A power supply controller on-orbit evaluation method, system and terminal

Abstract

The present application belongs to the technical field of power supply controller, and relates to a method and system for evaluating a power supply controller in orbit and a terminal. The method for evaluating the power supply controller in orbit comprises the following steps: collecting multi-dimensional telemetry data such as bus output voltage, power supply controller internal temperature, temperature measurement and heat pipe heat dissipation surface temperature; and then based on the telemetry data, evaluating from four dimensions of operation stability, working reliability, in-orbit performance and health state, wherein the operation stability is determined according to fluctuation analysis or classification model, the reliability is identified by relying on compliance statistics or mutation detection, the in-orbit performance evaluation is based on time series prediction or reconstruction error analysis, and the health state evaluation is combined with electrical parameter deviation, thermoelectric coupling characteristics and integrated learning model; and finally, the multi-dimensional results are fused to generate a comprehensive evaluation index. Through the cooperation of multi-source data fusion and multi-dimensional algorithm, the present application realizes all-around accurate characterization of the equipment state, effectively improves the evaluation accuracy and robustness, and has engineering practical value.

Description

Technical Field

[0001] This invention relates to the field of power controller technology, and in particular to a method, system and terminal for on-orbit evaluation of power controllers. Background Technology

[0002] The Power Control Unit (PCU) is the core unit of a spacecraft's energy system. It is mainly responsible for regulating the power of the solar array, controlling the charging and discharging of batteries, and ensuring stable power supply to the load. Its on-orbit health status and operational reliability directly determine the energy supply capability of the spacecraft throughout its entire life cycle. Therefore, conducting accurate and comprehensive on-orbit health and performance evaluations has crucial engineering application value.

[0003] At present, the on-orbit evaluation technology of spacecraft PCU has many shortcomings, including fragmented evaluation system, poor environmental adaptability, and insufficient data support. It cannot meet the high-precision and multi-dimensional evaluation requirements under complex dynamic conditions in orbit. The specific technical shortcomings are as follows.

[0004] First, existing evaluation systems have certain limitations, making it difficult to achieve comprehensive and systematic evaluation. Current mainstream evaluation systems are mainly divided into three categories, each with certain technical shortcomings. Firstly, the fault diagnosis and basic performance monitoring system based on built-in test (BIT) technology only monitors a few fixed parameters such as bus voltage ripple and shunt regulator efficiency. It primarily focuses on the identification and location of known fault modes, resulting in a relatively singular evaluation dimension. It does not fully consider key aspects such as equipment lifespan degradation trends and multi-parameter coupling adaptability. Furthermore, its fault diagnosis relies on pre-set fault samples, and its ability to identify unknown failure modes needs improvement, leading to incomplete evaluation results. Secondly, the performance evaluation system guided by ground simulation tests relies on ground simulation equipment and accelerated aging tests for evaluation. It is difficult to accurately reproduce real-world operating conditions such as extreme temperature difference cycles, space radiation, and dynamic energy allocation in orbit. The differences between the Earth and space environments cause a certain deviation between the evaluation results and the actual in-orbit state, making it difficult to fully reflect the true dynamic operating characteristics of the PCU. Third, the special evaluation system based on a single component or single indicator only evaluates a single component or a single health indicator. It does not establish a system-level evaluation framework, does not fully consider the coupling relationship between various internal functional modules, and is difficult to comprehensively characterize the overall health status of the PCU, thus having limited support for energy dispatch optimization decisions.

[0005] Second, the lack of basic evaluation data hinders accurate assessment. Existing technologies rely on two main types of data, both with fatal flaws. One type is ground-based simulation test data, which, due to differences between space and ground environments, fails to reflect extreme on-orbit conditions and dynamic operational characteristics, resulting in highly distorted performance predictions. The other type is traditional on-board telemetry data, which only collects a few parameters such as bus voltage and output current. This data is singular in dimension and lacks multi-source correlation, failing to support multi-dimensional and systematic evaluation, further exacerbating the limitations of the evaluation results.

[0006] In summary, existing PCU on-orbit evaluation technologies generally suffer from common problems such as limited evaluation dimensions, poor environmental adaptability, low systematization, and insufficient data support. The evaluation results have limited reference value and cannot form a systematic evaluation capability that adapts to complex on-orbit operating conditions, making it difficult to meet the operation and maintenance needs of spacecraft for long-term reliable on-orbit operation. Summary of the Invention

[0007] The purpose of this invention is to address the limitations of existing PCU on-orbit evaluation technologies, such as limited evaluation dimensions, poor environmental adaptability, low systematization, and insufficient data support. It proposes an on-orbit evaluation method, system, and terminal for power controllers. By enriching data acquisition, integrating multi-source algorithms, and constructing a multi-dimensional system, it achieves comprehensive and accurate characterization of the PCU's on-orbit status, improving evaluation accuracy and robustness, and ensuring the effective implementation of the evaluation.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides an on-orbit evaluation method for a power controller, comprising the following steps: Acquire telemetry data from multiple dimensions of the power controller during its on-orbit operation. The telemetry data should include at least the bus output voltage, the internal temperature of the power controller, the temperature measurement of the power controller, the temperature measurement of the power controller heat pipe, and the temperature measurement of the power controller heat dissipation surface. Based on telemetry data, the power controller is evaluated from multiple dimensions, including at least: operational stability evaluation, operational reliability evaluation, on-orbit performance evaluation, and health status evaluation. Among them, operational stability evaluation is achieved based on fluctuation analysis or state classification model of telemetry data; operational reliability evaluation is achieved based on parameter compliance statistics or mutation detection; on-orbit performance evaluation is achieved based on time series prediction model or reconfiguration error analysis; and health status evaluation is achieved based on electrical parameter deviation analysis, thermo-electric coupling analysis or ensemble learning model. By integrating the evaluation results from various dimensions, a comprehensive evaluation index for the power controller is generated.

[0009] As one possible approach, obtaining operational stability evaluation results through volatility analysis specifically includes: Based on the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement, calculate the standard deviation of each parameter within the sliding window, and obtain the mean of the standard deviation of the multi-window. The single-parameter stability score is calculated based on the ratio of the mean standard deviation of each parameter to the preset stability threshold. The smaller the ratio, the higher the score, and a score of 0 is given when the ratio exceeds 1. The entropy weight method is used to calculate the weight based on the dispersion of the standard deviation of each parameter window; the greater the dispersion, the higher the weight. The stability evaluation result is obtained by weighted summation of the stability scores of each parameter; or, The application of volatility analysis to obtain stability evaluation results specifically includes: Based on the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement, and power controller heat dissipation surface temperature measurement, the mean and standard deviation of each parameter within the sliding window are calculated, and then the coefficient of variation CV is calculated as: standard deviation / mean × 100%. The mean of the coefficients of variation for each parameter across all windows is calculated to obtain the mean of the multi-window CV. The single-parameter stability evaluation result is determined by comparing the mean CV of multiple windows with the preset threshold and according to the graded scoring rules. The thresholds include normal threshold, attention threshold, warning threshold and emergency threshold. The mean value of the stability evaluation results for each parameter is taken as the stability evaluation result; and / or The stability evaluation results are obtained by applying a state classification model, specifically including: Using the time-series data of bus output voltage and power controller internal temperature as input, a support vector machine model is used to output the current state identification label, which is either normal or abnormal. Within a sliding window, count the number of normal status identification tags, the number of abnormal status identification tags, and the total number of tags; The stability score is calculated using the following formula: Score = max{0, (number of normal state identification labels within the window / total number of labels) × 10 - (number of abnormal state identification labels × 0.5)}.

[0010] As one possible implementation method, obtaining operational reliability evaluation results through application parameter compliance statistics specifically includes: For each sampling point of the bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement, determine whether it is within the preset safety threshold range. If it is, the state quantity is recorded as 0; otherwise, it is recorded as 1. Add anti-shake logic: The parameter is only confirmed to be out of limit when three consecutive sampling points are all recorded as 1; otherwise, it is corrected to 0. Count the number of sampling points with a state value of 0 and the total number of valid sampling points within a day, and calculate the normality rate of each parameter; The average of the normality rates of each parameter is taken to obtain the comprehensive compliance rate of the multiple parameters, which is the result of the work reliability evaluation; and / or, The application of mutation detection to obtain operational reliability evaluation results specifically includes: Calculate the first-order difference sequence for the time series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement respectively. Points whose absolute difference exceeds the preset mutation threshold are identified as mutations, and the number of mutations and normal occurrences for each parameter within the period are counted. The reliability of each parameter is calculated using the following formula: Reliability = (Number of normal occurrences / Total number of occurrences) × 100%; The entropy weight method is used to calculate the weights based on the degree of dispersion of the difference sequences of each parameter; The reliability of each parameter is weighted and summed to obtain the comprehensive reliability of the multi-parameter parameters, which is the result of the operational reliability evaluation.

[0011] As one possible implementation, the time series prediction model includes a performance change trend score based on an LSTM network and a performance change trend score based on a Bayesian network. The LSTM-based performance change trend score includes: using time series data of bus output voltage and power controller internal temperature as input, and employing a Long Short-Term Memory (LSTM) network to output the anomaly probability value for each time step; calculating the first-order difference of the anomaly probability values ​​and counting the total number of first-order differences; summing the first-order differences and recording the sign of the result; counting the number of differences in the first-order difference sequence that have the same sign as the summation result and the number of differences that are different; and calculating the trend score using the following formula: Score = (Number of identical differences - Number of identical differences) / Total number × 10. The performance trend score based on Bayesian networks includes: using time-series data of bus output voltage and power controller internal temperature as input, the Bayesian network outputs the normal pattern similarity at the current moment; calculating the first-order difference of the similarity and counting the total number of first-order differences; summing the first-order differences and recording the sign of the result; counting the number of differences in the first-order difference sequence that have the same sign as the summation result and the number of differences that are different; calculating the trend score using the following formula: Score = (Number of identical differences - Number of different differences) / Total number × 10; or, The application of reconstruction error analysis methods to obtain on-orbit performance evaluation results specifically includes: The timing data of the bus output voltage and the internal temperature of the power controller are used as inputs, and the reconstruction error is output by a deep self-encoder. Calculate the temporal rate of change of the reconstruction error, i.e., the first-order difference; The proportion of negative rates of change within the sliding window is used as the proportion of state improvement. The health trend index is calculated using the following formula: Health trend index = percentage of improvement in condition × (1 - normalized value of average reconstruction error); The health trend index is normalized and multiplied by 10 to obtain the performance degradation trend score.

[0012] As one possible approach, obtaining health status assessment results through electrical parameter deviation analysis specifically includes: Obtain real-time data and rated voltage of the bus output voltage; Calculate the voltage deviation rate at each sampling point; The average deviation rate within the statistical data collection period; Based on the average deviation, the normalized bus output voltage health score is calculated, which is the health status evaluation result; and / or, The application of thermo-electric coupling analysis to obtain health status assessment results specifically includes: Acquire timing data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement, and power controller heat dissipation surface temperature measurement. Calculate the Pearson correlation coefficient between any two parameters and construct the original correlation coefficient matrix; Take the absolute value of the matrix elements and normalize them to obtain a normalized health-related matrix; The mean of the off-diagonal elements of the normalized matrix is ​​taken to obtain the global coupling health score, which is then multiplied by 10 to convert it into a 10-point score, thus obtaining the health status evaluation result; and / or, The ensemble learning model includes single-machine health scores based on the XGBoost model and single-machine health scores based on the random forest model. Specifically, the single-machine health score based on the XGBoost model is calculated by using the bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input features, and outputting anomaly probability values ​​using the XGBoost model. The health score is calculated as follows: Health score = (1 - anomaly probability value) × 10. The single-machine health score based on the random forest model includes: using bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input features, and using the random forest model to output anomaly probability values; the health score is calculated according to the following formula: Health score = (1 - anomaly probability value) × 10; and / or, Health status assessment also includes evaluation using unsupervised learning models. These unsupervised learning models include single-machine health scores based on the Local Outlier Factor (LOF) algorithm, single-machine health scores based on the Isolation Forest model, and single-machine health scores based on a one-dimensional residual network. Specifically, the single-machine health score based on the LEF algorithm involves: using the bus output voltage and power controller internal temperature as input samples, calculating the LOF score using the LEF algorithm; determining the maximum value of the LOF score in the training set as the normal threshold; normalizing the LOF score of the current sample: if the LOF score ≤ the normal threshold, the normalized value = 1 - (LOF score / normal threshold); if the LOF score > the normal threshold, the normalized value = 0.5 - (LOF score - normal threshold) / (2 × normal threshold); and calculating the health score using the following formula: Health score = normalized value × 10. The single-machine health score based on the isolated forest model includes: using the bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input features, and using the isolated forest model to output the anomaly probability value; the health score is calculated as follows: health score = (1 - anomaly probability value) × 10; The single-machine health score based on a one-dimensional residual network includes: taking the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input, and using a one-dimensional residual network to output the health state probability; the health score is calculated as follows: health score = health state probability × 10.

[0013] As one possible implementation, health status assessment also includes fusing the evaluation results of electrical parameter deviation analysis methods, thermo-electric coupling analysis methods, ensemble learning models, and unsupervised learning models to obtain a comprehensive health score, specifically including: Health scores based on XGBoost, Random Forest, Local Outlier Factor, Isolated Forest and One-Dimensional Residual Network are extracted, and the minimum value is taken as the health score of the neural network class. The comprehensive health score is obtained by arithmetically averaging the health score of the neural network class with the health score of the single-machine electrical parameters based on the voltage deviation rate and the health score of the thermo-electric coupling based on the thermo-electric parameter correlation matrix.

[0014] As one possible approach, the evaluation results from various dimensions are integrated to generate a comprehensive evaluation index for the power controller. Specifically, this includes mapping the evaluation results from each dimension to generate stability level, reliability level, performance level, and health level. Alternatively, the evaluation results from each dimension can be weighted and integrated to generate a comprehensive evaluation index for the power controller.

[0015] Secondly, the present invention provides an on-orbit evaluation system for a power controller, comprising: The data acquisition module is used to acquire telemetry data from multiple dimensions of the power controller during its on-orbit operation. The telemetry data includes at least the bus output voltage, the internal temperature of the power controller, the temperature measurement of the power controller, the temperature measurement of the power controller heat pipe, and the temperature measurement of the power controller heat dissipation surface. The multi-dimensional evaluation module is used to evaluate the power controller from multiple dimensions based on telemetry data: operational stability evaluation, operational reliability evaluation, on-orbit performance evaluation, and health status evaluation. The integrated evaluation module is used to integrate the evaluation results from various dimensions and generate a comprehensive evaluation index for the power controller.

[0016] Thirdly, the present invention provides a terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the on-orbit evaluation method for power controllers proposed in this invention.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. The present invention proposes an on-orbit evaluation method for power controllers, which enriches the dimensions of on-board telemetry data acquisition. It addresses the problems of insufficient evaluation data support and poor environmental adaptability in existing evaluation methods by expanding the data acquisition scope to cover key parameters such as PCU electrical, thermal management, and mission performance. This provides sufficient data support for multi-dimensional evaluation, ensures the implementation of the evaluation system, and improves the comprehensiveness and accuracy of the evaluation.

[0018] 2. The present invention proposes an on-orbit evaluation method for power controllers, which integrates innovative multi-source evaluation calculation methods and reliability evaluation dimensions. It adopts a collaborative calculation of basic index calculation, neural network prediction and data feature analysis. By quantifying core parameters, mining hidden failure patterns, extracting abnormal features and integrating evaluation results, it significantly improves the accuracy and robustness of PCU reliability evaluation and adapts to the evaluation needs of complex dynamic on-orbit operating conditions.

[0019] 3. The present invention proposes an on-orbit evaluation method for power controllers, which innovatively constructs a multi-dimensional evaluation system. Addressing the problem of the single evaluation dimension in existing methods, this method establishes an evaluation framework that covers operational stability, operational reliability, lifespan health, and on-orbit performance degradation trends, based on the core requirements of PCU on-orbit energy security. This comprehensively characterizes the on-orbit health status of the PCU, solves the problem of incomplete evaluation results that can lead to biased operation and maintenance decisions, and provides a reliable reference for on-orbit operation and maintenance decisions. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings: Figure 1This is a flowchart of the on-orbit evaluation method for the power controller in an embodiment of the present invention; Figure 2 This is a flowchart illustrating the process of obtaining the on-orbit stability evaluation index values ​​of the power controller in this embodiment of the invention. Figure 3 This is a flowchart illustrating the process of obtaining the on-orbit reliability evaluation index values ​​of the power controller in this embodiment of the invention. Figure 4 This is a flowchart illustrating the process of obtaining the on-orbit performance evaluation index values ​​of the power controller in this embodiment of the invention. Figure 5 This is a flowchart illustrating the process of obtaining the on-orbit health status index value of the power controller in an embodiment of the present invention. Detailed Implementation

[0021] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. For example, the first threshold and the second threshold are merely used to distinguish different thresholds and do not limit their order. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" are not necessarily different.

[0022] It should be noted that in this invention, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0023] In this invention, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one" or similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, "at least one of a, b, or c" can represent: a, b, c, a combination of a and b, a combination of a and c, a combination of b and c, or a, b, and c, where a, b, and c can be single or multiple.

[0024] In a first aspect, embodiments of the present invention provide an on-orbit evaluation method for a power controller, see [link to relevant documentation]. Figure 1 This includes the following steps: (a) Data collection Acquire telemetry data from multiple dimensions of the power controller during its on-orbit operation. The telemetry data should include at least the bus output voltage, the internal temperature of the power controller, the temperature measurement of the power controller, the temperature measurement of the power controller heat pipe, and the temperature measurement of the power controller heat dissipation surface. Among them, the power controller is the core control unit of the power system, undertaking the key functions of energy coordination and distribution, power supply stability assurance and fault protection. It needs to accurately connect with energy components such as solar cell arrays and batteries and various loads, and achieve efficient and reliable operation of the system by dynamically adjusting energy flow.

[0025] Among them, the bus output voltage is the core power indicator of the PCU supplying power to the system load. It directly determines whether the load can obtain stable and compatible power. Its stability is the key to ensuring the normal operation of the system. If the voltage is too high, it will burn out the load. If it is too low, the load will not be able to start.

[0026] The voltage is a continuous numerical time-domain data: under normal conditions, it fluctuates slightly around the rated reference value, with the fluctuation amplitude within the allowable error. The data distribution is concentrated and the time-domain waveform is flat, which can provide stable power supply. Under abnormal conditions, the fluctuation frequency increases and the amplitude exceeds the rated error. In extreme cases, overvoltage or undervoltage spikes occur, the time-domain waveform oscillates irregularly, and the dispersion is significantly increased, which may lead to abnormal downstream loads or even damage to sensitive electronic devices.

[0027] Among them, the internal temperature of the power controller is a basic indicator reflecting the thermal state of the core working environment of the equipment. It is directly related to the operational stability of the internal electronic components. Exceeding the temperature limit will accelerate component aging and cause performance drift. It is also an important basis for determining whether the equipment needs to activate heat dissipation protection.

[0028] The internal temperature is a continuous numerical time-domain data: under normal conditions, it is stable within the allowable operating temperature range of the device, and changes slowly and slightly with the ambient temperature, with minimal fluctuations and no sudden changes. The time domain shows a stable and gradual trend. When abnormal, the temperature shows a significant trend change or a sudden increase in temperature, and the value exceeds the safety threshold. The time domain waveform rises sharply, reflecting the internal device overheating fault. This will continue to lead to the degradation of controller performance or even permanent damage.

[0029] Among them, temperature measurement of the power controller is a means of monitoring the overall thermal status of the equipment. By accurately collecting temperature data, it provides support for assessing the safety of the equipment's working environment and predicting the risk of thermal failure. Inaccurate temperature measurement can lead to misjudgment or missed judgment by the thermal protection mechanism.

[0030] The temperature measurement is a high-precision continuous numerical time-domain data: under normal conditions, it can accurately reflect the actual internal temperature, fluctuating slightly around the true temperature, with small measurement error and strong anti-electromagnetic interference capability, and the time-domain changes are synchronized with the actual internal temperature changes; under abnormal conditions, the data drift increases, deviates from the actual temperature, or there are irregular jumps that are not synchronized with the internal temperature changes, which will lead to misjudgment of the temperature control strategy, failure to trigger temperature control measures in time, and indirectly affect the working stability of the controller.

[0031] Among them, the heat pipe temperature measurement of the power controller is a special monitoring indicator of the core heat dissipation component. It directly reflects the heat pipe heat dissipation efficiency and working status. Abnormal temperature will cause heat to accumulate in the core component, which is a key monitoring point to ensure the normal operation of the heat dissipation system.

[0032] The temperature measurement is a continuous numerical time-domain data: under normal conditions, it fluctuates slightly and smoothly around the reference range of the heat pipe's working temperature. The time-domain changes are consistent with the trends of the controller's internal temperature and ambient temperature, which can reflect the normal heat dissipation and conduction state of the heat pipe. In case of abnormality, the temperature changes suddenly or remains in the abnormal range. The time domain shows a significant trend deviation, which reflects the failure of the heat pipe's heat dissipation function. This will lead to the interruption of the controller's heat dissipation path and a continuous increase in internal temperature.

[0033] Among them, the temperature measurement of the heat dissipation surface of the power controller is a direct indicator for evaluating the heat dissipation effect of the equipment. If the temperature is too high, it will cause internal heat accumulation, and if it is too low, it may reflect that the heat dissipation system is over-operated. It is an important reference for optimizing heat dissipation strategies and ensuring the thermal balance of the equipment.

[0034] The temperature measurement is a continuous numerical time-domain data: under normal conditions, it is stable within the design operating temperature range, and adjusts slowly with the ambient temperature and the internal temperature of the controller. The fluctuation is minimal and the time-domain waveform is smooth, which can reflect the normal heat dissipation status of the heat dissipation surface. When abnormal, the temperature exceeds the design threshold, shows obvious trend changes or fluctuates irregularly, and the time-domain waveform fluctuates violently, reflecting problems such as dirt accumulation or damage on the heat dissipation surface, which leads to a decrease in heat dissipation efficiency and makes it impossible to ensure that the controller is in a safe operating temperature environment.

[0035] (II) Multi-dimensional evaluation Based on telemetry data, the power controller is evaluated from multiple dimensions, including at least: operational stability evaluation, operational reliability evaluation, on-orbit performance evaluation, and health status evaluation. 1. Operational stability evaluation Among them, see Figure 2 Operational stability evaluation is achieved based on fluctuation analysis or state classification models of telemetry data; As one possible approach, obtaining operational stability evaluation results through volatility analysis specifically includes: (1) Multi-parameter integrated stability Based on the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement, calculate the standard deviation of each parameter within the sliding window, and obtain the mean of the standard deviation of the multi-window. As an example, when collecting time-series data, the sampling frequency is 1Hz, which is 1 data point per second, conforming to the standard engineering configuration and requiring no additional hardware upgrades. The default scrolling window is 30 seconds (including 30 data points), which can be adjusted to 10 seconds or 60 seconds as needed. 10 seconds provides a fast response, while 60 seconds offers better smoothness. Extreme outliers are removed, such as single temperature jumps >5℃, current exceeding the rated range by 20%, and speed deviation from the design value by ±10%. Data from the previous moment is used to replace these outliers, avoiding interference with the standard deviation calculation. Data loss is ensured. If data is missing at a certain moment, the average value of the previous window is temporarily used to fill the gap, ensuring the continuity of the scrolling calculation.

[0036] As an example, the formula for calculating the standard deviation of each parameter within the sliding window is: Where n > 30, that is, continuous data points are taken according to the time series, and the number of data points is greater than 30.

[0037] As an example, the mean of multi-window standard deviations is calculated, which is the arithmetic mean of the standard deviations across all windows for each parameter. Window standard deviations and means can accurately capture the short-term fluctuation characteristics of parameters, helping to locate potential anomalies such as high-frequency fluctuations.

[0038] The single-parameter stability score is calculated based on the ratio of the mean standard deviation of each parameter to the preset stability threshold. The smaller the ratio, the higher the score, and a score of 0 is given when the ratio exceeds 1. As an example, the ratio of the mean standard deviation of each parameter to the preset stability threshold is denoted as r. Score = 10 - 10 × r (r ≤ 1) Rating = 0 (r > 1) When r > 1, the parameter fluctuation has exceeded the reasonable range, the stability is extremely poor, and a score of 0 is given directly; the smaller r is, the stronger the stability and the higher the score.

[0039] The entropy weight method is used to calculate the weight based on the dispersion of the standard deviation of each parameter window; the greater the dispersion, the higher the weight. The stability scores of each parameter are weighted and summed to obtain the stability evaluation result; and / or, As an example, the stability evaluation result = Σ(single parameter score × corresponding parameter weight) Among them, the stability evaluation results quantify the stability level of a single machine, and the higher the score, the more stable the operation.

[0040] In summary, the bus output voltage, internal temperature, measured temperature, heat pipe temperature, and heat dissipation surface temperature of the power controller are key indicators for the continuous and reliable on-orbit power supply of the spacecraft's power system. Abnormal fluctuations in these parameters can lead to decreased power supply voltage accuracy, load imbalance in the thermal control system, and long-term degradation may even trigger overheating protection of the power module and power outages. Using standard deviation as the core calculation indicator, through multi-step data processing, weight calculation, and comprehensive scoring, the stability evaluation results are ultimately presented as an intuitive normalized score, providing data support for monitoring the operating status and providing early warning of faults in the power controller.

[0041] (2) Voltage-temperature stability The application of volatility analysis to obtain stability evaluation results specifically includes: Based on the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement, and power controller heat dissipation surface temperature measurement, the mean and standard deviation of each parameter within the sliding window are calculated, and then the coefficient of variation CV is calculated as: standard deviation / mean × 100%. As an example, when collecting time-series data, the sampling frequency is consistent with the telemetry downlink frequency, usually 1~10Hz, to ensure that the data density supports window calculation; a fixed rolling window is used, such as 1 hour / 1 track task cycle, and the proportion of effective data points in each window is ≥95%; physical anomalies and special operating condition data, such as equipment start-up and shutdown, power supply mode switching periods, are removed.

[0042] Among them, the coefficient of variation (CV) of the bus output voltage, the internal temperature of the power controller, and the temperature measured by the power controller are the core indicators for quantifying its voltage-temperature stability. The rolling trend can accurately reflect the dynamic changes in the performance and health status of the power supply regulation system, thermal control system, and core components such as power modules and heat pipes.

[0043] The mean of the coefficients of variation for each parameter across all windows is calculated to obtain the mean of the multi-window CV. The single-parameter stability evaluation result is determined by comparing the mean CV of multiple windows with the preset threshold and according to the graded scoring rules. The thresholds include normal threshold, attention threshold, warning threshold and emergency threshold. As an example, the normal threshold (CV0) is 10%; the attention threshold (CV1=1.2×CV0); the warning threshold (CV2=1.5×CV0); and the emergency threshold (CV3=2.0×CV0). The normal threshold needs to be set according to the actual engineering situation and serves as a benchmark for judging whether the parameter fluctuation is normal. If it is exceeded, it is considered to be an abnormal stability.

[0044] As an example, the grading system: The mean CV value is less than the normal threshold CV0, which is 10 points, indicating stability. The mean CV value is between CV0 and CV1, around 8 points, and is basically stable. The mean CV value exceeds CV1 by 6 points, indicating increased volatility. The mean CV value exceeds the warning threshold CV2, score 4, indicating degradation; The mean CV value exceeds the emergency threshold CV3, score 2, which is abnormal.

[0045] The mean value of the stability evaluation results for each parameter is taken as the stability evaluation result; and / or In summary, this technical solution achieves normalization and standardization of fluctuation assessment through the coefficient of variation, overcoming the limitations of different parameter dimensions and mean differences, and enabling horizontal comparison of stability across parameter types. Based on a four-level threshold scoring rule, a full-link early warning system is constructed, with clear hierarchy and explicit criteria, facilitating rapid identification of equipment stability degradation stages. Single-parameter scoring can pinpoint weak points in stability, while the comprehensive score of a single unit reflects the overall stability level of the equipment, providing quantitative basis for on-orbit operation and maintenance decisions, and improving equipment operational reliability and the precision of operation and maintenance.

[0046] (3) On-orbit stability The stability evaluation results are obtained by applying a state classification model, specifically including: Using the time-series data of bus output voltage and power controller internal temperature as input, a support vector machine model is used to output the current state identification label, which is either normal or abnormal. As an example, the timing data of the bus output voltage and the internal temperature of the power controller are input into a support vector machine (SVM), and the SVM output is a state identification label for the power controller.

[0047] As an example, if the bus output voltage deviates from the rated value by more than 5% and the internal temperature is greater than 60°C, it is considered abnormal.

[0048] Support Vector Machine (SVM) is a supervised machine learning algorithm suitable for binary classification tasks with small samples and high-dimensional features. It is applicable to scenarios where the boundary between normal and abnormal categories is unclear in equipment condition monitoring, and can accurately capture the correlation patterns of power controller parameters. SVM searches for the optimal hyperplane in the high-dimensional feature space to maximize the margin between the two classes. For nonlinear data, it maps to the high-dimensional space using a kernel function to achieve linear separability, and the classification boundary is determined by the support vectors, ensuring generalization ability and avoiding overfitting. SVM clearly determines whether the operating condition is normal or abnormal, exhibits strong classification robustness, can resist slight noise interference from telemetry data, and by analyzing the parameter features corresponding to the support vectors, it can help understand the core basis for anomaly determination.

[0049] Among them, the bus output voltage directly reflects the stability of the power supply output and is the foundation for the reliable operation of the load equipment; the internal temperature of the power controller reflects the thermal operating safety of the power module, and abnormal temperature will accelerate component aging and trigger protection shutdown. Both are related to the core operating conditions of the power controller, providing data support for SVM to process high-dimensional feature classification tasks, which is in line with its application characteristics in equipment condition monitoring.

[0050] Within a sliding window, count the number of normal status identification tags, the number of abnormal status identification tags, and the total number of tags; The stability score is calculated using the following formula: Score = max{0, (number of normal state identification labels within the window / total number of labels) × 10 - (number of abnormal state identification labels × 0.5)}.

[0051] In summary, the output of SVM is used to calculate the on-orbit stability score of the power controller. This score quantifies the stability of the power controller in real time. The higher the score, the smaller the fluctuations in power supply and thermal operation and the stronger the stability. It can be directly used as a decision-making basis for temporarily adjusting the operating parameters of the power controller and carrying out fault early warning.

[0052] As an example, the arithmetic mean of the multi-parameter comprehensive stability score, the on-orbit operation stability score, and the voltage-temperature stability score is fused, and the result is used as the operation stability evaluation score, which is retained to one decimal place. The operation stability evaluation result is determined according to the graded scoring rules. The operational stability evaluation score is denoted as S. 9 ≤ S ≤ 10 points, "Excellent"; 8 ≤ S < 9 points, "Good"; 6≤S<8 points, “medium”; 3 ≤ S < 6 points, "Pass"; 0 ≤ S < 3 is classified as "range".

[0053] In summary, this technical solution integrates the advantages of accurate quantification of basic indicators, robust classification of neural networks, and cross-parameter adaptation of data features. It avoids the limitations of single calculation logic, does not require preset weights, and the results objectively reflect the overall stability level of the power controller, meeting the intuitive interpretation needs of engineering operation and maintenance.

[0054] 2. Operational reliability evaluation Operational reliability assessment is achieved based on parameter compliance statistics or mutation detection; see [link / reference]. Figure 3 ; As one possible implementation method, obtaining operational reliability evaluation results through application parameter compliance statistics specifically includes: (1) Multi-parameter comprehensive compliance rate For each sampling point of the bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement, determine whether it is within the preset safety threshold range. If it is, the state quantity is recorded as 0; otherwise, it is recorded as 1. Among them, the compliance of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement, and power controller heat dissipation surface temperature measurement is the core foundation for the reliable operation of the power controller. These parameters are directly related to the power supply stability and thermal safety of the power system. Exceeding the limit of any parameter will increase the risk of power failure, thereby affecting the power supply reliability of other single-unit load devices.

[0055] As an example, perform a two-branch logical operation on the real-time data of each sampling point: If the lower threshold ≤ parameter ≤ upper threshold, it is determined that the state is normal and the state quantity is recorded as 0. If the parameter is less than the lower threshold or greater than the upper threshold, it is determined that the state is out of limit and the state variable is recorded as 1.

[0056] That is: 0 = normal, 1 = exceeding the limit; Attached over-limit information: Over-limit type, such as internal temperature over-limit, voltage over-limit; Core over-limit data: current parameter value, over-limit start time (accurate to the second), duration; Add anti-shake logic: The parameter is only confirmed to be out of limit when three consecutive sampling points are all recorded as 1; otherwise, it is corrected to 0. As an example, anti-shake logic is added to avoid misjudgments caused by transient interference; if a single frame exceeds the limit and immediately returns to normal, it is judged as transient interference, and the state value is corrected to 0 to ensure the accuracy of the warning.

[0057] Count the number of sampling points with a state value of 0 and the total number of valid sampling points within a day, and calculate the normality rate of each parameter; As an example, calculate the normality rate of parameters: The result should be rounded to two decimal places.

[0058] The average of the normality rates of each parameter is taken to obtain the comprehensive compliance rate of the multiple parameters, which is the result of the work reliability evaluation; and / or, As an example, the comprehensive compliance rate of multiple parameters of a power controller is a quantitative indicator of its reliable operation level. The higher the score, the more the electrical and thermal parameters of the power controller are within the normal range, and the more stable and reliable the operation. If the score is below 80%, the status of the components corresponding to the out-of-limit parameters should be closely monitored to avoid the spread of the fault and affect the power supply to the load.

[0059] (2) Operational reliability The application of mutation detection to obtain operational reliability evaluation results specifically includes: Calculate the first-order difference sequence for the time series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement respectively. Among them, the first-order difference captures sudden changes in data by calculating the difference between adjacent data points. It can be used to monitor sudden changes in bus output voltage and various temperature measurement parameters, and to help determine abnormal parameter frequencies caused by power supply regulator module failure, temperature measurement link abnormality, etc.

[0060] As an example, the first-order difference = the next data point - the previous data point. The difference between each adjacent data point is calculated to form a first-order difference sequence. This sequence is used to traverse the time-domain data of each parameter to eliminate the influence of the overall data trend and highlight local changes.

[0061] Points whose absolute difference exceeds the preset mutation threshold are identified as mutations, and the number of mutations and normal occurrences for each parameter within the period are counted. The reliability of each parameter is calculated using the following formula: Reliability = (Number of normal occurrences / Total number of occurrences) × 100%; The entropy weight method is used to calculate the weights based on the degree of dispersion of the difference sequences of each parameter; The reliability of each parameter is weighted and summed to obtain the comprehensive reliability of the multi-parameter parameters, which is the result of the operational reliability evaluation.

[0062] As an example, multi-parameter integrated reliability calculation: in, For the reliability of each parameter, For weights.

[0063] In summary, this technical solution uses first-order differential to accurately capture instantaneous changes in parameters, overcoming the limitations of conventional threshold exceedances that rely on gradual trends. It can identify early degradation signals in advance, improving the foresight and accuracy of fault warnings. Single-parameter reliability can locate high-risk weak links and assist in specialized operation and maintenance. The comprehensive reliability of a single machine reflects the overall operating level of the equipment, providing a quantitative basis for on-orbit operation and maintenance related decisions.

[0064] As an example, the arithmetic mean of the multi-parameter comprehensive compliance rate and operational reliability is combined, and the result is rounded to two decimal places; to obtain the operational reliability evaluation score, and the operational reliability evaluation result is determined according to the graded scoring rules; Grading and scoring rules: The job reliability evaluation score is denoted as R. 98≤R≤100, Excellent; 90≤R<98, good; 80≤R<90, moderate; 65≤R<80, poor; R < 65, range.

[0065] In summary, this technical solution combines the complementary advantages of long-term time-series statistics of basic indicators and data feature mutation capture. It reflects the level of parameters meeting safety thresholds over a long period of time, while also taking into account the risk warning of instantaneous anomalies. It is unweighted and the results are comprehensive and meet the reliability requirements of power controllers for long-term stability and emergency protection.

[0066] 3. On-orbit performance evaluation On-orbit performance evaluation is achieved based on time-series prediction models or reconstruction error analysis. See [link / reference] Figure 4 ; (1) Time series prediction model As one possible implementation, time-series prediction models include performance change trend scores based on LSTM networks and performance change trend scores based on Bayesian networks.

[0067] ① Performance trends of LSTM-based networks Among them, the performance change trend score based on the LSTM network includes: taking the time series data of bus output voltage and power controller internal temperature as input, and using a long short-term memory network to output the abnormal probability value of each time step.

[0068] Among them, the bus output voltage reflects the stability of the power supply output and is the basis for the reliable operation of the load equipment; the internal temperature of the power controller reflects the thermal safety of the power module. Abnormal temperature will accelerate the aging of components and trigger protection shutdown.

[0069] Among them, Long Short-Term Memory (LSTM) networks, as an improved algorithm of Recurrent Neural Networks (RNNs), can effectively capture long-term dependencies in time-series data and are suitable for dynamic evaluation tasks of device performance change trends. Its core principle is to selectively retain and forget time-series information through a gating mechanism consisting of input gates, forget gates, and output gates, thereby alleviating the gradient vanishing problem that easily occurs in traditional RNNs. LSTM is applied to accurately learn the time-series evolution of power controller parameters, outputting the anomaly probability value at each time step, and thus deriving the performance change trend.

[0070] As an example, the probability value of an anomaly at each time step ranges from 0 to 1. The closer the probability value is to 1, the higher the probability that the power controller is in an abnormal state at the current moment; the closer it is to 0, the closer the operating state is to normal.

[0071] Calculate the first-order difference of the outlier probability values ​​and count the total number of first-order differences; Summing the first-order differences, and recording the sign of the result; As an example, if the result is 0, the performance change trend score will be output directly as 0; Count the number of differences in a first-order difference sequence that have the same sign as the summation result and the number of differences that are different; The trend score is calculated using the following formula: Score = (Number of identical differences - Number of different differences) / Total number × 10; The score quantifies the state evolution trend of the power controller on a 10-point scale. The higher the score, the more significant the trend of the state deteriorating, providing a dynamic basis for early warning of potential faults.

[0072] ②Performance trends based on Bayesian networks The performance change trend score based on Bayesian network includes: taking the time series data of bus output voltage and power controller internal temperature as input, and using Bayesian network to output the normal pattern similarity at the current moment; Among them, Yesian networks are graphical models based on probabilistic reasoning, with extremely high interpretability. They can efficiently handle uncertain relationships between variables, making them highly suitable for probabilistic assessment tasks of equipment performance change trends. Yesian networks use directed acyclic graphs (DAGs) to clearly depict the causal dependencies between various variables, and combined with Bayes' theorem, accurately calculate the similarity between the current operating state of the equipment and the standard normal mode, thereby deriving the evolution trend of equipment performance, perfectly adapting to the uncertain relationship characteristics of power controller parameters.

[0073] Calculate the first-order difference of similarity and count the total number of first-order differences; Summing the first-order differences, and recording the sign of the result; As an example, if the result is 0, the performance change trend score will be output directly as 0; Count the number of differences in a first-order difference sequence that have the same sign as the summation result and the number of differences that are different; The trend score is calculated using the following formula: Score = (Number of identical differences - Number of different differences) / Total number × 10; The score quantifies the state evolution trend of the power controller on a 10-point scale. The higher the score, the more significant the trend of the state deteriorating, providing a dynamic basis for early warning of potential faults.

[0074] or, (2) Reconstruction error analysis The application of reconstruction error analysis methods to obtain on-orbit performance evaluation results specifically includes: The timing data of the bus output voltage and the internal temperature of the power controller are used as inputs, and the reconstruction error is output by a deep self-encoder. Among them, the deep autoencoder (DAE) is an unsupervised deep learning algorithm that can automatically extract latent features from high-dimensional data, making it suitable for unsupervised evaluation tasks of device performance change trends. The DAE maps input data to a low-dimensional feature space through an encoder, and then reconstructs the input data through a decoder. Using reconstruction error as the core metric, it quantifies the degree of deviation of the input data from the normal pattern, thereby deriving the device performance change trend and adapting to the complex mode characteristics of power controller parameters.

[0075] The higher the reconstruction error value, the more the current operating state of the power controller deviates from the reconstruction result of the normal mode; the lower the reconstruction error value, the closer the state is to normal.

[0076] Calculate the temporal rate of change of the reconstruction error, i.e., the first-order difference; The proportion of negative rates of change within the sliding window is used as the proportion of state improvement. The health trend index is calculated using the following formula: Health trend index = percentage of improvement in condition × (1 - normalized value of average reconstruction error); The health trend index is normalized and multiplied by 10 to obtain the performance degradation trend score.

[0077] Among them, a 10-point scale is used to quantify the state trend of the power controller: the higher the score, the smaller the reconstruction error, the closer the state is to the normal mode, and the better the performance trend; the lower the score, the larger the reconstruction error and the more obvious the state deviation, providing dynamic quantitative basis for fault warning and health assessment.

[0078] As an example, the trend scores obtained from the time series prediction model and the reconstruction error analysis are fused by arithmetic mean, that is, the on-orbit performance evaluation score P = (LSTM performance change trend score + Bayesian network performance change trend score + DAE performance degradation trend score) ÷ 3, and the result is rounded to one decimal place. As an example, the on-orbit performance evaluation results are determined according to the graded scoring rules; Grading and scoring rules: 9≤P≤10, Excellent; 7 ≤ P < 9, good; 6≤P<7, qualified; P < 6, unqualified.

[0079] In summary, this technical solution integrates the algorithmic advantages of multiple neural networks: LSTM captures long-term trends, Bayesian networks adapt to uncertainties, and DAE unsupervised identification of anomalies. It avoids overfitting or feature omissions by a single model, requires no weight setting, and the results accurately reflect the true trend of power controller performance degradation or improvement.

[0080] 4. Health Status Assessment Health status assessment is implemented based on electrical parameter deviation analysis, thermo-electric coupling analysis, or ensemble learning models. See [link / reference]. Figure 5 ; (1) Analysis of electrical parameter deviation As one possible approach, obtaining health status assessment results through electrical parameter deviation analysis specifically includes: Obtain real-time data and rated voltage of the bus output voltage; As an example, real-time data is collected, and the sampling frequency is synchronized with the telemetry downlink frequency, for example, the sampling frequency is 1~10Hz to ensure real-time capture of voltage changes; lost telemetry data and voltage jump values ​​without cause are removed, for example, a single jump amplitude exceeding the rated voltage by 5% without external factors such as load adjustment or power supply switching; the proportion of valid sampling points within the data acquisition period is ≥98% to avoid evaluation bias caused by missing data.

[0081] Calculate the voltage deviation rate at each sampling point; Excessive voltage deviation rate directly reflects the decline in the voltage regulation performance of the power module, the aging of the filter link, or abnormal load adaptation. Long-term deviation will lead to reduced power supply accuracy, damage to the working stability of the load equipment, and ultimately trigger the overload protection of the power module or even power outage.

[0082] The average deviation rate within the statistical data collection period; As an example, in, This represents the relative deviation rate.

[0083] in, denoted as the average deviation rate, and M is the number of valid sampling points within the sampling period.

[0084] Based on the average deviation, the normalized bus output voltage health status is calculated, which is the health status evaluation result. As an example, calculate the normalized bus output voltage health: in, The health score ranges from 0 to 10. The closer the score is to 10, the better the health status of the bus output voltage.

[0085] And / or, In summary, this technical solution quantifies the health level of the bus output voltage through relative deviation rate calculation and normalization, providing a quantitative basis for on-orbit health management and fault early warning of the power controller's electrical system.

[0086] Among them, the voltage deviation rate and average deviation rate accurately capture the degree of deviation of the bus output voltage, and can directly locate abnormal fluctuations in the voltage regulation performance of the power controller; the normalized health score (0~10 points) intuitively reflects the health level of the bus output voltage, and can be linked with indicators such as temperature health and stability to form a comprehensive health assessment system for the power controller, providing quantitative support for on-orbit fault early warning and maintenance decisions.

[0087] (2) Thermal-electric coupling analysis Power controllers exhibit strong thermo-electric coupling characteristics. The synergistic relationship between voltage output and temperature field is the core indicator of their health. Stable coupling indicates efficient coordination between power supply regulation and thermal control system, while weakened coupling indicates deterioration of core components and decline in system health.

[0088] The application of thermo-electric coupling analysis to obtain health status assessment results specifically includes: Acquire timing data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement, and power controller heat dissipation surface temperature measurement. As an example, time-series data are collected, with continuous sampling over long time series, and a total sampling length of ≥200 points to ensure the statistical significance of correlation calculations; sampling is performed at equal time steps, with data integrity ≥98%, and data from transient operating conditions such as equipment start-up and shutdown and power supply mode switching are excluded; each parameter sequence has no constant value, as constant value sequences cannot reflect coupling relationships and have no analytical significance. If constant values ​​exist, they should be marked and the coupling calculation of that parameter pair should be excluded.

[0089] Calculate the Pearson correlation coefficient between any two parameters and construct the original correlation coefficient matrix; As an example, construct the original correlation coefficient matrix: Where, r ij Let be the Pearson correlation coefficient between any two parameters, i,j=1−5, where 1 corresponds to U, 2-5 correspond to the four temperature parameters, rij∈[−1,1], and the larger |rij| is, the stronger the coupling between the parameters; finally, a 5×5 original correlation coefficient matrix R is formed.

[0090] Take the absolute value of the matrix elements and normalize them to obtain a normalized health-related matrix; As an example, the normalized health-related matrix is ​​denoted as R. health The expression is: R after normalization health (i,j)∈[0,1], the larger the value, the tighter the coupling between parameters, R health Visualize the distribution of coupling strength between any two parameters; The mean of the off-diagonal elements of the normalized matrix is ​​taken to obtain the global coupling health score, which is then multiplied by 10 to convert it into a score of ten, thus obtaining the health status evaluation result. As an example, the global coupling health is denoted as H. corr The expression is: Among them, H corr ∈[0,1], the closer the value is to 1, the more stable the thermal-electric coupling relationship. Hcorr directly quantifies the overall health level of the thermal-electric coupling of the power controller.

[0091] The health status assessment results include an additional parameter pair coupling strength ranking table, which clearly marks the coupling strength ranking of each parameter pair, such as U and T1, T3 and T4, to help quickly locate the core coupling relationship.

[0092] In summary, this technical solution uses the correlation between thermal and electrical parameters as the core analytical indicator. By constructing a correlation matrix and performing threshold-free statistical operations, it achieves a quantitative evaluation of the high-order health status of the power controller. Without the need for preset thresholds, it can proactively detect hidden degradation signals such as decreased efficiency of the voltage regulator module and heat pipe cooling failure by dynamically changing the coupling strength. Through matrix operations, normalization processing, and core indicator extraction, the system health level is quantified by coupling strength. Simultaneously, using the normalized health correlation matrix and parameter pair ranking table, abnormal coupling points are accurately located. For example, a sudden drop in the coupling degree between U and T1 indicates a voltage regulator module failure, and a weakening coupling degree between T3 and T4 indicates an abnormal heat pipe cooling. Based on global coupling health, combined with stability, reliability, and other dimensional indicators, a comprehensive health assessment system encompassing coupling, stability, and reliability is constructed. This provides accurate data for in-depth health assessment and fault location of the power controller, improving the accuracy and foresight of operation and maintenance decisions.

[0093] And / or, (3) Ensemble learning model The ensemble learning models include single-machine health scores based on the XGBoost model and single-machine health scores based on the random forest model; ①XGBoost model Among them, the single-machine health score based on the XGBoost model includes: using the bus output voltage, power controller internal temperature, power controller temperature measurement and power controller heat pipe temperature measurement as input features, and using the XGBoost model to output the abnormal probability value; the health score is calculated according to the following formula: health score = (1 - abnormal probability value) × 10; XGBoost is an ensemble learning algorithm that is constructed serially from multiple decision trees. It boasts high training efficiency and strong generalization ability, making it suitable for accurate assessment of device health status. XGBoost selects the optimal splitting feature using a greedy algorithm, combined with regularization constraints to avoid overfitting. Each tree fits the residual of the previous tree, and the final output is the probability value of a sample belonging to an abnormal category, adapting to the nonlinear correlation features of multiple parameters in power controllers.

[0094] Among them, the bus output voltage reflects the stability of power supply, while the internal temperature of the power controller, the temperature measurement of the power controller, and the temperature measurement of the heat pipe of the power controller respectively reflect the thermal safety of the power module, the measurement point location, and the heat dissipation link.

[0095] The health score quantifies the overall health level of the power controller on a 10-point scale. A higher score indicates a lower degree of degradation in core modules such as the power supply system and thermal control system, and a better health status.

[0096] As an example, the anomaly probability value ranges from 0 to 1. The closer the probability value is to 1, the higher the probability that the sample belongs to an abnormal state; the closer it is to 0, the closer the operating state is to normal.

[0097] ② Random Forest Model The single-machine health score based on the random forest model includes: using bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input features, and using the random forest model to output anomaly probability values; the health score is calculated according to the following formula: Health score = (1 - anomaly probability value) × 10; and / or, Random forest is an ensemble learning algorithm composed of multiple decision trees. It has strong generalization ability and good resistance to noise interference, making it suitable for probabilistic assessment of equipment health status. Random forest constructs multiple independent decision trees through bootstrap sampling and random subspaces. During classification, the results are output by majority voting and converted into outlier probabilities, reducing the risk of overfitting from a single tree.

[0098] The anomaly probability value ranges from 0 to 1. The closer the probability value is to 1, the higher the probability that the sample belongs to an abnormal state; the closer it is to 0, the closer the operating state is to normal.

[0099] The health score quantifies the overall health level of the power controller on a 10-point scale. A higher score indicates a lower degree of degradation in core modules such as the power supply system and thermal control system, and a better health status.

[0100] Health status assessment also includes evaluation using unsupervised learning models, including single-machine health scores based on the local outlier factor algorithm, single-machine health scores based on the isolated forest model, and single-machine health scores based on a one-dimensional residual network. ③ Local Outlier Algorithm The single-machine health score based on the Local Outlier Factor (LOF) algorithm includes: using the bus output voltage and power controller internal temperature as input samples, calculating the LOF score using the LEF algorithm; determining the maximum value of the LOF score in the training set as the normal threshold; normalizing the LOF score of the current sample: if the LOF score ≤ the normal threshold, the normalized value = 1 - (LOF score / normal threshold); if the LOF score > the normal threshold, the normalized value = 0.5 - (LOF score - normal threshold) / (2 × normal threshold); and calculating the health score using the following formula: health score = normalized value × 10. Among them, LOF is an unsupervised outlier detection algorithm that requires no label data and can identify samples with abnormal local density, making it suitable for unsupervised assessment of equipment health status. LOF quantifies the degree of outlier status by calculating the ratio of the local density of a sample to its neighboring samples (LOF score). The higher the LOF score, the more the sample deviates from the normal local density, making it suitable for label-free monitoring scenarios of power controllers.

[0101] Among them, the LOF score indicates that the higher the score, the more the current operating state of the power controller deviates from the local density of normal samples; the lower the score, the closer the state is to normal.

[0102] Among them, the health score quantifies the health status of the power controller on a 10-point scale. The higher the score, the healthier the equipment is operating, providing a quantitative basis for operation and maintenance decisions in unsupervised scenarios.

[0103] ④Isolated Forest Model The single-machine health score based on the isolated forest model includes: using the bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input features, and using the isolated forest model to output the anomaly probability value; the health score is calculated as follows: health score = (1 - anomaly probability value) × 10; Among them, Isolation Forest is an unsupervised outlier detection algorithm with high detection efficiency and good adaptability to high-dimensional data, making it suitable for efficient assessment of equipment health status. Isolation Forest constructs multiple isolated trees by randomly partitioning the feature space. The shorter the path length of a sample in the tree, the easier it is to be isolated (i.e., the more abnormal), which is finally converted into anomaly probability values, adapting to the needs of efficient multi-parameter monitoring of power controllers.

[0104] The anomaly probability value ranges from 0 to 1. The closer the probability value is to 1, the higher the probability that the sample belongs to an abnormal state; the closer it is to 0, the closer the operating state is to normal.

[0105] The health score quantifies the overall health level of the power controller on a 10-point scale. A higher score indicates a lower degree of degradation in core modules such as the power supply system and thermal control system, and a better health status.

[0106] ⑤ One-dimensional residual network The single-machine health score based on a one-dimensional residual network includes: taking the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input, and using a one-dimensional residual network to output the health state probability; the health score is calculated as follows: health score = health state probability × 10.

[0107] Among them, the one-dimensional residual network (1D-ResNet) is a deep learning algorithm that can extract deep features of time-series data and alleviate gradient vanishing through residual connections, making it suitable for deep time-series assessment of device health status. 1D-ResNet constructs a deep network through residual blocks, directly learns the residuals between input and output, accurately captures the deep time-series features of power controller parameters, and outputs a health status probability value, adapting to its complex time-series pattern monitoring needs.

[0108] The anomaly probability value ranges from 0 to 1. The closer the probability value is to 1, the higher the probability that the sample belongs to an abnormal state; the closer it is to 0, the closer the operating state is to normal.

[0109] The health score quantifies the overall health level of the power controller on a 10-point scale. A higher score indicates a lower degree of degradation in core modules such as the power supply system and thermal control system, and a better health status.

[0110] (4) Integration of evaluation results As one possible implementation, health status assessment also includes fusing the evaluation results of electrical parameter deviation analysis methods, thermo-electric coupling analysis methods, ensemble learning models, and unsupervised learning models to obtain a comprehensive health score, specifically including: Health scores based on XGBoost, Random Forest, Local Outlier Factor, Isolated Forest and One-Dimensional Residual Network are extracted, and the minimum value is taken as the health score of the neural network class. The minimum value is denoted as H. minnet Taking the minimum value as the health score of the neural network class can capture the most significant health degradation risk and avoid underestimating the hidden dangers by a single model.

[0111] The comprehensive health score is obtained by arithmetically averaging the health score of the neural network class with the health score of the single-machine electrical parameters based on the voltage deviation rate and the health score of the thermo-electric coupling based on the thermo-electric parameter correlation matrix.

[0112] As an example, the overall health score is denoted as H; H = (Single machine electrical parameter health + Thermal-electric parameter coupling health + H) minnet ÷ 3, and round the result to one decimal place; As an example, the health status assessment results are determined according to the grading and scoring rules; 9.0≤H≤10, healthy; 8.0≤H<9.0, sub-healthy; 5.0≤H<8.0, indicating poor health; 0≤H<0.5, fault (extremely poor).

[0113] In summary, this technical solution integrates local core quantification of basic indicators, system-coupled evaluation of data features, and risk warning using multiple neural network models. It strengthens the warning of weak links by extracting the minimum value of the neural network evaluation result, and then fuses this result with the average of the evaluation results from the other two solutions. This fusion method does not rely on weight settings, comprehensively reflects the overall health level of the power controller, and effectively captures key hidden dangers, making the evaluation results both comprehensive and practical for engineering warnings.

[0114] (III) Integrating the evaluation results from various dimensions By integrating the evaluation results from various dimensions, a comprehensive evaluation index for the power controller is generated.

[0115] As one possible approach, the evaluation results from various dimensions are integrated to generate a comprehensive evaluation index for the power controller. Specifically, this includes mapping the evaluation results from each dimension to generate stability level, reliability level, performance level, and health level. Alternatively, the evaluation results from each dimension can be weighted and integrated to generate a comprehensive evaluation index for the power controller.

[0116] Secondly, embodiments of the present invention provide an on-orbit evaluation system for a power controller, comprising: a data acquisition module for acquiring telemetry data of multiple dimensions of the power controller during on-orbit operation, wherein the telemetry data includes at least bus output voltage, internal temperature of the power controller, temperature measurement of the power controller, temperature measurement of the heat pipe of the power controller, and temperature measurement of the heat dissipation surface of the power controller. The multi-dimensional evaluation module is used to evaluate the power controller from multiple dimensions based on telemetry data: operational stability evaluation, operational reliability evaluation, on-orbit performance evaluation, and health status evaluation. The integrated evaluation module is used to integrate the evaluation results from various dimensions and generate a comprehensive evaluation index for the power controller.

[0117] Thirdly, embodiments of the present invention provide a terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the on-orbit evaluation method for the power controller provided by the present invention.

[0118] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the description of the drawings, in carrying out the claimed invention. In this specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple components. A single processor or other unit can implement several of the functions listed in the specification. While certain measures are described in different embodiments, this does not mean that these measures cannot be combined to produce good results.

[0119] Although the invention has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made therein without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the invention and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if such modifications and modifications fall within the scope of the invention and its equivalents, the invention is also intended to include such modifications and modifications.

Claims

1. A method for on-orbit evaluation of a power controller, characterized in that, Includes the following steps: Acquire telemetry data from multiple dimensions of the power controller during its on-orbit operation. The telemetry data should include at least the bus output voltage, the internal temperature of the power controller, the temperature measurement of the power controller, the temperature measurement of the power controller heat pipe, and the temperature measurement of the power controller heat dissipation surface. Based on telemetry data, the power controller is evaluated from multiple dimensions, including at least: operational stability evaluation, operational reliability evaluation, on-orbit performance evaluation, and health status evaluation. Among them, operational stability evaluation is achieved based on fluctuation analysis or state classification model of telemetry data; operational reliability evaluation is achieved based on parameter compliance statistics or mutation detection; on-orbit performance evaluation is achieved based on time series prediction model or reconfiguration error analysis; and health status evaluation is achieved based on electrical parameter deviation analysis, thermo-electric coupling analysis or ensemble learning model. By integrating the evaluation results from various dimensions, a comprehensive evaluation index for the power controller is generated.

2. The on-orbit evaluation method for a power controller according to claim 1, characterized in that, The application of volatility analysis to obtain operational stability evaluation results specifically includes: Based on the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement, calculate the standard deviation of each parameter within the sliding window, and obtain the mean of the standard deviation of the multi-window. The single-parameter stability score is calculated based on the ratio of the mean standard deviation of each parameter to the preset stability threshold. The smaller the ratio, the higher the score, and a score of 0 is given when the ratio exceeds 1. The entropy weight method is used to calculate the weight based on the dispersion of the standard deviation of each parameter window; the greater the dispersion, the higher the weight. The stability evaluation result is obtained by weighted summation of the stability scores of each parameter; or, The application of volatility analysis to obtain stability evaluation results specifically includes: Based on the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement, and power controller heat dissipation surface temperature measurement, the mean and standard deviation of each parameter within the sliding window are calculated, and then the coefficient of variation CV is calculated as: standard deviation / mean × 100%. The mean of the coefficients of variation for each parameter across all windows is calculated to obtain the mean of the multi-window CV. The single-parameter stability evaluation result is determined by comparing the mean of the multi-window CV with the preset threshold and according to the graded scoring rules. The thresholds include normal threshold, attention threshold, warning threshold and emergency threshold. The mean value of the stability evaluation results for each parameter is taken as the stability evaluation result; and / or The stability evaluation results are obtained by applying a state classification model, specifically including: Using the time-series data of bus output voltage and power controller internal temperature as input, a support vector machine model is used to output the current state identification label, which is either normal or abnormal. Within a sliding window, count the number of normal status identification tags, the number of abnormal status identification tags, and the total number of tags; The stability score is calculated using the following formula: Score = max{0, (number of normal state identification tags within the window / total number of tags) × 10 - (number of abnormal state identification tags × 0.5)}.

3. The on-orbit evaluation method for a power controller according to claim 1, characterized in that, The application parameter compliance statistics used to obtain operational reliability evaluation results specifically include: For each sampling point of the bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement, determine whether it is within the preset safety threshold range. If it is, the state quantity is recorded as 0; otherwise, it is recorded as 1. Add anti-shake logic: The parameter is only confirmed to be out of limit when three consecutive sampling points are all recorded as 1; otherwise, it is corrected to 0. Count the number of sampling points with a state value of 0 and the total number of valid sampling points within a day, and calculate the normality rate of each parameter; The average of the normality rates of each parameter is taken to obtain the comprehensive compliance rate of the multiple parameters, which is the result of the work reliability evaluation; and / or, The application of mutation detection to obtain operational reliability evaluation results specifically includes: Calculate the first-order difference sequence for the time series data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement and power controller heat dissipation surface temperature measurement respectively. Points whose absolute difference exceeds the preset mutation threshold are identified as mutations, and the number of mutations and normal occurrences for each parameter within the period are counted. The reliability of each parameter is calculated using the following formula: Reliability = (Number of normal occurrences / Total number of occurrences) × 100%; The entropy weight method is used to calculate the weights based on the degree of dispersion of the difference sequences of each parameter; The reliability of each parameter is weighted and summed to obtain the comprehensive reliability of the multi-parameter parameters, which is the result of the operational reliability evaluation.

4. The on-orbit evaluation method for a power controller according to claim 1, characterized in that, The time-series prediction model includes performance change trend scores based on LSTM networks and performance change trend scores based on Bayesian networks. Specifically, the performance change trend score based on LSTM networks involves: using time-series data of bus output voltage and power controller internal temperature as input, and employing a Long Short-Term Memory (LSTM) network to output the anomaly probability value for each time step; calculating the first-order difference of the anomaly probability values ​​and counting the total number of first-order differences; summing the first-order differences and recording the sign of the result; counting the number of differences in the first-order difference sequence that have the same sign as the summation result and the number of different differences; and calculating the trend score using the following formula: Score = (Number of identical differences - Number of different differences) / Total number × 10. The performance trend score based on Bayesian networks includes: using time-series data of bus output voltage and power controller internal temperature as input, the Bayesian network outputs the normal pattern similarity at the current moment; calculating the first-order difference of the similarity and counting the total number of first-order differences; summing the first-order differences and recording the sign of the result; counting the number of differences in the first-order difference sequence that have the same sign as the summation result and the number of differences that are different; calculating the trend score using the following formula: Score = (Number of identical differences - Number of different differences) / Total number × 10; or, The application of reconstruction error analysis methods to obtain on-orbit performance evaluation results specifically includes: The timing data of the bus output voltage and the internal temperature of the power controller are used as inputs, and the reconstruction error is output by a deep self-encoder. Calculate the temporal rate of change of the reconstruction error, i.e., the first-order difference; The proportion of negative rates of change within the sliding window is used as the proportion of state improvement. The health trend index is calculated using the following formula: Health trend index = percentage of improvement in condition × (1 - normalized value of average reconstruction error); The health trend index is normalized and multiplied by 10 to obtain the performance degradation trend score.

5. The on-orbit evaluation method for a power controller according to claim 1, characterized in that, The application of electrical parameter deviation analysis to obtain health status evaluation results specifically includes: Obtain real-time data and rated voltage of the bus output voltage; Calculate the voltage deviation rate at each sampling point; The average deviation rate within the statistical data collection period; Based on the average deviation, the normalized bus output voltage health score is calculated, which is the health status evaluation result; and / or, The application of thermo-electric coupling analysis to obtain health status assessment results specifically includes: Acquire timing data of bus output voltage, power controller internal temperature, power controller temperature measurement, power controller heat pipe temperature measurement, and power controller heat dissipation surface temperature measurement. Calculate the Pearson correlation coefficient between any two parameters and construct the original correlation coefficient matrix; Take the absolute value of the matrix elements and normalize them to obtain a normalized health-related matrix; The mean of the off-diagonal elements of the normalized matrix is ​​taken to obtain the global coupling health score, which is then multiplied by 10 to convert it into a 10-point score, thus obtaining the health status evaluation result; and / or, The ensemble learning model includes single-machine health scores based on the XGBoost model and single-machine health scores based on the random forest model. Specifically, the single-machine health score based on the XGBoost model includes: using bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input features, and using the XGBoost model to output anomaly probability values; the health score is calculated as follows: Health score = (1 - anomaly probability value) × 10; The single-machine health score based on the random forest model includes: using bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input features, and using the random forest model to output anomaly probability values; the health score is calculated according to the following formula: Health score = (1 - anomaly probability value) × 10; and / or, The health status assessment also includes evaluation using unsupervised learning models. These unsupervised learning models include single-machine health scores based on the Local Outlier Factor (LOF) algorithm, single-machine health scores based on the Isolation Forest model, and single-machine health scores based on a one-dimensional residual network. Specifically, the single-machine health score based on the LEF algorithm involves: using the bus output voltage and power controller internal temperature as input samples, calculating the LOF score using the LEF algorithm; determining the maximum value of the LOF score in the training set as the normal threshold; normalizing the LOF score of the current sample: if the LOF score ≤ the normal threshold, the normalized value = 1 - (LOF score / normal threshold); if the LOF score > the normal threshold, the normalized value = 0.5 - (LOF score - normal threshold) / (2 × normal threshold); and calculating the health score using the following formula: Health score = normalized value × 10. The single-machine health score based on the isolated forest model includes: using the bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input features, and using the isolated forest model to output the anomaly probability value; the health score is calculated as follows: health score = (1 - anomaly probability value) × 10; The single-machine health score based on a one-dimensional residual network includes: taking the time-series data of bus output voltage, power controller internal temperature, power controller temperature measurement, and power controller heat pipe temperature measurement as input, and using a one-dimensional residual network to output the health state probability; the health score is calculated as follows: health score = health state probability × 10.

6. The on-orbit evaluation method for a power controller according to claim 5, characterized in that, Health status assessment also includes fusing the evaluation results from electrical parameter deviation analysis, thermo-electric coupling analysis, ensemble learning models, and unsupervised learning models to obtain a comprehensive health score, specifically including: Health scores based on XGBoost, Random Forest, Local Outlier Factor, Isolated Forest and One-Dimensional Residual Network are extracted, and the minimum value is taken as the health score of the neural network class. The comprehensive health score is obtained by arithmetically averaging the health score of the neural network class with the health score of the single-machine electrical parameters based on the voltage deviation rate and the health score of the thermo-electric coupling based on the thermo-electric parameter correlation matrix.

7. The on-orbit evaluation method for a power controller according to claim 1, characterized in that, By integrating the evaluation results from various dimensions, a comprehensive evaluation index for the power controller is generated, which specifically includes: mapping the evaluation results from each dimension to generate stability level, reliability level, performance level, and health level. Alternatively, the evaluation results from each dimension can be weighted and integrated to generate a comprehensive evaluation index for the power controller.

8. An on-orbit evaluation system for a power controller, characterized in that, include: The data acquisition module is used to acquire telemetry data from multiple dimensions of the power controller during its on-orbit operation. The telemetry data includes at least the bus output voltage, the internal temperature of the power controller, the temperature measurement of the power controller, the temperature measurement of the power controller heat pipe, and the temperature measurement of the power controller heat dissipation surface. The multi-dimensional evaluation module is used to evaluate the power controller from multiple dimensions based on telemetry data: operational stability evaluation, operational reliability evaluation, on-orbit performance evaluation, and health status evaluation. The integrated evaluation module is used to integrate the evaluation results from various dimensions and generate a comprehensive evaluation index for the power controller.

9. A terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the power controller on-orbit evaluation method as described in any one of claims 1 to 7.

Images