Power amplification system suitable for wide-temperature environment and stability control method

Abstract

The invention relates to the technical field of power amplification control, and discloses a power amplification system suitable for a wide-temperature environment and a stability control method, and the method comprises the steps: obtaining a real-time sampling sequence of an output signal of the power amplification system and current temperature data; recording a mass center position sequence in a plurality of continuous time windows, and fitting a direction vector of a mass center migration track; calculating the cosine similarity between the migration direction vector and the direction vector of each fault attractor, and predicting the predicted time when the mass center reaches each fault attractor based on the cosine similarity and the distance between the current mass center and each fault attractor; and selecting a main risk fault type according to the predicted time, outputting a fault evolution prediction result, and generating a control strategy enabling the phase space trajectory centroid to move towards the normal attractor direction. The fault evolution trend and occurrence time can be predicted in advance, and the stability of the power amplification system in the wide-temperature environment is improved.

Description

Technical Field

[0001] This invention relates to the field of power amplification control technology, and more specifically, to a power amplification system and stability control method adapted to a wide temperature range. Background Technology

[0002] In applications such as aerospace and polar exploration, power amplification systems need to operate stably in a wide temperature range. However, temperature variations can lead to performance degradation of power devices and abnormal sensor responses, ultimately causing system failures. These failures are usually not sudden but rather undergo a gradual evolution: the phase space trajectory of the system output signal gradually deviates from the normal operating state and migrates towards a fault state as temperature changes and the fault develops.

[0003] Existing phase space stability analysis methods primarily rely on static judgments based on the phase space characteristics at the current moment. They fail to utilize the correlation between temperature changes and phase space characteristics to predict the evolution trend of system stability and cannot estimate the time of failure in advance. Furthermore, existing fault location methods require waiting for multiple sensors to actually generate abnormal signals before analyzing the fault propagation path, making it impossible to proactively predict the original fault source in the early stages of fault evolution.

[0004] This prevents the system from taking timely preventive control measures, affecting the stable operation of the system in a wide temperature range. There is a technical problem that the power amplification system cannot predict the fault evolution trend and estimate the fault occurrence time in advance under temperature change conditions. Summary of the Invention

[0005] This invention provides a power amplification system and stability control method adapted to a wide temperature environment, solving the technical problem in related technologies that power amplification systems cannot predict the fault evolution trend and estimate the fault occurrence time in advance under temperature change environments.

[0006] This invention provides a power amplifier stability control method adaptable to a wide temperature range, comprising:

[0007] Acquire the real-time sampling sequence of the power amplifier system's output signal and the current temperature data;

[0008] The output signal is reconstructed in phase space using a time delay embedding method to generate a phase space vector sequence and calculate the centroid of the phase space trajectory within the current time window. The attractor center under normal operating conditions and the attractor center corresponding to each typical fault mode are obtained from the historical database.

[0009] Record the centroid position sequence within multiple consecutive time windows, and fit the direction vector of the centroid migration trajectory; the cosine similarity between the migration direction vector and the direction vector of each fault attractor is obtained by dividing the inner product of the migration direction vector and the direction vector of the fault attractor by the product of the norms of the two vectors; based on the cosine similarity and the distance between the current centroid and each fault attractor, predict the estimated time for the centroid to reach each fault attractor.

[0010] Based on the expected time, the primary risk fault type is selected, the fault evolution prediction result is output, and a control strategy is generated to move the centroid of the phase space trajectory towards the direction of the normal attractor.

[0011] Furthermore, the estimated time is obtained by dividing the distance between the current centroid and the fault attractor by the product of the migration velocity and the absolute value of the cosine similarity, wherein the migration velocity is the norm of the migration direction vector.

[0012] Furthermore, the phase space reconstruction of the output signal using the time delay embedding method includes:

[0013] The output signal sampling sequence is reconstructed using delay coordinates according to the time delay parameter and embedding dimension, generating a phase space vector sequence composed of the current sample value and its multiple delayed sample values.

[0014] Further, the direction vector of the fitted centroid migration trajectory includes:

[0015] The least squares method is used to perform linear fitting on the centroid position sequence within multiple consecutive time windows to obtain the direction vector representing the centroid movement trend.

[0016] Furthermore, it also includes:

[0017] Obtain the rate of change of the current temperature; predict the future temperature based on the current temperature and the rate of change of temperature.

[0018] Query the expected phase space feature corresponding to the future temperature from the pre-stored temperature-phase space feature mapping table;

[0019] Calculate the temperature correlation deviation vector between the current centroid and the desired phase space feature;

[0020] The direction of the temperature-related deviation vector is fused with the migration direction vector. When both point to the same fault attractor direction, the prediction confidence of this fault type is increased.

[0021] Furthermore, increasing the prediction confidence of this fault type includes:

[0022] Calculate the cosine of the angle between the temperature-related deviation vector and the migration direction vector;

[0023] The adjusted prediction confidence is obtained by multiplying the base confidence by the temperature correlation factor determined by the cosine of the included angle and the temperature correlation weighting coefficient.

[0024] Furthermore, it also includes:

[0025] Obtain the Mahalanobis distance sequence of multiple sensor channels distributed at various locations in the system. The Mahalanobis distance represents the degree of deviation of the output characteristics of each sensor channel from the normal state distribution.

[0026] Calculate the anomaly trend of Mahalanobis distance for each sensor channel;

[0027] Obtain the pre-stored system topology correlation matrix, which represents the physical correlation strength between each sensor;

[0028] Based on the aforementioned abnormal trends and topological correlation matrix, Bayesian inference is used to calculate the probability that each sensor location is a potential source of failure.

[0029] Furthermore, the calculation of the probability of each sensor location being a potential fault source using Bayesian inference includes:

[0030] The prior probability of each sensor location as a fault source is determined based on the historical fault statistics of each location in the system topology.

[0031] Calculate the likelihood probability of observing the current abnormal trend distribution under the condition that a fault occurs at each sensor location based on the topological correlation matrix and the law of physical propagation.

[0032] The likelihood probability is multiplied by the prior probability and normalized to obtain the posterior probability of each sensor location as a potential fault source.

[0033] Furthermore, it also includes:

[0034] The predicted fault type is associated and matched with the probability of sensor fault source. Based on the fault mode-location association knowledge, the set of locations where the fault type is most likely to occur is selected, and the most likely fault type-fault source combination is determined.

[0035] When outputting the fault evolution prediction results, the location information of the most likely original fault source is also output.

[0036] This invention provides a power amplifier stability control system adapted to a wide temperature range, comprising:

[0037] The data acquisition module is used to acquire the real-time sampling sequence of the power amplifier system's output signal and the current temperature data;

[0038] The phase space reconstruction module is used to reconstruct the phase space of the output signal using the time delay embedding method, generate a phase space vector sequence, calculate the centroid of the phase space trajectory within the current time window, and obtain the attractor center under normal operating conditions and the attractor center corresponding to each typical fault mode from the historical database.

[0039] The migration trajectory analysis module is used to record the centroid position sequence within multiple consecutive time windows, fit the direction vector of the centroid migration trajectory, calculate the cosine similarity between the migration direction vector and the direction vector of each fault attractor, and predict the estimated time for the centroid to reach each fault attractor based on the cosine similarity and the distance between the current centroid and each fault attractor.

[0040] The control strategy output module is used to select the primary risk fault type based on the estimated time, output the fault evolution prediction result, and generate a control strategy that moves the centroid of the phase space trajectory toward the direction of the normal attractor.

[0041] The beneficial effects of this invention are as follows:

[0042] This invention overcomes the limitations of existing methods that rely solely on static judgments based on current phase space characteristics by tracking the migration trajectories of phase space attractors and calculating the consistency of migration speed and direction from the centroid to each fault attractor. This enables dynamic prediction of fault evolution trends and allows for advance estimation of fault occurrence time windows. Furthermore, by introducing a temperature-phase space correlation mapping, it uses the rate of temperature change to predict the expected phase space characteristics corresponding to future temperatures and integrates temperature correlation deviations with attractor migration directions. This overcomes the shortcomings of existing methods that fail to utilize the influence of temperature changes on phase space characteristic correlations, thus improving the accuracy of fault evolution trend prediction under temperature-changing environments.

[0043] Furthermore, by performing Bayesian inference based on the system topological correlation matrix and the anomaly trend of the multi-sensor Mahalanobis distance, potential fault sources can be inferred from the nascent state of the anomaly trend and the physical propagation law before all sensors have fully displayed anomaly signals. This overcomes the lag in existing methods that require waiting for multiple sensors to actually exhibit anomalies before locating the fault source, achieving early and forward-looking prediction of the fault source. In summary, this invention integrates three methods: temperature correlation prediction, attractor migration analysis, and multi-sensor temporal inference. It can simultaneously provide comprehensive predictive information on fault modes, occurrence time, evolution rate, and the location of the original fault source, providing sufficient decision support for the system to take preventative control measures. It solves the technical problem that power amplifier systems cannot predict fault evolution trends and estimate fault occurrence time in advance under temperature variation environments, achieving the technical effect of improving the stability of power amplifier systems under wide temperature environments. Attached Figure Description

[0044] Figure 1This is a flowchart of the power amplification stability control method adapted to a wide temperature environment according to the present invention;

[0045] Figure 2 This is a line graph illustrating the evolution trend of the multi-sensor Mahalanobis distance according to the present invention;

[0046] Figure 3 This is a bar graph showing the relationship between temperature change and system output power according to the present invention;

[0047] Figure 4 This is a diagram showing the sensor topology network relationship of the present invention;

[0048] Figure 5 This is a comparison of the confidence levels of fault evolution prediction in this invention. Detailed Implementation

[0049] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0050] At least one embodiment of the present invention discloses a power amplification stability control method adapted to a wide temperature range, such as... Figure 1 As shown, it includes the following steps:

[0051] Step 100: Obtain system operation data and generate a multi-source monitoring data set.

[0052] Acquire the real-time sampling sequence of the output signal of the power amplifier system ,in For the first The output signal value at each sampling time. This represents the number of sampling points. The number of sampling points should meet the following constraints: It should be large enough to cover the complete failure evolution process, it is recommended This ensures that sufficient phase space reconstruction and centroid calculation can be performed; the time interval between adjacent sampling times. The sampling frequency should be kept constant to ensure the effectiveness of the time delay embedding; the sampling frequency should be selected based on the Nyquist frequency of the system, and should be at least twice the highest frequency component of the system.

[0053] Simultaneously acquire current temperature data and the rate of temperature change The temperature parameters should meet the following constraints: temperature value It should be within the system's designed operating temperature range, typically... It is recommended to record the system's nominal operating temperature range and temperature change rate. It reflects the rate of temperature change, with the unit being the change in temperature per unit time; when When the temperature remains stable, Time indicates a rise in temperature. The time indicates a decrease in temperature; the rate of temperature change can be obtained by calculating the temperature difference between two consecutive sampling times: .

[0054] In addition, data distributed across various locations within the system are obtained. Mahalanobis distance sequence of each sensor channel ,in Indicates the first The output characteristics of each sensor channel are represented by the Mahalanobis distance relative to the normal state distribution. The Mahalanobis distance should satisfy the following constraint: the Mahalanobis distance ranges from... Where 0 indicates that the feature perfectly conforms to the normal distribution; under normal operating conditions, the Mahalanobis distance should be close to 0 (usually within 0). (within the range), when Significantly increased (e.g., exceeding a preset abnormal threshold) When the value is 3 to 5, it indicates that the sensor channel is abnormal; the sensitivity of Mahalanobis distance depends on the covariance matrix used for its calculation. The covariance matrix should be calculated based on sufficient historical data of normal operating conditions, and it is recommended that the number of samples be no less than 10 times the feature dimension.

[0055] Furthermore, the method for calculating the Mahalanobis distance is as follows: for the th Statistical analysis was performed on historical feature data collected by each sensor under normal operating conditions to calculate the mean value of the sensor's feature vector under normal operating conditions. Covariance Matrix ,in For the first The feature dimensions extracted by each sensor. Mean. Covariance Matrix The following constraints should be met: mean vector The arithmetic mean of historical feature data: ,in For the number of historical samples, Index of historical samples; covariance matrix It should be a symmetric positive definite matrix. The element is the first and the The covariance of dimensional features; the covariance matrix should be invertible, i.e. This is to compute its inverse matrix; to avoid matrix singularity caused by collinearity, a regularization term can be added: ,in The regularization coefficient (usually taken as...) arrive The trace of the covariance matrix is ​​multiplied by a factor of 1 to improve numerical stability.

[0056] Then, at the current moment, the real-time feature vector of the sensor is... Substitute into the Mahalanobis distance formula Perform calculations, where This represents transpose, and the result is... The value should be close to zero under normal operating conditions (usually within). (within the range), when A significant increase (exceeding the abnormal threshold) indicates an abnormality in the sensor channel.

[0057] Figure 2 The evolution trend of the Mahalanobis distance of the five sensor channels over time is shown.

[0058] Figure 3 This demonstrates the correlation between the rapid temperature rise and the decrease in system output power.

[0059] Figure 4 The physical topology of the system's various sensors is shown.

[0060] Step 200: Reconstruct the phase space based on the output signal and calculate the centroid of the phase space trajectory.

[0061] First, sample the output signal sequence. Z-score normalization is performed to eliminate the influence of signals of different magnitudes on phase space reconstruction. The normalized sampled values ​​should meet the following requirements. The mean is 0 and the standard deviation is 1. A time-delay embedding algorithm is used to reconstruct the phase space of the standardized output signal. The time-delay embedding algorithm uses the standardized sampling sequence... Time delay parameters and embedding dimension As input, a sequence of phase space vectors is generated. As output.

[0062] phase space vector The following constraints must be met: the vector dimension is... Each component is a standardized sample value; vector construction requires historical time data, i.e., time intervals. The vector at that location needs to be at least The sampling data at time t, therefore only when Only when the time is right can a valid phase space vector be calculated; the total number of valid phase space vectors is .

[0063] Calculate the centroid of the phase space trajectory within the current time window. ,in The time window length, This is the index of the moment within the time window. Centroid vector. The following constraints must be met: each component of the centroid is located in the corresponding dimension. The arithmetic mean of the vector components; the range of the centroid value is usually related to... The distribution is related, meaning that each component should be within a reasonable standardized range (usually 1). (Multiple standard deviations); a vector sequence used to calculate the centroid. Each vector in the vector must be a valid phase space vector.

[0064] Furthermore, the time delay parameter According to the sampling interval The selection of system dynamics characteristics ensures that the sampled values ​​at the delayed time have an appropriate correlation with the sampled values ​​at the current time, generally within a range of [value range missing]. Each sampling period; embedding dimension It should be greater than or equal to 2, and not exceed 2. This ensures that the reconstructed phase space vectors have sufficient dimension to represent the dynamic characteristics of the system; time window length It is recommended to select a vector that allows the centroid calculation to include a sufficient number of phase space vector samples. The range of values ​​is That is, the window contains at least The position of the center of mass at a given moment.

[0065] The aforementioned time delay embedding algorithm combines the values ​​of a one-dimensional time series signal under multiple time delays into a vector, mapping the time series to a high-dimensional phase space, thereby revealing the dynamic characteristics of the system.

[0066] Retrieve the attraction subcenter under normal operating conditions from the historical database. The attractor center represents the location where the phase space trajectories converge during normal system operation. Simultaneously, the attractor centers corresponding to each typical fault mode are acquired. ,in Given the number of known failure modes, For the first Attractor centers for various failure modes.

[0067] Attractor Center and The following constraint must be satisfied: Dimensional consistency: the dimension of all attractor centers is 0. The dimensions are the same as those of the phase space vector and the centroid; the dissimilarity condition: the attractor centers of different failure modes should satisfy... Furthermore, the distance between the normal attractor and the faulty attractor should satisfy the following conditions: To ensure that different working conditions can be effectively distinguished, among which Minimum separation distance; physical rationality: the component values ​​of each attractor center should be within a reasonable range, that is, each component is usually within... Within the range, The standard deviation of the Z-score after standardization (value is 1).

[0068] Furthermore, the minimum separation distance The method for determining the distances is as follows: First, calculate the pairwise Euclidean distances between all established attractor centers to obtain the distance set. ,in , and Index the attractors for different operating conditions; then calculate the average of this distance set as the typical distance between attractors. ,in This represents the number of attractor pairs; the recommended minimum separation distance is [value missing]. This is to ensure that there is sufficient differentiation between different attractors.

[0069] Furthermore, the method for establishing the attractor center is as follows: During the system design and testing phases, a large number of output signal samples are collected from the system under normal operating conditions and various typical fault modes. For each operating condition, phase space reconstruction using a time delay embedding algorithm is performed to generate a set of phase space trajectory points under that operating condition. Then, the geometric center of these phase space trajectory points is calculated as the attractor center for this working condition: ,in This represents the total number of phase space trajectory points under this operating condition. These are the indexes for the trajectory points. These attractor centers are pre-calculated and stored in a historical database before the system runs, serving as a reference standard during system operation.

[0070] Step 300: Track the migration trajectory of the centroid of the phase space trajectory to predict the fault evolution trend and arrival time.

[0071] Recording Continuous Centroid position sequence within a time window The direction vector of the centroid migration trajectory is fitted using the least squares method, and the least squares method uses the centroid position sequence. As input, the output is a direction vector representing the trajectory migration trend. The magnitude of the direction vector This represents the migration velocity of the centroid per unit time. Calculate the direction vector from the current centroid position to each fault attractor. .

[0072] Furthermore, the specific calculation method for fitting the centroid migration direction using the least squares method is as follows: The centroid position sequence... Using the time window index as the independent variable and the various dimensional components of the centroid position as the dependent variable, a linear fit is performed on each dimension to obtain the rate of change for each dimension; specifically, for the ... Dimensions Establish a linear relationship ,in Relative time index, The rate of change in this dimension. The intercept is obtained by minimizing the sum of squared residuals. Solve for the rate of change of each dimension Finally, the migration direction vector is obtained. .

[0073] Furthermore, the least squares method is used to solve for the rate of change. and intercept The specific formula is: for optimizing the objective function To each and Taking the partial derivatives and setting them to zero, we obtain the normal system of equations: and ;

[0074] Solving this system of equations yields the rate of change. and intercept The time index The summation terms can be calculated in advance: and .

[0075] The cosine similarity algorithm is used to calculate the similarity between the migration direction and the directions of each fault attractor. The cosine similarity algorithm uses the migration direction vector as the basis for the similarity calculation. and fault attractor direction vector As input, output cosine similarity. ,in Represents the dot product of vectors. Represents the vector norm.

[0076] Migration direction vector The following constraints must be satisfied: the vector norm satisfies That is, the vector magnitude is a positive real number, representing the velocity of the centroid's migration per unit time; the vector direction is... The unit vector indicates the direction of centroid migration; when the vector magnitude is close to zero (e.g. ,in The characteristic length of the phase space can be taken as the typical distance between attractors. When the fault attractor direction vector is in the first place, subsequent cosine similarity calculations should be skipped to avoid numerical instability; The module length must meet the following requirements. If the centroid has reached the fault attractor (i.e. ,in Recommended value If the fault mode is not specified, then that fault mode will not be included in the prediction. Cosine similarity. When the value is close to 1, it indicates that the phase space trajectory is more inclined to this fault mode.

[0077] Calculate the distance between the current centroid and each fault attractor. ,in The distance is dimensionless and standardized. Based on cosine similarity and current distance, the estimated time for the centroid to reach each fault attractor is predicted. The estimated time reflects the system's evolution to the [number]th [stage]. The time required for each fault state.

[0078] Furthermore, the estimated arrival time The calculation should satisfy the following constraints: only when The calculation is performed in real time, meaning that only fault modes where the angle between the migration direction and the fault direction is less than 90 degrees have a finite estimated arrival time; when When this occurs, it indicates that the direction of the centroid migration is contrary to the fault attractor, and the estimated time should be set to infinity or marked as "direction mismatch - will not reach"; when Close to zero (e.g., less than a preset distance threshold) The recommended value is When the time reaches 0, it indicates that the centroid is close to the fault attractor, and the estimated time should be set to 0 or marked as "Approaching Target"; estimated time The unit of measurement is the number of time windows, and the actual physical time is... ,in The duration of a single time window.

[0079] Furthermore, the number of time windows It should be at least 2 to ensure that least squares fitting can be performed; this is generally recommended. To obtain a more stable estimate of the migration direction. When the calculated migration velocity... Approaching zero (e.g., less than a preset speed threshold) The recommended value is When the value is zero, it indicates that the system's centroid is in a relatively static state within the current observation window. In this case, the predicted fault arrival time should be set to approach infinity or marked as "static state - unpredictable" to avoid computational instability caused by the denominator approaching zero.

[0080] Figure 5 The prediction confidence levels for different failure modes are compared.

[0081] Step 400: Output the fault evolution prediction results and stability control strategy.

[0082] The failure mode with the shortest expected arrival time is selected as the primary risk. The failure evolution prediction results are output, including the predicted failure mode and its expected occurrence time. Migration speed A control strategy is generated based on the prediction results, wherein the direction of control parameter adjustment is achieved by moving the target's centroid in phase space in the specified direction. Decoding is performed to obtain: based on the projection components of the target direction vector onto the dimensions of each key control parameter of the system, the direction and magnitude of adjustment for each control parameter are determined, so that by adjusting these parameters, the centroid of the phase space trajectory of the system output signal can be directed towards the normal attractor. Directional movement.

[0083] Furthermore, the target direction vector in phase space The decoding method for the actual control parameters is as follows: through the output signal Related to key system control parameters (such as the gain of the power amplifier) Operating point voltage Establish a parameter sensitivity matrix based on the quantization relationship between (etc.) ,in Embedding dimension of phase space, To control the number of parameters, The row and column elements Representing the One control parameter For the first The partial derivative influence coefficients of each phase space dimension. Since different control parameters have different dimensions and numerical ranges, the parameter sensitivity matrix needs to be normalized: divide each column of the sensitivity matrix by the operating range of the parameter in that column. The normalized sensitivity matrix is ​​obtained. This eliminates the impact of dimensional differences on the optimization calculation. Then, based on the target direction... The normalized adjustment of each control parameter is obtained by solving the normalized parameter sensitivity matrix using the least squares method or the pseudo-inverse matrix method. ,in for The pseudo-inverse matrix is ​​obtained, and finally the normalized adjustment is denormalized to obtain the actual adjustment. .

[0084] Furthermore, the parameter sensitivity matrix The establishment method is as follows: during the system testing phase, for each control parameter Within its operating range, several test points are selected, and the system output signals are collected at different parameter values ​​while keeping other parameters constant. Phase space reconstruction and centroid calculation are performed on each set of output signals to obtain the centroid positions for different parameter values. Then, the... Given a parameter, calculate the numerical derivative of each dimension component of the centroid with respect to that parameter: ,in For small changes in parameters, it is recommended to take a value of 1% to 5% of the parameter's operating range; by calculating the partial derivatives for all control parameters, a complete parameter sensitivity matrix can be finally established. And it is stored in the system for use at runtime.

[0085] Parameter adjustment amount The following constraints must be met: the adjustment amount of each parameter must be within the allowable range, i.e. ,in and For the first The physical upper and lower limits of each parameter are usually determined by hardware constraints (such as the maximum gain and maximum output power of a power amplifier); the parameter adjustment range should meet the stability requirements, i.e. ,in The maximum adjustment range for a single step is recommended to be between 5% and 20% of the working range of each parameter to avoid control instability caused by over-adjustment.

[0086] When pseudo-inverse computation When the problem exceeds the constraints, the constrained least squares method should be used to solve it again, with the optimization objective being... Then, inverse normalization is performed to obtain the actual adjustment amount; finally, the calculated parameter adjustment amount is... Applying this to the current operating point yields new control parameter setpoints. .

[0087] It should be noted that the cosine similarity calculation in step 300 above refers to normalizing the result after performing an inner product operation between the currently fitted centroid migration direction vector and the direction vectors of each pre-stored fault attractor, thereby quantifying the directional consistency between the migration trend and various fault modes. The cosine similarity value range is... When the value is close to 1, it means that the migration direction is highly consistent with the fault direction; when the value is close to 0, it means that the two are orthogonal; when the value is close to -1, it means that the two are in opposite directions.

[0088] It should be noted that the above estimated arrival time The calculation is performed in the normalized phase space. Since all phase space vectors in step 200 have been Z-score normalized, the dimensions of each dimension in the phase space are unified, therefore the distance... Migration speed All distances are dimensionless standardized distances, and the centroid migration velocity represents the standardized displacement of the phase space trajectory per unit time. The estimated time is calculated assuming the centroid migrates at the current velocity. Assuming the system continues to move along the current migration direction, the theoretical arrival time is obtained by dividing the standardized distance by the effective velocity component (i.e., the projected velocity along the fault direction). This time is the estimated time window for the system to evolve to the corresponding fault state.

[0089] In this embodiment of the application, in order to improve the accuracy of fault evolution trend prediction, when obtaining historical database information in step 200, the method further includes obtaining temperature-phase space feature mapping tables for different temperature ranges. The temperature-phase space feature mapping table records the mapping relationship between temperature values ​​and corresponding phase space feature parameters. When the mapping table is established, for each temperature range, its corresponding phase space feature parameter has been standardized, having the same dimensions as the phase space vector in step 200 (i.e., standardized dimensionless quantity). Before performing migration trajectory analysis in step 300, the following steps are added:

[0090] Temperature-Phase Space Characteristic Mapping Table The following constraints must be met: the domain of the mapping table is the temperature range. The range is a phase space characteristic. Number of temperature ranges Choose a reasonable size; it is recommended. Too few features will result in insufficient feature representation, while too many will increase the computational burden; features for each temperature range Calculations should be based on sufficient historical data, and it is recommended that each temperature range contain at least 50 to 100 phase space trajectory points; all feature values ​​in the mapping table should be Z-score standardized and dimensionless.

[0091] Furthermore, the temperature-phase space characteristic mapping table The method for establishing it is as follows:

[0092] In the system's historical operating data, the entire temperature range (e.g., from...) arrive ) divided into Equally spaced temperature intervals, for example ,in These are the boundary values ​​of the temperature range. This is the lowest temperature boundary. The highest temperature boundary, the width of the temperature range is For each temperature range The output signals collected within this temperature range are selected from historical data, and phase space reconstruction is performed using a time-delay embedding algorithm. The center positions of all phase space trajectory points within this temperature range are calculated as the desired phase space features at this temperature. ,in This represents the number of phase space trajectory points within this temperature range. This is the index of the phase space trajectory points. The lookup method for the mapping table is: given the predicted temperature... First, determine the temperature range to which it belongs. Then, query the expected phase space features corresponding to that interval from the mapping table. If the predicted temperature falls exactly on the boundary between two intervals, then a linear interpolation method is used to calculate the expected characteristics. This is to improve the accuracy of temperature feature mapping.

[0093] Based on the current temperature and rate of temperature change Predicting future temperatures ,in For predicting the time step. For predicting the temperature. The following constraint must be met: If the predicted temperature exceeds the temperature range of the established mapping table, i.e. or In this case, boundary value processing should be used, that is... This is to avoid unreasonable extrapolations that exceed the known range.

[0094] In the mapping table Query predict temperature Corresponding expected phase space features (The desired phase space feature is already in a standardized form). Calculate the temperature correlation deviation vector between the current centroid and the desired feature at the predicted temperature. .

[0095] Temperature correlation deviation vector The following constraints must be met: the vector dimension is... The dimensions should be consistent with those of the phase space vector; the values ​​of each component of the vector should be within a reasonable standardized range, typically within... Between two standard deviations, exceeding the range indicates an excessive difference between the centroid and the expected feature; vector magnitude This reflects the phase space characteristic shift distance caused by temperature changes, where For the dimension index of the eigenvectors of the phase space; if Close to zero (e.g., less than a preset threshold) The recommended value is This indicates that the current centroid is very close to the expected characteristics at that temperature, and the temperature information has little guiding significance.

[0096] Furthermore, predict the time step. The selection should be based on the system's temperature change rate and fault evolution timescale; generally, it is recommended... exist Within a certain range, that is, at least half the duration of a time window and no more than the total duration of all observation windows, to ensure that temperature prediction is within a reasonable time range.

[0097] During the migration trajectory analysis in step 300, the temperature-related deviation vector is... Direction and attractor migration direction Perform fusion analysis:

[0098] Calculate the cosine of the angle between the two. If the cosine of the included angle is greater than the direction consistency threshold (It is recommended to set the directional consistency threshold to...) Within the specified range, a value of 0.707 generally corresponds to a 45-degree angle, indicating that both point in similar directions. This determines that the temperature change trend and the trajectory migration trend are consistent. For cases where both the migration direction and the temperature deviation direction point to a fault attractor, the prediction confidence of that fault mode is increased. ,in The base confidence level is calculated based on the migration direction. Temperature-related weighting coefficients (value range: (This is used to control the influence of temperature on the confidence level). When outputting the prediction results in step 400, the confidence level of each fault mode is also output, and the fault mode with the highest confidence level and the shortest expected arrival time is selected as the main risk.

[0099] Confidence The following constraints must be met: the base confidence level must be satisfied. and ,in For indexing failure modes, ensure that the base confidence of all failure modes follows a probability distribution; temperature-related weighting coefficients. ,when Temperature information does not affect confidence level when Temperature information completely affects the confidence level; a value of 0.3 to 0.7 is recommended. (The corrected confidence level is...) Should meet (When the cosine of the angle between the temperature direction and the migration direction is in) (When within the range), normalization should be performed again after correction so that... Finally, the confidence level that satisfies the following criteria is selected. Failure modes are used as the primary risk prediction targets.

[0100] Furthermore, the aforementioned basic confidence level The result is obtained by normalizing the mapping between the migration direction and the cosine similarity of the directions of each fault attractor. The specific calculation method is as follows: ,in The index is used to determine the failure modes. It ensures that the sum of the base confidence of all failure modes is 1, and only failure modes with a cosine similarity greater than zero (i.e., the angle between the migration direction and the failure attractor direction is less than 90 degrees) have a non-zero base confidence.

[0101] In this embodiment of the application, in order to achieve early location of the fault source, when acquiring system operation data in step 100, the method further includes acquiring a pre-stored system topology correlation matrix. The elements of the system's topological correlation matrix Indicates sensor With sensors The strength of the physical correlation between them. While performing fault evolution prediction in step 300, the following steps are added:

[0102] Furthermore, the topological correlation matrix The following constraints must be met: Each element Indicates sensor With sensors The strength of the physical correlation between them; the matrix should be a symmetric matrix or satisfy... , indicating the symmetry of physical relationships; diagonal elements The weight representing the sensor itself is generally set to the same order of magnitude as other non-zero elements; the elements in each row should be normalized so that the sum of the elements in each row is 1, i.e. ,in For the sensor index, representing the index from the sensor Distribution of fault propagation intensity at the origin.

[0103] Calculate the Mahalanobis distance anomaly trend for each sensor channel. The anomaly trend of Mahalanobis distance reflects the rate of change of the output characteristics of each sensor from the normal state. The anomaly trend of Mahalanobis distance can be obtained by calculating the change in Mahalanobis distance over two consecutive time windows: ,in and These are the Mahalanobis distance values ​​for the current and previous time windows, respectively. This is the time window interval. When... When, it indicates the first The anomaly level of each sensor channel is increasing; when When, it indicates that the degree of abnormality is decreasing; when When the value is close to zero, it indicates that the abnormal trend is relatively stable.

[0104] Mahalanobis distance anomaly trend The constraints are: the dimension of the Mahalanobis distance anomaly trend is the change in Mahalanobis distance per unit time; to reduce the influence of noise, the trend of multiple time windows can be smoothed, for example, by using a moving average. ,in To smooth window size, The time window index within the smoothing window is generally recommended to be between 2 and 5; the anomaly trigger threshold. It should be set according to the trend statistics under normal operating conditions. It is recommended to take a value that is 2 to 3 times the standard deviation of the rate of change of Mahalanobis distance under normal operating conditions.

[0105] The probability of each sensor location being a potential fault source is calculated using a Bayesian inference algorithm. The Bayesian inference algorithm is based on the abnormal trends of each sensor. Topological Independence Matrix The system takes historical fault statistics as input and outputs the posterior probability of the fault source at each sensor location. .

[0106] Prior probability The following constraints must be met: Probability range: Completeness requirements: ,in The total number of sensors, This serves as an index for the sensors, ensuring that the prior probabilities of all locations form a probability distribution.

[0107] Furthermore, prior probability The specific method for obtaining this information is as follows: The locations of each sensor are statistically analyzed from the system's long-term fault history records. Number of failure events as the original source of failure (i.e., the location where the failure first occurred) Calculate the relative frequency of this location as a fault source: ,in The denominator is the index of the sensor location. The total number of fault events across all sensor locations; if historical fault data is insufficient for some locations (e.g.) Less than the preset minimum number of samples (A value of 10 to 50 is recommended). Laplace smoothing or a uniform distribution can be used as the initial prior. (Laplace smoothing, where) This is a smoothing parameter, typically set to 1 or... (Uniformly distributed).

[0108] Likelihood probability The following constraints must be met:

[0109] Probability range: The conditional probability of observing the current abnormal trend pattern when a fault occurs at a given location should be modeled comprehensively using system topology correlation matrix, anomaly propagation delay, probability function, etc.

[0110] The likelihood probability is calculated as follows: if If it is a source of failure, then the sensors connected to it through the topological correlation matrix are... (Right now A significant abnormal growth trend should be observed, and an abnormality trigger threshold should be set. Then the likelihood probability is ,in For the sensor index, Let be a probability function that depends on propagation delay and topological strength, specifically, if the ... Each sensor satisfies and Relatively large, then A larger value indicates a larger value; conversely, a smaller value indicates a smaller value.

[0111] Furthermore, the abnormal trigger threshold The value should be set according to the fluctuation range of the Mahalanobis distance under normal operating conditions. It is recommended to take a value that is 2 to 3 times the standard deviation of the rate of change of the Mahalanobis distance under normal operating conditions to avoid normal fluctuations being misjudged as abnormal; the probability function It can be expressed in the form of a Gaussian kernel function. ,in Represents an exponential function. For the kernel width parameter, it is generally recommended to set it to a value of [value to be filled in]. 0.5 times, used to control the steepness of the function's response curve.

[0112] Posterior probability The following constraints must be met: Probability range: ;

[0113] Completeness requirements: The posterior probability should be normalized using Bayes' theorem. ,in For the index of sensor locations, ensure that the sum of the posterior probabilities of all locations is 1; the fault source confidence should be selected from the location with the highest posterior probability, i.e. .

[0114] After calculating the predicted fault mode in step 300, the fault mode is associated and matched with the sensor fault source probability: for each predicted fault mode, based on known fault mode-location association knowledge, the set of locations where the fault mode is most likely to occur is selected, and the posterior probability of the sensor fault source corresponding to these locations is normalized and compared to determine the most likely fault mode-fault source combination. Specifically, for the predicted fault mode... Let the set of possible locations be . (in This represents the number of locations where this failure mode may occur. (where the sensor location is where this fault mode may occur), then the conditional probability within this fault mode is: ,in The location index for the possible location of this failure mode is selected as the most likely failure source under this failure mode.

[0115] When outputting the prediction result in step 400, the system should also output the most likely original fault source location information and its posterior probability, so that the system can specifically check the original fault source location or take local control measures. The output format should include: predicted fault mode, estimated arrival time, confidence level, most likely fault source location, and fault source posterior probability.

[0116] This implementation overcomes the limitations of existing methods that rely solely on static judgments based on current phase space characteristics by tracking the migration trajectories of phase space attractors and calculating the consistency of migration speed and direction from the centroid to each fault attractor. It achieves dynamic prediction of fault evolution trends and can estimate the time window for fault occurrence in advance. By introducing a temperature-phase space correlation mapping, it uses the rate of temperature change to predict the expected phase space characteristics corresponding to future temperatures and integrates temperature correlation deviations with attractor migration directions. This overcomes the shortcomings of existing methods that fail to utilize the influence of temperature changes on phase space characteristic correlations, improving the accuracy of fault evolution trend prediction under temperature-changing environments.

[0117] Furthermore, by performing Bayesian inference based on the system topological correlation matrix and the anomaly trend of the Mahalanobis distance from multiple sensors, potential fault sources can be inferred from the nascent state of the anomaly trend and the physical propagation law before all sensors have fully displayed anomaly signals. This overcomes the lag in existing methods that require waiting for multiple sensors to actually exhibit anomalies before locating the fault source, achieving early and forward-looking prediction of the fault source. In summary, this implementation integrates three methods: temperature correlation prediction, attractor migration analysis, and multi-sensor temporal inference. It can simultaneously provide comprehensive predictive information on fault modes, occurrence time, evolution rate, and the location of the original fault source, providing sufficient decision support for the system to take preventative control measures and improving the stability of the power amplifier system in a wide temperature range.

[0118] Considering the application of an uplink power amplifier on a certain satellite platform, this system needs to be used in polar orbit missions. Stable operation over a wide temperature range. The system uses microwave power devices with a target power output of 100W and an operating frequency of 8GHz.

[0119] Step 100 Implementation Example:

[0120] Table 1 shows the system at 1000 sampling points ( Real-time data collection within the system:

[0121]

[0122] Output power The first few values ​​of the standardized sample sequence are: .

[0123] Rate of temperature change This indicates the heating rate, meaning the system is in a rapid heating state. The Mahalanobis distance values ​​of each sensor gradually increase, indicating that the sensor channel characteristic values ​​deviate from the normal distribution.

[0124] Step 200 implementation example:

[0125] Using time delay parameters Embedded dimension Time window length Phase space reconstruction is performed on the standardized output signal.

[0126] Table 2. Locations of phase space trajectory points and centroids:

[0127]

[0128] Normal attractor center Fault attractor 1 (gain attenuation) center Fault attractor 2 (operating point drift) center .

[0129] The center of mass gradually moves closer to the fault attractor 1 from the normal state, and the distance shrinks to 0.2 (dimensionless unit) in the fourth time window.

[0130] Step 300 Implementation Example:

[0131] The centroid position sequence within time window 2-5 is fitted using the least squares method to obtain the migration direction vector. Migration speed per unit time (Dimensionless units / time window).

[0132] Calculate the cosine similarity between the migration direction and the directions of each fault attractor:

[0133] Fault attractor 1 direction: ,

[0134] Fault attractor 2 direction: ,

[0135] Baseline confidence level: ,

[0136] The estimated arrival times from the current centroid (5th time window) to each fault attractor are:

[0137]

[0138]

[0139] Predicted temperature: The desired phase space characteristics are obtained by querying the temperature-phase space mapping table. .

[0140] Temperature correlation deviation vector , module length .

[0141] The cosine value of the direction of temperature deviation and the direction of migration The temperature change was determined to be consistent with the fault evolution trend.

[0142] Correct confidence: (After normalization)

[0143] Mahalanobis distance anomaly trend: This indicates that the abnormality of sensor channel 1 is worsening.

[0144] Bayesian inference is used to calculate the posterior probability of each sensor location as a fault source. Based on the topological correlation matrix and anomaly trends, the following is known:

[0145]

[0146]

[0147]

[0148] Step 400 Implementation Example:

[0149] The predicted fault mode calculated in step 300 is "gain decay type fault" (corresponding to fault attractor 1). The fault mode with the shortest expected arrival time is selected as the main risk, with an expected arrival time of 3.6 minutes and a confidence level of 90%.

[0150] The predicted fault modes are correlated and matched with the sensor fault source probabilities calculated by Bayesian inference: gain attenuation faults mainly occur at the power device location, which has a posterior probability of 0.78. Therefore, the most likely fault source location is the power device, with a posterior probability of 78%.

[0151] Predicted Failure Mode: "Gain Attenuation Failure"

[0152] Estimated time of occurrence: within 3.6 minutes

[0153] Migration speed: 0.48 dimensionless units / minute

[0154] Confidence level: 90%

[0155] Fault source location: power devices

[0156] Posterior probability of fault source: 78%

[0157] Based on phase space target direction After establishing the parameter sensitivity matrix, the adjustment amount of the control parameters is calculated.

[0158] Table 3: Control Parameter Adjustment Amounts

[0159]

[0160] The system immediately executes a control strategy, reducing gain to prevent excessive power output and increasing the operating point voltage and bias current to compensate for device performance degradation caused by temperature rise. The centroid of the phase space trajectory begins to move towards the normal attractor direction, and system stability is restored.

[0161] During its operation in Mars orbit, the X-band downlink power amplifier system of a deep space probe experienced a temperature increase from [missing information - likely a temperature range] as it entered a region exposed to sunlight. Quickly rise to The rate of temperature change reached The power amplifier system operates at a frequency of 8.4 GHz and has a rated output power of 80 W. The system is equipped with five sensor channels to monitor the operating status of the power devices, matching network, heat dissipation module, power supply module, and output port.

[0162] The system performs real-time monitoring and fault prediction at 600 consecutive sampling points after entering the sunlight area, following the method described in this embodiment:

[0163] In the early stages of temperature rise, the raw monitoring data collected by the system showed that the output power began to fluctuate slightly, and the Mahalanobis distance values ​​of each sensor channel were all within the normal range.

[0164] Table 4: Mahalanobis distance values ​​for each sensor channel:

[0165]

[0166] After Z-score normalization of the output power sampling sequence, the time delay parameter is used. Embedded dimension Time window length Phase space reconstruction is performed. The calculated sequence of centroid positions of the phase space trajectory and their distances to each attractor are as follows:

[0167] Table 5. Sequence of the centroid positions of phase space trajectories and their distances from each attractor:

[0168]

[0169] As the temperature continued to rise, data from time windows 4-6 showed a clear downward trend in output power, with the most significant increase in the Mahalanobis distance at sensor 1 (power device monitoring). This was observed over four consecutive time windows ( The centroid position sequence is fitted using the least squares method to obtain the centroid migration direction vector. migration speed Dimensionless units / time window.

[0170] Calculate the cosine similarity between the migration direction and the directions of two pre-stored fault attractors: the direction vector of fault attractor A (power device thermal failure). cosine similarity ; Direction vector of fault attractor B (matching network drift) cosine similarity The basic confidence level was calculated as follows. , .

[0171] Based on the current temperature and rate of temperature change The predicted temperature in 1 minute is The desired phase space characteristics at a given temperature are obtained by querying the temperature-phase space characteristic mapping table. Calculate the temperature correlation deviation vector , module length .

[0172] Calculate the cosine of the angle between the direction of temperature deviation and the direction of migration. The temperature change trend was determined to be consistent with the fault evolution trend. A temperature correlation weighting coefficient was used. Correct the confidence level of fault attractor A After normalization Expected time to reach fault attractor A The time window is approximately 4.4 minutes.

[0173] Simultaneously calculate the anomaly trend of Mahalanobis distance for each sensor channel:

[0174] Table 6: Anomaly trends in Mahalanobis distance for each sensor channel:

[0175]

[0176] System topology correlation matrix The association strengths between the power device and the matching network and the heat dissipation module are 0.45 and 0.35, respectively. Using a Bayesian inference algorithm, combining the prior probability (the historical failure frequency of the power device is 0.35) and the likelihood probability (calculated based on topological association and anomaly trends), the posterior probability of each location as a fault source is obtained:

[0177] Table 7 Posterior probabilities of fault sources:

[0178]

[0179] The system outputs the comprehensive prediction results: the predicted fault mode is "power device thermal failure", the expected occurrence time is within 4.4 minutes, the confidence level is 85%, the most likely fault source location is the power device, the posterior probability is 74%, and the current migration speed is 0.52 dimensionless units / minute.

[0180] Based on the prediction results, the system calculates the target direction vector in phase space. Based on the pre-established parameter sensitivity matrix, the control parameter adjustment amount is solved using the pseudo-inverse matrix method:

[0181] Table 8 Control Parameter Adjustment Amounts:

[0182]

[0183] After the system executes the control strategy, the migration direction of the phase space trajectory centroid changes within the next three time windows, and it begins to move towards the normal attractor. Approaching. At time window 9, the centroid position is... The distance to the normal attractor decreased to 0.35, while the distance to the faulty attractor A increased to 2.10, and the output power recovered to 79.5W. The abnormal trend of the Mahalanobis distance in each sensor channel began to decrease, and the power device... The system stability was effectively maintained by reducing the value from 1.5 to 0.3.

[0184] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A power amplifier stability control method for accommodating a wide temperature environment, characterized by, The method comprises the following steps: obtaining a real-time sampling sequence of an output signal of a power amplification system and current temperature data; reconstructing a phase space of the output signal by using a time delay embedding method to generate a phase space vector sequence and calculating a centroid of a phase space trajectory in a current time window; obtaining an attractor center under a normal working condition and attractor centers corresponding to each typical fault mode from a historical database; recording a centroid position sequence in a plurality of continuous time windows and fitting a direction vector of a centroid migration trajectory; a cosine similarity between the migration direction vector and a direction vector of each fault attractor is obtained by an inner product of the migration direction vector and the direction vector of the fault attractor divided by a product of norms of the two vectors; based on the cosine similarity and a distance between a current centroid and each fault attractor, a predicted time for the centroid to reach each fault attractor is predicted; selecting a primary risk fault type according to the predicted time and outputting a fault evolution prediction result and generating a control strategy for moving the centroid of the phase space trajectory to a direction of a normal attractor.

2. The power amplification stability control method for adapting to a wide temperature environment according to claim 1, wherein, The predicted time is obtained by dividing the distance between the current centroid and the fault attractor by a product of an absolute value of the cosine similarity and a migration speed, wherein the migration speed is a norm of the migration direction vector.

3. The method of claim 1, wherein, The phase space reconstruction of the output signal by using the time delay embedding method comprises: delay coordinate reconstruction of the output signal sampling sequence according to a time delay parameter and an embedding dimension to generate a phase space vector sequence composed of a current sampling value and a plurality of delayed sampling values thereof.

4. The method of claim 1, wherein, The fitting of the direction vector of the centroid migration trajectory comprises: linear fitting of the centroid position sequence in the plurality of continuous time windows by using a least square method to obtain a direction vector representing a moving trend of the centroid.

5. The method of claim 1, wherein, Further comprising: obtaining a change rate of the current temperature; predicting a future temperature according to the current temperature and the change rate of the temperature; querying an expected phase space feature corresponding to the future temperature from a pre-stored temperature-phase space feature mapping table; calculating a temperature correlation deviation vector between the current centroid and the expected phase space feature; fusing and analyzing directions of the temperature correlation deviation vector and the migration direction vector, and increasing a prediction confidence of a fault type when the two directions point to a same direction of a fault attractor.

6. The method of claim 5, wherein, The increasing of the prediction confidence of the fault type comprises: calculating a cosine value of an included angle between the temperature correlation deviation vector and the migration direction vector; multiplying a basic confidence by a temperature correlation factor determined by the cosine value of the included angle and a temperature correlation weight coefficient to obtain an adjusted prediction confidence.

7. The method of claim 1, wherein, Further comprising: obtaining a Mahalanobis distance sequence of a plurality of sensor channels distributed at positions of the system, the Mahalanobis distance representing a deviation degree of an output feature of each sensor channel relative to a normal state distribution; calculating a Mahalanobis distance abnormal trend of each sensor channel; obtaining a pre-stored system topology correlation matrix, the topology correlation matrix representing a physical correlation strength between sensors; based on the abnormal trend and the topology correlation matrix, calculating a probability of each sensor position as a potential fault source by using Bayesian inference.

8. The method of claim 7, wherein, The calculation of the probability of each sensor position as the potential fault source by using the Bayesian inference comprises: According to the fault history statistics of each position in the system topology, the prior probability of each sensor position as a fault source is determined; According to the topology correlation matrix and the physical propagation law, the likelihood probability of observing the current abnormal trend distribution under the condition that a fault occurs at each sensor position is calculated; The likelihood probability is multiplied by the prior probability and normalized to obtain the posterior probability of each sensor position as a potential fault source.

9. The method of claim 7, wherein, Also includes: Correlate and match the predicted fault type with the sensor fault source probability, filter out the position set where the fault type is most likely to occur according to the fault mode-position correlation knowledge, and determine the most likely fault type-fault source combination; When outputting the fault evolution prediction result, the most likely original fault source position information is also output.

10. A power amplifier stability control system for wide temperature environments for performing the power amplifier stability control method for wide temperature environments of any one of claims 1-9, wherein Includes: A data acquisition module for acquiring real-time sampling sequences of power amplification system output signals and current temperature data; A phase space reconstruction module for reconstructing the output signals using the time delay embedding method to generate a phase space vector sequence, calculating the center of mass of the phase space trajectory in the current time window, and obtaining the attractor center under normal working conditions and the attractor center corresponding to each typical fault mode from a historical database; A migration trajectory analysis module for recording the center of mass position sequence in a plurality of consecutive time windows, fitting the direction vector of the center of mass migration trajectory, calculating the cosine similarity between the migration direction vector and each fault attractor direction vector, and predicting the predicted time for the center of mass to reach each fault attractor based on the cosine similarity and the distance between the current center of mass and each fault attractor; A control strategy output module for selecting the primary risk fault type according to the predicted time, outputting the fault evolution prediction result, and generating a control strategy for moving the phase space trajectory center of mass in the direction of the normal attractor.

Images