Spacecraft Hall propulsion device test working medium flow stable control system

Abstract

The invention discloses a spacecraft Hall propulsion device test working medium flow stability control system, which comprises a sensing module for detecting working condition state parameters of working medium supply equipment, and an adaptation module for exploring current working condition control parameters by combining detection data of the sensing module and similar working condition control parameter empirical data fusion, the adjusting module is used for receiving the control parameters and dynamically adjusting the adjusting weight of the adjusting device according to different phase change periods, an experience library unit and a self-adaption unit are arranged in the adaption module, the experience library unit provides experience data of similar working conditions and control parameters, and the self-adaption unit provides self-adaption data of the control parameters. And the autonomous adaptation unit fuses and explores control parameters by constructing a neural network adaptation model, so that working condition automatic identification and control parameter autonomous adaptation under the environment that the working medium type or the working medium state rapidly changes are realized; the problems of low control precision and slow response speed caused by manual presetting of control parameters of different working media in a traditional control technology are solved, and stable control of the working medium flow is realized.

Description

Technical Field

[0001] This invention belongs to the field of electric propulsion technology for spacecraft, specifically, it relates to a stable control system for the working fluid flow of a Hall thruster for spacecraft. Background Technology

[0002] In today's aerospace field, the development of spacecraft plays a crucial role in exploring the universe, expanding the boundaries of human knowledge, and driving industrial upgrading and new economic growth. Hall thrusters, as an advanced electric propulsion technology, provide precise and continuous thrust to spacecraft. With their significant advantages of high specific impulse, low thrust, re-start capability, light weight, and long lifespan, they are widely used in many critical mission stages of spacecraft, such as attitude control, orbital transfer and correction, position holding, and deorbit handling. To improve the operational performance of Hall thrusters during mission execution, practical applications typically involve conducting Hall thruster experiments to verify the consistency between theoretical design and engineering implementation.

[0003] In Hall thruster experiments, stable control of the working propellant flow rate is a key factor in ensuring its efficient and stable operation. Fluctuations in the working propellant flow rate directly affect important performance indicators such as thrust stability, specific impulse, and efficiency. Unstable working propellant flow rate can lead to thrust fluctuations, causing deviations in spacecraft orbit control and attitude adjustment, affecting the precise execution of missions; it can also cause instability in the plasma state inside the thruster, triggering abnormal discharges and thus shortening the thruster's lifespan. However, Hall thrusters need to switch between different working propellants according to mission requirements, such as switching between gaseous working propellants and phase-change working propellants like solid iodine based on the required thrust and endurance. Furthermore, the same working propellant can exhibit different states in the space environment, directly leading to significant differences in operating parameters. This diversity of working propellant states requires the working propellant flow control system to have cross-working propellant adaptability. Traditional control technologies rely on manually preset control parameters for different working propellants, which cannot cope with the requirements for control accuracy and response speed in the face of rapid changes in working propellant type or state. Therefore, how to improve the control accuracy and response speed of working propellant flow rate in complex and variable working propellant environments has become an urgent problem to be solved. Summary of the Invention

[0004] To overcome the above-mentioned shortcomings in the prior art, the present invention provides a working fluid flow stability control system for spacecraft Hall thruster test in environments with rapid changes in working fluid type or state, automatic identification of working conditions, autonomous adaptation of control parameters, and dynamic adjustment of adjustment weights of the adjustment device according to different phase transition periods, thereby improving control accuracy and response speed.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: The experimental working fluid flow stability control system for a Hall thruster in a spacecraft includes a sensing module for detecting the operating status parameters of the working fluid supply equipment, an adaptation module for fusing the detection data from the sensing module with empirical data of control parameters under similar operating conditions to explore the current operating condition control parameters, and an adjustment module for receiving the control parameters explored by the adaptation module and controlling the working fluid supply equipment. The sensing module includes a flow deviation calculation unit, a pressure sensing unit, a temperature sensing unit, a working fluid density calculation unit, a drive current sampling unit, and a valve opening sampling unit for detecting the operating status parameters of the working fluid supply equipment. The adaptation module includes an experience base unit and an autonomous adaptation unit. The experience base unit provides experience data on similar operating conditions and control parameters corresponding to the operating condition parameters detected by the sensing module. The autonomous adaptation unit constructs a neural network adaptation model to identify the operating conditions of the sensing module's detection data and fuses the data from the experience base unit to explore control parameters. The control parameters include phase change parameters, heating power parameters, PID parameters, and throttling opening parameters. The adjustment module includes a phase changer, a heating power regulator, a PID parameter regulator, and a throttling opening regulator that receive the control parameters. The phase changer dynamically adjusts the adjustment weights of the heating power regulator, PID parameter regulator, and throttling opening regulator according to different phase change periods. During the phase change start-up period, the adjustment weight of the heating power regulator is increased to prioritize working fluid vaporization. During the phase change stabilization period, the adjustment weight of the throttling opening regulator is increased to prioritize flow accuracy. The heating power regulator controls the working fluid vaporization state of the working fluid supply equipment, and the PID parameter regulator and throttling opening regulator control the test working fluid flow rate of the working fluid supply equipment.

[0006] Furthermore, the neural network adaptation model includes an input layer, a hidden layer, a fusion exploration layer, and an output layer. The input layer preprocesses the data detected by the perception module and generates a standardized state vector. The hidden layer extracts state features from the standardized state vector. The fusion exploration layer combines the state feature data and the experience data in the experience base unit to explore the current operating condition control parameters. The output layer transmits the explored current operating condition control parameters to the adjustment module. Furthermore, the exploration process of the fusion exploration layer is as follows: Extract the current working condition working fluid density, temperature, and pressure state characteristics from the hidden layer and calculate the temperature fluctuation variance that distinguishes between stable temperature and extreme temperature difference, and the pressure change rate that distinguishes between static and dynamic working conditions. The working fluid density is used to distinguish between gaseous working fluid and phase change working fluid. The K-means algorithm is used to classify historical operating conditions into five categories: gaseous stable, gaseous dynamic, phase transition stable, phase transition dynamic, and extreme environment. The current operating condition is determined by calculating the Euclidean distance between it and the center of each category to determine the most similar operating condition type. Based on the historical optimal control parameters of similar operating conditions, the initial control parameters of the current operating condition are used to shorten the convergence time; A dynamic decay exploration strategy is adopted to balance the exploration of the current operating condition with control stability. Different exploration frequencies are used for the initial, intermediate, and stable stages to complete the exploration and optimization of control parameters. Two convergence conditions are set: convergence of flow error and convergence of change in control parameter value. If either condition is met, the system is considered to have converged and enters a stable control mode.

[0007] Furthermore, in the dynamic decay exploration, the random exploration probability is set to 0.5 in the initial stage, 0.1 in the middle stage, and 0.01 in the stable stage.

[0008] Furthermore, the convergence condition for flow error is set to a flow error change of less than 0.1 sccm for fifty consecutive steps, and the convergence condition for control parameter value change is set to a control parameter value change of less than 0.01 for one hundred consecutive steps.

[0009] Furthermore, the experience base unit constructs the experience base and performs training iterations through the following steps: S2011: Five typical working conditions were simulated in the laboratory environment, namely gaseous stability, gaseous dynamics, phase transition stability, phase transition dynamics, and extreme environment. 20,000 experience points were collected for each working condition to build an initial experience database of 100,000 samples. S2012: The Adam optimizer is used to train for one hundred epochs. The trained model is then solidified into the ONNX format and converted into an engine file executable by the computing device using TensorRT. S2013: After the system is powered on, the offline training model is loaded, new working condition experience is collected in real time, one old experience is replaced for every ten new experiences, and the model is fine-tuned every hundred steps. S2014: Automatically backs up the online optimized model daily, and rolls back to the previous day's optimal model if an anomaly occurs.

[0010] Furthermore, the experience base unit includes a priority processing subunit. The priority processing subunit sets a time difference error function to measure the value priority of experience, allocates sampling probability through a sampling probability function, and sets a sampling weight function to balance the sampling advantage of high-priority experience with the training value of low-priority experience.

[0011] Furthermore, the time difference error function is:

[0012] In the formula, This represents the absolute value of the time difference error. This represents the immediate reward value that measures the immediate effect of the current action. This represents a discount factor that balances immediate rewards with future rewards. Indicates the current state. Indicates the current action. Indicates the next state. Indicates the next action. This means taking the maximum value among all possible actions for the next state. This represents the predicted action value for the next state. This represents the predicted action value for the current state; The sampling probability function is:

[0013] In the formula, Indicates the first The sampling probability of the empirical evidence, Indicates the first Prioritizing experience points This represents the priority weight coefficient. This represents the weighted sum of the priorities of all experiences; The sampling weight function is:

[0014] In the formula, Indicates the first The importance of each experience point is sampled by weight. This represents the total capacity of the experience base units. Indicates the first The sampling probability of the empirical evidence, This represents the deviation correction factor. Represents the unnormalized original weights. This represents the maximum value of all original empirical weights.

[0015] Furthermore, based on the hardware characteristics of the adaptation module, the adaptation module is divided into a PS control area using a serial computing architecture and a PL operation area using a parallel computing architecture. The experience library unit is placed in the PS control area, and the autonomous adaptation unit is placed in the PL operation area.

[0016] Furthermore, the PS control area and PL calculation area interact with each other via the AXI bus.

[0017] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention detects the working condition parameters of the working fluid supply equipment through a sensing module, explores the current working condition control parameters through an adaptation module, and receives the control parameters and controls the working fluid supply equipment through an adjustment module. The sensing module detects six-dimensional working condition parameters, including flow deviation, pressure, temperature, working fluid density, driving current, and valve opening, which overcomes the problem of inaccurate judgment of working condition by a single parameter. The adaptation module processes these six-dimensional working condition parameters simultaneously by constructing a neural network adaptation model and explores four types of control parameters: phase change parameters, heating power parameters, PID parameters, and throttling opening parameters. Through the coordinated operation of the three modules, this invention realizes automatic identification of working condition and autonomous adaptation of control parameters in environments with rapid changes in working fluid type or working fluid state. It also optimizes the control by dynamically adjusting the adjustment weight of the adjustment device according to different phase change periods. This invention overcomes the problem of low control accuracy and slow response speed caused by the reliance on manual preset control parameters for different working fluids in traditional control technology.

[0018] (2) The neural network adaptation model of this invention, through its built-in input layer, hidden layer, fusion exploration layer, and output layer, realizes autonomous exploration of the mapping relationship between operating conditions and control parameters without the need for preset control rules. The fusion exploration layer identifies the most similar operating condition type based on the characteristic data of working fluid density, temperature fluctuation variance, and pressure change rate, determines the initial control parameters of the current operating condition, and completes the exploration and optimization of control parameters by adopting a dynamic decay exploration strategy. It also sets two convergence conditions: flow error convergence and control parameter value change convergence. When either convergence condition is met, it enters a stable control mode, reducing parameter fluctuations. In addition, this invention balances operating condition exploration and control stability by setting different random exploration probabilities in the initial, intermediate, and stable stages.

[0019] (3) The experience base constructed by the experience base unit of this invention covers five typical working conditions. The working condition experience sample data has a large capacity. Moreover, through offline training and online learning, the experience base data has been optimized and iterated. In addition, the priority handling subunit measures the value priority of experience by setting a time difference error function, allocates sampling probability by sampling probability function, and sets a sampling weight function to balance the sampling advantage of high priority experience and the training value of low priority experience, thus providing better experience data for the exploration sampling of the fusion exploration layer, thereby improving the control accuracy.

[0020] (4) This invention divides the adaptation module into a PS control area and a PL calculation area. The PS control area adopts a serial computing architecture as the general control core, while the PL calculation area adopts a parallel computing architecture as the parallel computing core. This partitioning allows for full utilization of the respective hardware advantages and avoids performance bottlenecks caused by a single module needing to be both fast and comprehensive. The PS control area and the PL calculation area are connected via an AXI bus. The AXI bus has high bandwidth and low latency, enabling efficient data interaction and transmission. This provides a real-time communication link for efficient collaboration between the PS control area and the PL calculation area, and for the autonomous adaptation unit to explore experience library unit data in real time. Attached Figure Description

[0021] Figure 1 This is a schematic diagram illustrating the structural principle of the experimental working fluid flow stabilization control system for the Hall thruster of the present invention.

[0022] Figure 2 This is a schematic diagram of the exploration process of the neural network adaptation model fusion exploration layer of the present invention.

[0023] Figure 3 This is a schematic diagram illustrating the steps involved in constructing the experience base and the training iteration process for the experience base unit of this invention.

[0024] Figure 4 This is a schematic diagram showing the composition of the working fluid supply equipment of the present invention and its relationship with the sensing module and the adjustment module.

[0025] In the above figures, the component names corresponding to the reference numerals are as follows: 1-Sensing Module, 101-Flow Deviation Calculation Unit, 102-Pressure Sensing Unit, 103-Temperature Sensing Unit, 104-Working Fluid Density Calculation Unit, 105-Drive Current Sampling Unit, 106-Valve Opening Sampling Unit, 2-Adaptation Module, 201-Experience Library Unit, 201a-Priority Handling Subunit, 202-Autonomous Adaptation Unit, 3-Regulation Module, 301-Phase Changer, 302-Heating Power Regulator, 303-PID Parameter Regulator, 304-Throttle Opening Regulator, 4-Working Fluid Supply Equipment, 401-Working Fluid Storage Tank, 402-Heating Module, 403-Gas Supply Pipeline, 404-Valve, 405-Actuator, 5-Hall Effect Propulsion Device. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments.

[0027] Example like Figures 1 to 4As shown, this embodiment provides a test working fluid flow stability control system for a spacecraft Hall thruster, including a sensing module 1, an adaptation module 2, and an adjustment module 3. The sensing module 1 includes a flow deviation calculation unit 101, a pressure sensing unit 102, a temperature sensing unit 103, a working fluid density calculation unit 104, a drive current sampling unit 105, and a valve opening sampling unit 106, which detect six dimensions of working conditions of the working fluid supply device 4: flow deviation, pressure, temperature, working fluid density, drive current, and valve opening. The adaptation module 2 receives the six-dimensional working condition status parameter data from the sensing module 1 and explores the current working condition control parameters by combining the neural network adaptation model in the built-in autonomous adaptation unit 202 with the data from the built-in experience base unit 201. The adjustment module 3 is equipped with a phase changer 301, a heating power regulator 302, a PID parameter regulator 303, and a throttling opening regulator 304, which receive four types of control parameters explored by the adaptation module 2: phase change parameters, heating power parameters, PID parameters, and throttling opening parameters. In this embodiment, the working fluid supply device 4 includes a working fluid storage tank 401 for storing the working fluid, a heating module 402 for heating and vaporizing the working fluid in the working fluid storage tank 401, a gas supply pipeline 403 for transmitting the gaseous working fluid in the working fluid storage tank 401 to the Hall thruster 5, a valve 404 provided on the gas supply pipeline 403, and a driver 405 for controlling the opening degree of the valve 404.

[0028] In this embodiment, the pressure sensing unit 102 and temperature sensing unit 103 detect the pressure and temperature of the working fluid storage tank 401, respectively; the valve opening sampling unit 106 detects the opening of the valve 404; the drive current sampling unit 105 detects the drive current of the driver 405; and the flow deviation calculation unit 101 calculates the deviation between the actual flow rate and the target flow rate, using the calculated flow deviation as the operating condition parameter data. In this embodiment, the working fluid density is a key parameter connecting the working fluid state and the stable supply of flow rate. Its calculation accuracy directly affects the accuracy of the adaptation module 2 in identifying the operating condition and the rationality of the control parameter output. The working fluid density calculation unit 104 derives and calculates the density by combining the detected pressure and temperature data with the thermodynamic characteristics of the working fluid. In this embodiment, the following formula is used to calculate the density of the gaseous working fluid:

[0029] In the formula, This represents the density of the gaseous working fluid. Indicates the pressure value. Indicates molar mass. Represents the gas constant. Indicates the temperature value; The density of a phase change working fluid is calculated using the following formula:

[0030] In the formula, This represents the density of the phase change working fluid. Indicates the temperature value. This represents the saturation density curve based on experimental fitting.

[0031] In this embodiment, the heating power regulator 302 adjusts the vaporization state of the working fluid by controlling the heating module 402 in the working fluid supply device 4. In this embodiment, the PID parameter regulator 303 uses the abbreviation PID for Proportional Integral Derivative, which means proportional, integral, and derivative. The PID parameter regulator 303 is essentially a closed-loop control algorithm based on deviation feedback. Its core function is to calculate the deviation between the actual value and the target value, and combine the synergistic effect of the three parameters of proportional, integral, and derivative to output a precise control signal, so that the actual value stably approaches the target value. The PID parameter adjustment control method has been applied in practice, and its specific control principle will not be elaborated in this embodiment. The PID parameter regulator 303 and the throttling opening regulator 304 work together to control the actuator 405, and the actuator 405 in turn controls the opening of the valve 404. For phase change working fluids, this embodiment sets up a phase changer 301 to monitor the phase change state. The phase changer 301 dynamically adjusts the adjustment weights of the heating power regulator 302, PID parameter regulator 303, and throttling opening regulator 304 according to different phase change stages. During the phase change start-up stage, the adjustment weights of the heating power regulator 302, PID parameter regulator 303, and throttling opening regulator 304 are set to 50%, 30%, and 20%, respectively. By increasing the adjustment weight of the heating power regulator 302, the vaporization of the working fluid is prioritized. During the phase change stabilization stage, the adjustment weights of the heating power regulator 302, PID parameter regulator 303, and throttling opening regulator 304 are set to 20%, 30%, and 50%, respectively. By increasing the adjustment weight of the throttling opening regulator 304, the flow accuracy is prioritized.

[0032] This embodiment's neural network adaptation model is based on a multilayer perceptron and long short-term memory network neural network structure. It achieves intelligent algorithm model for parameter prediction by learning the mapping relationship between input and output parameters in the system. This embodiment's neural network adaptation model includes an input layer, a hidden layer, a fusion exploration layer, and an output layer. The input layer preprocesses the data detected by the perception module 1 and generates a standardized state vector. The hidden layer extracts state features from the standardized state vector. The fusion exploration layer combines the state feature data and the experience data in the experience base unit 201 to explore the current operating condition control parameters. The output layer transmits the explored current operating condition control parameters to the adjustment module 3. The exploration process of the fusion exploration layer in this embodiment is as follows: Extract the current working condition working fluid density, temperature, and pressure state characteristics from the hidden layer, and calculate the temperature fluctuation variance that distinguishes between stable temperature and extreme temperature difference, and the pressure change rate that distinguishes between static and dynamic working conditions. Among them, the working fluid density is used to distinguish between gaseous working fluid and phase change working fluid. The K-means algorithm is used to classify historical operating conditions into five categories: gaseous stable, gaseous dynamic, phase transition stable, phase transition dynamic, and extreme environment. The current operating condition is determined by calculating the Euclidean distance between it and the center of each category to determine the most similar operating condition type. Based on the historical optimal control parameters of similar operating conditions, the initial control parameters of the current operating condition are used to shorten the convergence time; A dynamic decay exploration strategy is adopted to balance the exploration of the current operating condition with control stability. Different exploration frequencies are used for the initial, intermediate, and stable stages to complete the exploration and optimization of control parameters. Two convergence conditions are set: convergence of flow error and convergence of change in control parameter value. If either condition is met, the system is considered to have converged and enters a stable control mode.

[0033] In this embodiment, the K-means algorithm, also known as the K-means clustering algorithm, is an unsupervised learning algorithm. Its core function is to automatically divide an unlabeled dataset into K non-overlapping subsets based on data similarity. Data within each subset is highly similar, while data differences between different subsets are significant. In this embodiment, Euclidean distance is the core metric for measuring the similarity between the current working condition feature vector and the cluster centers of each category. A smaller distance indicates greater similarity between the working conditions. Its calculation formula is set as follows:

[0034] In the formula, Indicates the current operating condition and the first Euclidean distance between cluster centers Indicates the feature dimension number. Indicates the current operating condition. Standardized values ​​of each feature Indicates the first Cluster center The mean of each feature.

[0035] In this embodiment, during the dynamic decay exploration, the random exploration probability is set to 0.5 in the initial stage, increasing the exploration rate by increasing the exploration frequency; in the intermediate stage, the random exploration probability is set to 0.1, reducing the exploration frequency and optimizing control parameters based on existing experience; in the stable stage, the random exploration probability is reduced to 0.01, primarily utilizing optimal experience to ensure stable control. The flow error convergence condition in this embodiment is set as follows: the flow error change is less than 0.1 sccm for fifty consecutive steps, meaning that in fifty consecutive flow measurements, the difference between the flow value at each step and the previous step does not exceed 0.1 standard cubic centimeters per minute. The control parameter value change convergence condition is set as follows: the control parameter value change is less than 0.01 for one hundred consecutive steps, meaning that in one hundred consecutive control cycles, the difference between the control parameter value at each step and the previous step does not exceed 0.01.

[0036] In this embodiment, the experience base unit 201 constructs the experience base and performs training iterations through the following steps: S2011: Five typical working conditions were simulated in the laboratory environment, namely gaseous stability, gaseous dynamics, phase transition stability, phase transition dynamics, and extreme environment. 20,000 experience points were collected for each working condition to build an initial experience database of 100,000 samples. S2012: The Adam optimizer is used to train for one hundred epochs. The trained model is then solidified into the ONNX format and converted into an engine file executable by the computing device using TensorRT. S2013: After the system is powered on, the offline training model is loaded, new working condition experience is collected in real time, one old experience is replaced for every ten new experiences, and the model is fine-tuned every hundred steps. S2014: Automatically backs up the online optimized model daily, and rolls back to the previous day's optimal model if an anomaly occurs.

[0037] In this embodiment, step S2012, through the process of Adam optimizer training, ONNX format solidification, and TensorRT conversion, achieves efficient execution of the experience base from algorithm design to computing device. The Adam optimizer, also known as the Adaptive Momentum Estimation Optimizer, dynamically adjusts the update step size of each parameter by calculating the first and second momentum of the parameter gradients, achieving fast and stable convergence with different learning rates for different parameters. An Epoch refers to a period of time; in experience base training, one Epoch is the process of the model completely traversing all training set samples. The ONNX format is an open-source model serialization format across deep learning frameworks, and TensorRT is a deep learning inference optimization software development kit. This embodiment, by constructing and training an iterative experience base, provides a high-quality data foundation for the fusion of the exploration layer.

[0038] To further improve the efficiency of exploration and adoption in the fusion exploration layer, the experience base unit 201 includes a priority processing subunit 201a. The priority processing subunit 201a sets a temporal difference error function to measure the value priority of experiences, allocates sampling probabilities through a sampling probability function, and sets a sampling weight function to balance the sampling advantage of high-priority experiences with the training value of low-priority experiences. The temporal difference error function is as follows:

[0039] In the formula, This represents the absolute value of the time difference error. This represents the immediate reward value that measures the immediate effect of the current action. This represents a discount factor that balances immediate rewards with future rewards. Indicates the current state. Indicates the current action. Indicates the next state. Indicates the next action. This means taking the maximum value among all possible actions for the next state. This represents the predicted action value for the next state. This represents the predicted action value for the current state; The sampling probability function is:

[0040] In the formula, Indicates the first The sampling probability of the empirical evidence, Indicates the first Prioritizing experience points This represents the priority weight coefficient. This represents the weighted sum of the priorities of all experiences; The sampling weighting function is:

[0041] In the formula, Indicates the first The importance of each experience point is sampled by weight. This represents the total capacity of the experience base units. Indicates the first The sampling probability of the empirical evidence, This represents the deviation correction factor. Represents the unnormalized original weights. This represents the maximum value of all original empirical weights.

[0042] In this embodiment, the adaptation module 2 is divided into a PS control area and a PL computation area. The PS control area uses a serial computing architecture as its general control core, while the PL computation area uses a parallel computing architecture as its parallel computing core. This partitioning fully leverages the advantages of each hardware component, avoiding performance bottlenecks caused by a single module trying to be both fast and comprehensive. In this embodiment, the experience library unit 201 is placed in the PS control area to avoid occupying the computing power required for the autonomous adaptation unit 202. The autonomous adaptation unit 202 is placed in the PL computation area for efficient computation, improving exploration efficiency. The PS control area and the PL computation area are connected via an AXI bus. AXI stands for Advanced eXtensible Interface, which offers high bandwidth and low latency, enabling efficient data exchange and transmission. This provides a real-time communication link for efficient collaboration between the PS control area and the PL computation area, and for the autonomous adaptation unit 202 to explore data from the experience library unit 201 in real time.

[0043] In use, this invention deploys the components of a working fluid flow stabilization control system in a test environment and connects the system to a working fluid supply device 4 containing gaseous and phase-change working fluids. The working fluid supply device 4 is connected to a Hall thruster 5. Simultaneously, an experience database is built in the experience database unit 201, and a neural network adaptation model is built in the autonomous adaptation unit 202. This invention possesses cross-working fluid adaptation capabilities and can simultaneously process six-dimensional operating condition data, including flow deviation, pressure, temperature, working fluid density, drive current, and valve opening. It achieves automatic identification of complex operating conditions in environments with rapid changes in working fluid type or state, and supports four types of control parameters: phase change parameters, heating power parameters, PID parameters, and throttling opening parameters. By autonomously adapting parameters and dynamically adjusting the adjustment weights of the regulating device according to different phase transition periods to optimize control, this invention overcomes the problem that traditional control technologies rely on manually preset control parameters for different working media, which cannot cope with the control accuracy and response speed issues caused by rapid changes in working media type or state. Moreover, this invention also adopts a dynamic attenuation exploration strategy to balance the exploration of the current working condition with control stability. By setting a priority processing subunit 201a in the experience base unit 201 to optimize the experience base data, and by dividing the adaptation module 2 into a PS control area and a PL calculation area and using an AXI bus for data interaction, the system control accuracy and response speed are further improved, thereby achieving stable control of the working media flow.

[0044] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any changes made based on the design principles of the present invention, or any non-creative modifications made thereon, shall fall within the scope of protection of the present invention.

Claims

1. A stable control system for the working fluid flow rate of a Hall thruster test device for a spacecraft, characterized in that... The system includes a sensing module (1) for detecting the working condition parameters of the working fluid supply equipment (4), an adaptation module (2) for exploring the current working condition control parameters by combining the detection data of the sensing module (1) and the experience data of similar working condition control parameters, and an adjustment module (3) for receiving the control parameters explored by the adaptation module (2) and controlling the working fluid supply equipment (4). The sensing module (1) includes a flow deviation calculation unit (101), a pressure sensing unit (102), a temperature sensing unit (103), a working fluid density calculation unit (104), a drive current sampling unit (105), and a valve opening sampling unit (106) for detecting the working condition parameters of the working fluid supply equipment (4). The adaptation module (2) includes an experience base unit (201) and an autonomous adaptation unit (202). The experience base unit (201) provides experience data on similar working conditions and control parameters corresponding to the working condition parameters detected by the sensing module (1). The autonomous adaptation unit (202) constructs a neural network adaptation model to identify the working conditions of the data detected by the sensing module (1) and combines the data from the experience base unit (201) to explore control parameters. The control parameters include phase change parameters, heating power parameters, PID parameters, and throttling opening parameters. The adjustment module (3) includes a phase changer (301), a heating power regulator (302), and a PID parameter regulator (303) that receive the control parameters. 03) Throttling opening regulator (304), wherein the phase changer (301) dynamically adjusts the adjustment weights of the heating power regulator (302), PID parameter regulator (303), and throttling opening regulator (304) according to different phase change periods. During the phase change start-up period, the adjustment weight of the heating power regulator (302) is increased to prioritize the vaporization of the working fluid. During the phase change stabilization period, the adjustment weight of the throttling opening regulator (304) is increased to prioritize the flow accuracy. The heating power regulator (302) controls the vaporization state of the working fluid in the working fluid supply device (4), and the PID parameter regulator (303) and throttling opening regulator (304) control the test working fluid flow rate of the working fluid supply device (4).

2. The spacecraft Hall thruster experimental working propellant flow stabilization control system according to claim 1, characterized in that: The neural network adaptation model includes an input layer, a hidden layer, a fusion exploration layer, and an output layer. The input layer preprocesses the data detected by the perception module (1) and generates a standardized state vector. The hidden layer extracts state features from the standardized state vector. The fusion exploration layer combines the state feature data and the experience data in the experience base unit (201) to explore the current operating condition control parameters. The output layer transmits the explored current operating condition control parameters to the adjustment module (3).

3. The spacecraft Hall thruster experimental working fluid flow stabilization control system according to claim 2, characterized in that, The exploration process of the fusion exploration layer is as follows: Extract the current working condition working fluid density, temperature, and pressure state characteristics from the hidden layer and calculate the temperature fluctuation variance that distinguishes between stable temperature and extreme temperature difference, and the pressure change rate that distinguishes between static and dynamic working conditions. The working fluid density is used to distinguish between gaseous working fluid and phase change working fluid. The K-means algorithm is used to classify historical operating conditions into five categories: gaseous stable, gaseous dynamic, phase transition stable, phase transition dynamic, and extreme environment. The current operating condition is determined by calculating the Euclidean distance between it and the center of each category to determine the most similar operating condition type. Based on the historical optimal control parameters of similar operating conditions, the initial control parameters of the current operating condition are used to shorten the convergence time; A dynamic decay exploration strategy is adopted to balance the exploration of the current operating condition with control stability. Different exploration frequencies are used for the initial, intermediate, and stable stages to complete the exploration and optimization of control parameters. Two convergence conditions are set: convergence of flow error and convergence of change in control parameter value. If either condition is met, the system is considered to have converged and enters a stable control mode.

4. The spacecraft Hall thruster experimental working fluid flow stabilization control system according to claim 3, characterized in that, In the dynamic decay exploration, the random exploration probability is set to 0.5 in the initial stage, 0.1 in the middle stage, and 0.01 in the stable stage.

5. The experimental working fluid flow stability control system for the Hall thruster of a spacecraft according to claim 3, characterized in that, The convergence condition for flow error is set to the flow error change being less than 0.1 sccm for 50 consecutive steps, and the convergence condition for control parameter value change is set to the control parameter value change being less than 0.01 for 100 consecutive steps.

6. The experimental working fluid flow stabilization control system for the Hall thruster of a spacecraft according to claim 3, characterized in that, The experience base unit (201) constructs the experience base and performs training iterations through the following steps: S2011: Five typical working conditions were simulated in the laboratory environment, namely gaseous stability, gaseous dynamics, phase transition stability, phase transition dynamics, and extreme environment. 20,000 experience points were collected for each working condition to build an initial experience database of 100,000 samples. S2012: The Adam optimizer is used to train for one hundred epochs. The trained model is then solidified into the ONNX format and converted into an engine file executable by the computing device using TensorRT. S2013: After the system is powered on, the offline training model is loaded, new working condition experience is collected in real time, one old experience is replaced for every ten new experiences, and the model is fine-tuned every hundred steps. S2014: Automatically backs up the online optimized model daily, and rolls back to the previous day's optimal model if an anomaly occurs.

7. The experimental working fluid flow stability control system for the Hall thruster of a spacecraft according to claim 6, characterized in that, The experience base unit (201) includes a priority processing subunit (201a). The priority processing subunit (201a) sets a time difference error function to measure the value priority of experience, allocates sampling probability through a sampling probability function, and sets a sampling weight function to balance the sampling advantage of high-priority experience with the training value of low-priority experience.

8. The experimental working fluid flow stability control system for the Hall thruster of a spacecraft according to claim 7, characterized in that, The time difference error function is: ； In the formula, This represents the absolute value of the time difference error. This represents the immediate reward value that measures the immediate effect of the current action. This represents a discount factor that balances immediate rewards with future rewards. Indicates the current state. Indicates the current action. Indicates the next state. Indicates the next action. This means taking the maximum value among all possible actions for the next state. This represents the predicted action value for the next state. This represents the predicted action value for the current state; The sampling probability function is: ； In the formula, Indicates the first The sampling probability of the empirical evidence, Indicates the first Prioritizing experience points This represents the priority weight coefficient. This represents the weighted sum of the priorities of all experiences; The sampling weight function is: ； In the formula, Indicates the first The importance of each experience point is sampled by weight. This indicates the total capacity of the experience base units. Indicates the first The sampling probability of the empirical evidence, This represents the deviation correction factor. Represents the unnormalized original weights. This represents the maximum value of all original empirical weights.

9. The experimental working fluid flow stabilization control system for the Hall thruster of a spacecraft according to any one of claims 1-8, characterized in that: Based on the hardware characteristics of the adapter module (2), the adapter module (2) is divided into a PS control area with a serial computing architecture and a PL operation area with a parallel computing architecture. The experience library unit (201) is placed in the PS control area, and the autonomous adaptation unit (202) is placed in the PL operation area.

10. The experimental working fluid flow stabilization control system for the Hall thruster of a spacecraft according to claim 9, characterized in that: The PS control area and PL calculation area exchange data via the AXI bus.

Images