraft; the preprocessing module preprocesses and extracts features from the time-series data and inputs it into the decision layer; The decision-making layer includes AI predictive models and reinforcement learning decision engines; AI prediction models are used to generate predicted values ​​of spacecraft potential changes over a future period of time based on features from the perception layer. The reinforcement learning decision engine uses the features corresponding to the real-time potential of the spacecraft and the predicted value of the potential change of the spacecraft to form a state space, and sends control commands to the execution layer as the action space. The goal of maintaining the potential stability within the safe range is to serve as the reward. The engine learns the optimal control strategy by continuously interacting with the environment. The reinforcement learning decision engine generates action command 'a' based on the optimal control strategy and sends it to the execution layer. The execution layer includes a control command generation module and a charged particle emission component; the control command generation module is used to convert the action command a into a control command, driving the charged particle emission component to emit charged particles into space to neutralize the charge on the surface of the spacecraft.

[0017] As can be seen, this invention makes decisions on charged particle emission schemes based on real-time monitoring and prediction of potential. Even if the current potential is normal, if the AI ​​predicts that entering a high-density plasma region will cause a sharp negative bias in the potential, the decision engine will increase the controller's emission current in advance and gradually, achieving "preemptive" smooth control and improving the reliability of spacecraft potential control.

[0018] Compared with the prior art, the present invention has the following significant advantages: 1) Proactive and Predictive Control: Through AI prediction models, the system can anticipate potential changes, transforming passive response into proactive intervention, greatly eliminating control lag and avoiding drastic potential fluctuations. In one preferred embodiment, the AI ​​prediction model also predicts changes in space plasma environment data, providing another dimension of information to support preventative control and enhancing resilience to future space environment disturbances.

[0019] 2) High adaptability: The reinforcement learning-based controller does not rely on a precise physical model and can automatically adjust the control strategy through online learning, so as to cope with various complex and unknown spatial environment disturbances.

[0020] 3) High control precision and stability: It achieves smooth and precise control, avoiding the oscillation phenomenon that may occur in traditional PID control, and provides an "ultra-clean" background electrical environment for high-precision scientific exploration missions. At the same time, it can achieve high-precision potential control, providing a guarantee for future high-precision operations such as spacecraft docking and formation.

[0021] 4) Optimize resource consumption: Through intelligent decision-making, the system launches only when necessary and at the most appropriate power, effectively saving the working fluid or energy consumed by the electronic launch device and extending its on-orbit service life.

[0022] 5) Robust fault tolerance: The system can learn coping strategies in the event of partial sensor failure or actuator performance degradation, thereby improving the reliability of the entire system. Attached Figure Description

[0023] Figure 1 This is a flowchart of an artificial intelligence-based spacecraft potential prediction and adaptive control method according to Embodiment 1 of the present invention.

[0024] Figure 2 This is a block diagram of the spacecraft potential prediction and adaptive control system based on artificial intelligence, according to Embodiment 2 of the present invention.

[0025] Figure 3 This is a schematic diagram of the hardware composition of Embodiment 2 of the present invention. Detailed Implementation

[0026] This invention provides an artificial intelligence-based spacecraft potential prediction and adaptive control scheme. The basic idea is to predict the potential change of the spacecraft over a period of time based on the real-time potential of the spacecraft and plasma environment data. A deep reinforcement learning agent combines the real-time potential of the spacecraft with the AI-predicted future potential changes, and uses the goal of maintaining the potential stability within a safe range as a "reward" to intelligently decide the action of the charged particle emission component, thereby driving the charged particle emission component to work and emit charged particles into space to neutralize the charge on the surface of the spacecraft.

[0027] This invention uses real-time and predicted potentials to determine the charged particle emission scheme. Even if the current potential is normal, if the AI ​​predicts that entering a high-density plasma region will cause a sharp negative potential bias, the decision engine will increase the controller's emission current in advance and gradually, achieving "preemptive" smooth control. Therefore, this invention can autonomously and preemptively adjust the spacecraft's potential according to the dynamic evolution of the space environment. Through online learning, it can continuously adapt to new environments that the spacecraft may encounter during its on-orbit lifespan, further improving control accuracy, and the AI ​​decision-making response speed is fast.

[0028] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0029] Example 1 Figure 1 The flowchart of the spacecraft potential prediction and adaptive control method based on artificial intelligence in this embodiment is shown in the figure. The method includes the following steps: Step 1: Potential and Environmental Condition Measurement: Collect time-series data, including the spacecraft's real-time potential and the space plasma environment data around the spacecraft.

[0030] This step is the operation of the perception layer, which realizes environmental and state perception and data preprocessing. The perception layer uses spacecraft potential probes, plasma probes, and other methods to achieve spacecraft potential perception and perception of the space plasma environment around the spacecraft.

[0031] Among these features, spacecraft potential probes can be installed on the sun-facing side, the sun-facing side, and the sidewalls of the spacecraft. Plasma probes are deployed on an unobstructed area outside the satellite to measure the plasma density (Ne) and electron temperature (Te) of the environment.

[0032] Then, the data preprocessing section preprocesses the time-series data acquired in real time by the sensor, which may include filtering, normalization, and feature extraction, and then inputs it to the decision layer II.

[0033] Step 2: Decision layer II, based on potential and environmental state measurements, determines the action command A for charged ion radiation through AI prediction and intelligent decision-making.

[0034] In this step, Decision Layer II employs an AI-powered intelligent decision-making center to make action decisions. This center comprises a predictive model and a reinforcement learning decision engine. The AI-powered intelligent decision-making center runs as software on a radiation-hardened onboard AI computing unit within the spacecraft. This computing unit includes a multi-core CPU and a GPU or NPU to support computation, meeting the computational performance and power consumption requirements of the AI ​​model inference. All perceived data is transmitted to this computing unit via a data bus.

[0035] The AI ​​prediction model generates predicted spacecraft potential changes over a future period, such as tens of milliseconds to several seconds, based on time-series data of the spacecraft's real-time potential and the surrounding space plasma environment acquired by the perception layer. This AI prediction model is a trained machine learning model, which can be a Long Short-Term Memory (LSTM) network, a Transformer time-series model, or a hybrid model of Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN), to predict the trend of spacecraft potential changes over a future period.

[0036] The reinforcement learning decision engine is a deep reinforcement learning (DRL) agent, such as an agent based on DDPG or PPO algorithms. This agent uses the features corresponding to the spacecraft's real-time potential and predicted potential changes as its state space, issues control commands to actuators as its action space, and aims to maintain potential stability within a safe range as its reward. Through continuous interaction with the environment, the agent learns which optimal control strategy to adopt in which state. When the AI ​​predicts that the future potential value will exceed the safe range, it instructs the charged ion emission component to act in advance for preventative control.

[0037] Action instruction 'a' is generated based on the optimal control strategy and provided to execution layer III. The action instruction 'a' output by the reinforcement learning decision engine can include the magnitude of the emission current, the emission timing, and the magnitude of the emission direction adjustment, such as "increase the emission current to X mA".

[0038] In a preferred embodiment, the AI ​​prediction model is configured to further generate predicted data on the space plasma environment around the spacecraft over a future period. Since the space plasma environment data is part of the state space of the reinforcement learning decision engine, the state space of the reinforcement learning decision engine not only includes real-time acquired and predicted data on the spacecraft's potential, but also further includes real-time acquired and predicted data on the space plasma environment around the spacecraft, providing another dimension of information to support preventative control.

[0039] During the offline training phase, historical on-orbit satellite data and high-precision space environment simulation data are used to extensively train the AI ​​prediction model and reinforcement learning decision engine, enabling them to initially grasp the control laws. In the on-orbit deployment phase, the trained model is embedded in the AI ​​chip of the onboard computer. During on-orbit operation, the system runs in "active mode" while maintaining a "shadow mode" to run in parallel and continuously collect data, periodically fine-tuning and updating the model.

[0040] Step 3, Precise Execution: Execution layer III, based on the action command a generated by decision layer II, emits charged particles into space to neutralize the charge on the spacecraft's surface.

[0041] The actuators in Execution Layer III mainly include an instruction conversion section and a charged particle emission assembly.

[0042] The instruction conversion section converts the action instructions a generated by the reinforcement learning decision engine, such as increasing the emission current to X mA, into control instructions u that can be recognized by the charged particle emission component.

[0043] The charged particle emission assembly, installed near the exterior of the satellite, comprises a control unit and charged particle emitters. Multiple emitters can be deployed near critical surfaces, such as large-area dielectric surfaces, when it is necessary to control their potential. Driven by control command u, the charged particle emission assembly emits charged particles into space to neutralize the charge on the spacecraft's surface.

[0044] Step 4: Closed-loop feedback and online learning: After the charged particle emission component performs its action, the time-series data collected by the perception layer changes, generating new states and rewards, which are fed back to the reinforcement learning decision engine for online fine-tuning of the optimal control strategy, enabling it to continuously adapt to new environments that the spacecraft may encounter during its on-orbit lifespan.

[0045] The working principle of this embodiment is as follows: As the satellite is about to enter the sunlight zone from the shadow zone, the plasma environment will undergo a dramatic change. The AI ​​prediction model predicts in advance that the potential will shift negatively. Based on this, the reinforcement learning decision engine instructs the charged ion emission component to smoothly increase the current. When the satellite enters the sunlight zone, due to the advance intervention, the spacecraft potential only fluctuates slightly and quickly returns to stability. After this shadowing event, the system stores the complete data sequence (state, action, reward, new state) from T0 to T2 in an experience replay buffer. When onboard computing resources are idle, the system samples small batches of data from the buffer to fine-tune the model network online, thereby improving its ability to handle similar shadowing events and achieving lifelong on-orbit learning.

[0046] As can be seen, this invention is an intelligent closed-loop control and learning scheme from perception to execution, specifically including: a perception layer responsible for real-time data acquisition, as well as subsequent data preprocessing and feature extraction, providing high-quality input data for intelligent decision-making. The AI ​​intelligent decision-making center comprises two key steps: an AI prediction model and reinforcement learning decision-making, enabling the prediction of potential change trends and the generation of optimal control strategies. The execution layer translates decisions into specific control commands and drives actuator actions. Closed-loop feedback and learning: the system feeds the execution results back to the decision center for continuous model optimization and online learning, demonstrating the system's adaptability.

[0047] Example 2 like Figure 2 As shown, this embodiment provides an artificial intelligence-based spacecraft potential prediction and adaptive control system, including a perception layer I, a decision layer II, and an execution layer III.

[0048] Sensing layer I includes a potential and environmental state measurement module and a preprocessing module. Among them: The potential and environmental state measurement module is used to acquire time-series data, including real-time spacecraft potential and space plasma environment data surrounding the spacecraft. The potential and environmental state measurement module may include a spacecraft potential probe 1, a plasma probe 2, and other relevant spacecraft sensors 3 arranged on the spacecraft surface 9, such as… Figure 3 As shown, this method jointly achieves spacecraft potential sensing and surrounding space plasma environment sensing. The spacecraft potential probe is installed on the sun-facing side, the sun-facing side, and the sidewalls of the spacecraft; the plasma probe collects the density and temperature of the surrounding space plasma environment. The sensed time-series data is sent to the preprocessing module.

[0049] The preprocessing module preprocesses and extracts features from the time-series data and inputs it into decision layer II.

[0050] Decision Layer II is the AI ​​Intelligent Decision Center 5, which includes AI prediction models and reinforcement learning decision engines.

[0051] AI prediction models are used to generate predicted values ​​of spacecraft potential changes over a future period based on features from the perception layer.

[0052] The reinforcement learning decision engine constructs a state space from the features corresponding to the spacecraft's real-time potential and the predicted value of the spacecraft's potential change. It uses the action space to issue control commands to the execution layer and the reward R to maintain potential stability within a safe range. Through continuous interaction with the environment, it learns the optimal control strategy. Based on the optimal control strategy, the reinforcement learning decision engine generates action command 'a' and sends it to execution layer III.

[0053] Execution layer III includes a control command generation module and a charged particle emission assembly. Specifically: The control command generation module is used to convert the action command a from the decision layer II into a control command u and send it to the charged particle emission component.

[0054] like Figure 3 As shown, the charged particle emission assembly includes a charged particle emitter 4, and a control unit 7, a power supply unit 6, and a working propellant storage and supply unit 9 that support the parsing of control commands and the implementation of actions of the charged particle emitter 4. The control unit 7 is connected to the control command generation module and the working propellant storage and supply unit 8, and the control unit 7 and the working propellant storage and supply unit 8 are connected to the charged particle emitter 4. The control unit 7 receives and parses the control command u, driving the charged particle emitter 4 to emit charged particles into space to neutralize the charge on the surface of the spacecraft. Figure 3 The control command generation module is integrated into the control unit 7.

[0055] In practice, charged particle emission components can employ one or more of electron emission devices, ion emission devices, or plasma contactors.

[0056] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based spacecraft potential prediction and adaptive control method, characterized in that, The method comprises the following steps: Step 1: collecting time series data, including real-time spacecraft potential and spacecraft surrounding space plasma environment data; Step 2: an AI prediction model generates a spacecraft potential change prediction value for a future period of time based on the time series data; Step 3: a reinforcement learning decision engine uses the spacecraft real-time potential and the spacecraft potential change prediction value as a state space, issues a control instruction as an action space to an actuator, and maintains the potential stable in a safe interval as a reward, so as to learn an optimal control strategy through continuous interaction with the environment; and an action instruction a is generated based on the optimal control strategy; Step 4: the action instruction a generated by the reinforcement learning decision engine is converted into a control instruction to drive the charged particle emission assembly to work and emit charged particles into space to neutralize the charge on the surface of the spacecraft.

2. The method of claim 1, wherein, In step 2, the AI prediction model further generates prediction data of the spacecraft surrounding space plasma environment data for a future period of time based on the time series data; In step 3, the reinforcement learning decision engine further adds real-time collected data and prediction data of the spacecraft surrounding space plasma environment data to the state space.

3. The method of claim 1, wherein, The charged particle emission assembly is one or a combination of an electron emitter, an ion emission device, or a plasma contactor.

4. The method of claim 1, wherein, The step 1 further comprises filtering and normalizing the time series data.

5. The method of claim 1, wherein, The AI prediction model is a trained machine learning model; and the machine learning model is a long short-term memory network, a Transformer time series model, or a hybrid model of a convolutional neural network and a recurrent neural network.

6. The method of claim 1, wherein, The action instruction a output by the reinforcement learning decision engine comprises an emission current size, an emission timing, and an emission direction adjustment size.

7. The method of claim 1, wherein, The method further comprises the following steps: after the charged particle emission assembly performs the action, the collected time series data changes, new states and rewards are generated, and the new states and rewards are fed back to the reinforcement learning decision engine, so as to fine-tune the optimal control strategy online and make the optimal control strategy continuously adapt to new environments possibly encountered by the spacecraft during an on-orbit service life.

8. The method of claim 1, wherein, In step 1, the time series data is collected by a spacecraft potential probe and a plasma probe; The spacecraft potential probe is installed on a sun-facing surface, a sun-facing back surface, and a side wall of the spacecraft; The plasma probe collects density and temperature of a spacecraft surrounding space plasma environment.

9. The method of claim 1, wherein, The AI prediction model and the reinforcement learning decision engine are deployed in an AI chip of an on-board computer; during on-orbit operation, the AI chip is operated in a master control mode, a shadow mode is reserved and operated in parallel and continuously collects data, and the model is fine-tuned and updated regularly.

10. An artificial intelligence based spacecraft potential prediction and adaptive control system, characterized in that, The method comprises a perception layer, a decision layer, and an execution layer; The perception layer comprises a potential and environment state measurement module and a preprocessing module, which are used to collect time series data, including real-time spacecraft potential and spacecraft surrounding space plasma environment data; The preprocessing module pre-processes and extracts features of the time series data and inputs the time series data to the decision layer; The decision layer comprises an AI prediction model and a reinforcement learning decision engine; The AI prediction model is used to generate a spacecraft potential change prediction value for a future period of time based on features from the perception layer; The reinforcement learning decision engine is used to generate an action instruction a based on the optimal control strategy. The reinforcement learning decision engine constitutes a state space with the spacecraft real-time potential and the feature corresponding to the spacecraft potential change prediction value, issues a control instruction as an action space to an execution layer, takes maintaining the potential stable in a safety interval as a target as a reward, and learns an optimal control strategy by continuously interacting with the environment; The reinforcement learning decision engine generates an action instruction a based on the optimal control strategy and sends it to the execution layer; The execution layer includes a control instruction generation module and a charged particle emission component; the control instruction generation module is used to convert the action instruction a into a control instruction and drive the charged particle emission component to emit charged particles into space to neutralize the charge on the spacecraft surface.

Images