CN113283113B - Solar cell array power generation current prediction model training method, abnormality detection method, device and medium - Google Patents
Solar cell array power generation current prediction model training method, abnormality detection method, device and medium Download PDFInfo
- Publication number
- CN113283113B CN113283113B CN202110656835.2A CN202110656835A CN113283113B CN 113283113 B CN113283113 B CN 113283113B CN 202110656835 A CN202110656835 A CN 202110656835A CN 113283113 B CN113283113 B CN 113283113B
- Authority
- CN
- China
- Prior art keywords
- solar cell
- cell array
- spacecraft
- power generation
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Development Economics (AREA)
- Geometry (AREA)
- Game Theory and Decision Science (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Photovoltaic Devices (AREA)
Abstract
The invention provides a spacecraft solar cell array power generation current prediction model training method, an anomaly detection method, equipment and a medium. The method comprises the steps of training a plurality of LS-SVM base models, calculating base model weights by utilizing the prediction precision of each base model on a verification set, and obtaining an integrated LS-SVM prediction model through weighting to serve as a finally trained solar cell array power generation current prediction model. Acquiring the deviation between the actual value of the solar cell array generating current of the spacecraft and the expected value predicted by the solar cell array generating current prediction model; and setting an abnormality judgment criterion about the deviation, and detecting whether the solar cell array is abnormal or not based on the deviation. The method can better solve the prediction problem of the spacecraft telemetry data.
Description
Technical Field
The invention relates to the technical field of solar cell array performance detection, in particular to a method for training a generation current prediction model of a spacecraft solar cell array and an abnormality detection method.
Background
The solar cell array is used as an indispensable component in a spacecraft power supply system, absorbs solar radiation to generate electricity when the spacecraft runs to an illumination zone, and provides necessary energy for the spacecraft to perform various activities and tasks on orbit. However, since the solar cell array works in a severe space environment and is exposed to corrosion of atomic oxygen, protons, electrons and the like for a long time, the power generation reliability of the solar cell array inevitably and continuously decreases until faults such as component damage and the like occur. And the failure of the solar cell array further causes the conditions that the storage battery pack cannot reach full charge, the spacecraft cannot realize energy balance and the like, so that the spacecraft cannot normally execute an on-orbit task. In view of this, how to detect the possible abnormality of the solar cell array of the spacecraft in time before or in an early stage of the occurrence of the fault of the solar cell array of the spacecraft so as to take corresponding measures in time is one of the challenges faced by ground management personnel of the spacecraft.
At present, most ground monitoring stations adopt a detection method based on manual monitoring and threshold value to carry out anomaly detection on telemetering data of a spacecraft solar cell array: and (3) detecting whether the parameter data exceeds a preset interval by combining the telemetering data acquired in real time with a preset threshold value by a ground technician, judging that the data is abnormal if the parameter data exceeds the preset interval, and otherwise, judging that the data is normal. The method is simple to operate, most of the abnormal telemetry data of the solar cell array can be screened out, but the requirement on the specificity of threshold setting for the telemetry parameters is high, and the threshold detection method cannot detect the abnormality of parameter values such as abnormal series-open circuit of the solar cell array in a threshold range.
The anomaly detection method based on data driving can well solve short boards appearing in threshold method anomaly detection, and the change rule of telemetering data is characterized and modeled by adopting theories such as statistics, machine learning and deep learning so as to identify an anomaly mode in the telemetering data. According to the difference of the metric indexes, the anomaly detection method based on data driving can be further divided into three methods, namely a statistical-based method, a similarity-based method and a deviation-based method. The statistical-based method usually assumes overall distribution of data, and considers a value with a higher probability of occurrence in the distribution as normal data, and considers a value with a probability of occurrence in the distribution lower than a certain threshold as abnormal data. However, actual telemetry data is often more difficult to characterize with a predetermined distribution, which makes statistical-based methods much less effective in the face of anomaly detection of spacecraft telemetry data. The similarity-based method screens abnormal parts in data by measuring similarity indexes, data points with lower similarity index performance are subjected to abnormal marking, the abnormal detection performance of the similarity-based method is very sensitive to the similarity indexes, and if a proper similarity measurement index is not found, the abnormal detection accuracy of the method is often unsatisfactory. The deviation-based method is used for establishing a model according to normal data, bringing test data into the model to obtain a predicted expected value of a target parameter, and performing abnormality detection by calculating the deviation between the expected value and an actual sample. The deviation-based method mainly comprises a classification algorithm and a prediction algorithm: the classification algorithm requires that training data are provided with labels, the number of normal data points is close to that of abnormal data points, most of spacecraft telemetering data are normal data, and abnormal data labels are not available, so that the classification algorithm cannot achieve a good abnormal detection effect on the spacecraft telemetering data. In comparison, the prediction algorithm is more suitable for anomaly detection of spacecraft telemetry data, the model is trained through normal data with significant telemetry data, the data is not dependent on whether sufficient anomaly data and anomaly labels exist in the data, and the learning capability is strong.
However, when the spacecraft solar cell array data is faced, due to the fact that the data volume is large and the number of noise points in the data is large, the traditional prediction algorithm is prone to overfitting, the average value and the variance of the overall prediction error are increased, and therefore the spacecraft solar cell array data is poor in detection capability of the spacecraft solar cell array anomaly and low in anomaly detection accuracy.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a spacecraft solar cell array power generation current prediction model training method, an anomaly detection method, equipment and a medium.
In order to achieve the technical purpose, the invention adopts the following specific technical scheme:
the method for training the generation current prediction model of the spacecraft solar cell array comprises the following steps:
constructing n LS-SVM base models;
acquiring on-orbit telemetry data of a spacecraft solar cell array as an off-line data set for training n LS-SVM base models, wherein the on-orbit telemetry data of the spacecraft solar cell array is solar cell array generating current and on-orbit telemetry parameter data related to the solar cell array generating current;
taking out a part of spacecraft solar cell array on-orbit telemetry data from an off-line training data set as a training set, randomly extracting n groups of sub-training sets from the training set, respectively using the sub-training sets for training n LS-SVM base models, and constructing a state evolution rule between solar cell array generating current and on-orbit telemetry parameters related to the solar cell array generating current by adopting an LS-SVM base model to obtain n trained LS-SVM base models;
and taking out a part of the spacecraft solar cell array on-orbit telemetering data from the off-line training data set as a verification set, substituting the spacecraft solar cell array on-orbit telemetering data in the verification set into each trained LS-SVM base model, distributing weights to each trained LS-SVM base model by utilizing the error values of each trained LS-SVM base model on the verification set, and outputting and weighting the n trained LS-SVM base models to obtain an ILS-SVM prediction model which is used as a finally trained solar cell array power generation current prediction model.
The method acquires the on-orbit telemetry data of the spacecraft solar cell array at the initial on-orbit stage as an off-line data set for training n LS-SVM base models. The initial stage of the on-orbit is a period (such as half a year, 1 year and the like) counted from the time when the spacecraft is put into service and starts to work normally, and the solar cell array of the spacecraft in the initial stage of the on-orbit is not abnormal and is at a normal health level. The on-orbit telemetering data of the spacecraft solar cell array refer to the on-orbit telemetering parameter data with strong correlation between the generating current of the solar cell array and the generating current of the solar cell array, and can be selected by adopting correlation analysis methods such as Pearson correlation coefficient, maximum mutual information coefficient and the like. On-orbit telemetry parameters strongly correlated with the solar cell array generating current include, but are not limited to: solar cell array temperature, solar radiation intensity, orbital plane angle and other parameters.
The invention provides an anomaly detection method for a spacecraft solar cell array, which comprises the following steps:
acquiring on-orbit telemetering data of a spacecraft solar cell array;
obtaining an expected value of the solar cell array generating current corresponding to the spacecraft solar cell array on-orbit telemetry data to be detected by utilizing the trained solar cell array generating current prediction model obtained by any one of the above solar cell array generating current prediction model training methods;
calculating the deviation between the actual value of the solar cell array generating current of the spacecraft and the expected value predicted by the solar cell array generating current prediction model;
and setting an abnormality judgment criterion about the deviation, and detecting whether the solar cell array is abnormal or not based on the deviation.
Further, the anomaly determination criterion of the present invention is: and if the acquired deviations corresponding to the on-orbit telemetry data of the continuous m spacecraft solar cell arrays exceed a set deviation threshold value, judging that the spacecraft solar cell arrays are abnormal, wherein m is more than or equal to 1, and if m is 3.
The method for determining the deviation threshold value comprises the following steps: taking out a part of on-orbit telemetry data of the spacecraft solar array from the off-line training data set as a test set, inputting the test set into a trained solar array power generation current prediction model to obtain a solar array power generation current expected value predicted by the solar array power generation current prediction model, and comparing the actual value of the solar array power generation current of the test set with the solar array power generation current expected value predicted by the solar array power generation current prediction model to obtain a prediction deviation sequence e; considering that the predicted deviation sequence e obeys or approximately obeys normal distribution in the case of a sufficiently large test set data amount, the deviation threshold value for anomaly detection is calculated by using the Laviand criterion as follows: assuming that the mean of the predicted deviation series e is μ and the standard deviation is σ, the deviation threshold for anomaly detection is eth＝μ+3σ。
The invention provides electronic equipment which comprises a processor, a memory and a computer program which is stored on the memory and can run on the processor, wherein when the computer program is executed by the processor, the method for training the solar cell array generating current prediction model or the method for detecting the abnormality of the spacecraft solar cell array is realized.
The invention provides a computer-readable storage medium, wherein a computer program is stored on the computer-readable storage medium, and when being executed by a processor, the computer program realizes the solar cell array power generation current prediction model training method or the spacecraft solar cell array anomaly detection method.
In summary, compared with the prior art, the invention can at least bring the following beneficial effects:
1. the invention provides a method for training a generation current prediction model of a spacecraft solar cell array, which comprises the following steps: the method has the advantages that a plurality of LS-SVM base models are trained, the base model weight is calculated by utilizing the prediction precision of each base model on a verification set, an integrated LS-SVM (subsequently abbreviated as ILS-SVM) prediction model is obtained through weighting and serves as a finally trained solar cell array generating current prediction model, and the method has higher prediction precision and generalization performance compared with a common neural network method and a single LS-SVM prediction model, and can better process the prediction problem of spacecraft telemetering data;
2. the invention designs a parameter-free and unsupervised abnormity judgment criterion based on the constructed ILS-SVM prediction model for identifying the abnormity in the telemetering data, and the method can effectively avoid the condition that the noise point is identified as the abnormity, thereby reducing the abnormity detection false alarm rate;
3. the anomaly detection method for the spacecraft solar cell array is based on a data driving layer, the requirement on the professional knowledge of ground operators is almost zero, and meanwhile, the ILS-SVM prediction model is pre-trained, so that the method can be operated simply, has high accuracy and is high in detection speed when the real-time anomaly detection of the solar cell array is carried out.
Drawings
FIG. 1 is an overall flow chart of an embodiment of the present invention;
fig. 2 is a graph illustrating a variation of parameters associated with a solar cell array according to an embodiment of the present invention;
FIG. 3 is a flowchart of training an ILS-SVM predictive model according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating the prediction effect of the solar array generated current of the ILS-SVM prediction model according to an embodiment of the present invention;
FIG. 5 is a diagram of a sequence of predicted deviations of an ILS-SVM predictive model in a test set according to an embodiment of the present invention, wherein (a) is a diagram of an anomaly determination result using a conventional anomaly determination method, and (b) is a diagram of an anomaly determination result using the anomaly determination criterion of the present invention;
FIG. 6 is a diagram of the anomaly detection results (in the form of a deviation sequence chart) for the 9 th and 18 th day telemetry data in accordance with an embodiment of the present invention;
fig. 7 is a graph of the anomaly detection results (in the form of graphs) for the 9 th and 18 th day telemetry data in an embodiment of the present invention, wherein (a) is a graph comparing expected values and actual values of the model between samples 200 and 390; (b) a graph of the model expected values versus actual values between samples 390-580;
fig. 8 is a graph showing the results of anomaly detection (in the form of a variation graph) for the 9/19 th day telemetry data in accordance with an embodiment of the present invention, wherein (a) is a graph showing the expected value versus the actual value for the model for the first cycle to the fifth cycle of the ten cycles before the 9/19 th day, and (b) is a graph showing the expected value versus the actual value for the model for the sixth cycle to the tenth cycle of the ten cycles before the 9/19 th day.
Detailed Description
In order to make the technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
An embodiment of the invention provides a method for training a generation current prediction model of a spacecraft solar cell array, which comprises the following steps:
s1, constructing n LS-SVM base models;
s2, acquiring on-orbit telemetering data of the spacecraft solar cell array as an off-line data set for training n LS-SVM base models, wherein the on-orbit telemetering data of the spacecraft solar cell array are on-orbit telemetering parameter data related to the power generation current of the solar cell array and the power generation current of the solar cell array.
In the embodiment, the historical healthy on-orbit telemetry data of the spacecraft solar cell array is obtained and used as off-line data for training an ILS-SVM model. The historical health on-orbit telemetry data of the spacecraft solar cell array refers to data of the spacecraft solar cell array in a normal health level and in an abnormal period recorded by ground technicians. If on-orbit telemetry data of the spacecraft solar cell array at the initial stage of the on-orbit is acquired, the initial stage of the on-orbit refers to a period (such as half a year, 1 year and the like) counted from the time when the spacecraft is in orbit service and starts to work normally, and the spacecraft solar cell array at the initial stage of the on-orbit is not abnormal and is at a normal health level. The on-orbit telemetering data of the spacecraft solar cell array refer to the on-orbit telemetering parameter data with strong correlation between the generating current of the solar cell array and the generating current of the solar cell array, and can be selected by adopting correlation analysis methods such as Pearson correlation coefficient, maximum mutual information coefficient and the like. On-orbit telemetry parameters strongly correlated with the solar cell array generating current include, but are not limited to: solar cell array temperature, solar radiation intensity, orbital plane angle and other parameters.
S3, taking out a part of on-orbit telemetry data of the spacecraft solar cell array from the off-line training data set as a training set, randomly extracting n groups of sub-training sets from the training set, respectively using the sub-training sets for training n LS-SVM base models, and constructing a state evolution rule between the solar cell array generating current and on-orbit telemetry parameters related to the solar cell array generating current by adopting an LS-SVM base model to obtain n trained LS-SVM base models;
and S4, taking out a part of on-orbit telemetering data of the spacecraft solar cell array from the off-line training data set as a verification set, substituting the on-orbit telemetering data of the spacecraft solar cell array in the verification set into each trained LS-SVM base model, distributing weights to each trained LS-SVM base model by utilizing the error values of each trained LS-SVM base model in the verification set, outputting the weights to the n trained LS-SVM base models to obtain an ILS-SVM prediction model, and taking the ILS-SVM prediction model as a finally trained solar cell array power generation current prediction model.
According to the method, an LS-SVM basic model is adopted to construct a state evolution rule between the solar cell array power generation current and the on-orbit remote measurement parameters related to the solar cell array power generation current. The LS-SVM base model adopted in the invention is an improvement on the basis of a kernel SVM model, the kernel SVM model maps the on-orbit telemetering data space of the spacecraft solar cell array to a high-dimensional feature space through a kernel function, and a hyperplane is utilized to fit the nonlinear relation between the feature parameters and the target parameters. The LS-SVM base model adopted by the invention converts the complex quadratic programming problem in the kernel SVM model into a linear equation set problem to be solved, and replaces inequality constraints in the kernel SVM model optimization problem with equality constraints, so that the complexity of the problem is reduced, and the operation time of model solution is reduced. The method enables the LS-SVM base model to quickly fit the strong nonlinear relation between the solar cell array generating current and the on-orbit telemetry parameters related to the solar cell array generating current when the on-orbit telemetry data of the spacecraft solar cell array with huge data amount are faced.
An embodiment of the present invention provides an anomaly detection method for a spacecraft solar cell array, including:
acquiring on-orbit telemetering data of a spacecraft solar cell array;
obtaining an expected value of the solar cell array generating current corresponding to the spacecraft solar cell array on-orbit telemetry data to be detected by utilizing the trained solar cell array generating current prediction model obtained by the solar cell array generating current prediction model training method in the embodiment;
calculating the deviation between the actual value of the solar cell array generating current of the spacecraft and the expected value predicted by the solar cell array generating current prediction model;
and setting an abnormality judgment criterion about the deviation, and detecting whether the solar cell array is abnormal or not based on the deviation.
In the embodiment, the acquired on-orbit telemetry data of the spacecraft solar cell array is input into a trained solar cell array power generation current prediction model, and the deviation between the actual value of the spacecraft solar cell array power generation current and the expected value predicted by the solar cell array power generation current prediction model is calculated. In the process of acquiring and transmitting the on-orbit telemetry data of the spacecraft solar cell array, individual data received by the ground station are distorted due to the influence of factors such as space environment and the like, so that the deviation between the measured data value and the expected value is obviously increased. In order to avoid determining this phenomenon as abnormal, the false alarm rate of abnormal detection is reduced. The abnormity judgment criterion set by the invention is as follows: and if the acquired deviations corresponding to the on-orbit telemetry data of the continuous m spacecraft solar cell arrays exceed a set deviation threshold value, judging that the spacecraft solar cell arrays are abnormal, wherein m is more than or equal to 1, and if m is 3.
The method for determining the deviation threshold in this embodiment is as follows: taking out a part of spacecraft solar array on-orbit remote measurement data from an off-line training data set as a test set, inputting the test set into a trained solar array power generation current prediction model to obtain a solar array power generation current expected value predicted by the solar array power generation current prediction model, and comparing the actual value of the solar array power generation current of the test set with the solar array power generation current expected value predicted by the solar array power generation current prediction model to obtain a prediction deviation sequence e; considering that the predicted deviation sequence e obeys or approximately obeys normal distribution in the case of a sufficiently large test set data amount, the deviation threshold value for anomaly detection is calculated by using the Laviand criterion as follows: assuming that the mean of the predicted deviation series e is μ and the standard deviation is σ, the deviation threshold for anomaly detection is eth＝μ+3σ。
Referring to fig. 1 to 8, a specific embodiment of the method for detecting an anomaly of a spacecraft solar cell array according to the present invention is shown. The research of this embodiment is a certain sun synchronous orbit satellite, and the satellite ground monitoring personnel catches that the satellite storage battery pack fails to reach the full charge state for a plurality of continuous cycles within 9 months and 24 days in 2020, and it is known through traceability analysis that the solar battery array has a battery series open circuit phenomenon, wherein four solar battery series open circuits, and the generation current of the solar battery array is reduced by 1.0A compared with the conventional state, so that the charging current of the storage battery pack in the illumination area is also correspondingly reduced by 0.8A, and finally the storage battery pack fails to reach the full charge state. In the embodiment, the first 80% of data from 2019, 8 months to 2020, 8 months of the satellite is used as a training set to train a base model LS-SVM, the last 20% of data is used as a verification set to distribute weights to all base models, weighting integration is carried out to obtain an ILS-SVM prediction model, data from 9 months 1 to 9 months 15 days in 2020 is used as a test set to calculate a deviation threshold, and then the trained ILS-SVM model and an abnormality judgment criterion are used for carrying out abnormality detection on subsequent solar cell array telemetry data. The flow chart of the whole method is shown in fig. 1, and specifically comprises the following steps:
step 1: and acquiring on-orbit telemetry data of the satellite solar cell array.
In this embodiment, the on-orbit telemetry parameters related to the solar cell array power generation current selected according to the correlation analysis include: the solar cell array power generation current monitoring method comprises the following steps of + Y windsurfing board outer plate temperature (recorded as parameter + Y temp), -Y windsurfing board outer plate temperature (recorded as-Y temp), DSS1 solar radiation intensity (recorded as parameter DSS1), DSS2 solar radiation intensity (recorded as parameter DSS2), ASS1 output angle (recorded as parameter ASS1), ASS21 output angle (recorded as parameter ASS21) and ASS3 field monitoring signal (recorded as parameter ASS3), wherein target parameters are solar cell array power generation current (recorded as parameter SA current), and variation curves of the parameters are shown in FIG. 2. The satellite data set time span acquired by the embodiment is 8 months in 2019 to 9 months in 2020.
Step 2: the method comprises the steps of preprocessing acquired satellite solar cell array on-orbit telemetry data, using the preprocessed satellite solar cell array on-orbit telemetry data as an off-line data set for training n LS-SVM base models, and dividing the on-orbit telemetry data into a training set, a verification set and a test set according to a certain proportion.
In the embodiment, in the process of acquiring and transmitting the satellite telemetry data, partial data distortion is caused due to factors such as noise interference and instantaneous sensor failure, and the distortion data causes certain interference on the subsequent training of an ILS-SVM prediction model, so that the distortion value elimination is performed on the acquired satellite solar cell array in-orbit telemetry data by adopting the Lauda rule. Ether typeThe positive battery array generated current parameter is taken as an example to explain the specific distortion value elimination process: calculating the mean value mu of the generated current of the solar cell array during the period from 8 months in 2019 to 9 months in 2020SAStandard deviation σSAConsidering that the data amount is large enough, the solar cell array power generation current value can be approximately regarded as a normal distribution in which the values are distributed in (μ)SA-3σSA，μSA+3σSA) The probability of (2) is 0.9974, namely the probability of exceeding the range is less than 0.3%, so that the solar cell array power generation current value exceeding the range is regarded as a distortion value and eliminated. And subsequently segmenting data in a time period from 8 months in 2019 to 9 months in 2020, wherein the first 80% of data in the period from 8 months in 2019 to 8 months in 2020 is used as a training set, the last 20% of data is used as a verification set, and the data in the period from 1 day in 9 months in 2020 to 15 days in 9 months in 2020 is used as a test set.
And step 3: and training the n LS-SVM base models by using the data in the training set.
Step 3.1: randomly extracting n groups of sub-training sets from the training sets, and respectively training n LS-SVM base models;
step 3.2: setting key parameters of n LS-SVM base models, including regularization parameters C and kernel function parameters gamma, and then respectively inputting n sub-training sets into corresponding LS-SVM base models for model training to obtain n well-trained LS-SVM base models.
Step 3.3: and storing the structure and parameters of each trained LS-SVM base model for the subsequent construction of an ILS-SVM prediction model.
In this embodiment, 5 groups of sub-training sets are randomly extracted from the training set in a put-back manner, each group of sub-training sets accounts for 10% of the total training set, and are used for training 5 LS-SVM basis models, wherein parameters of the 5 LS-SVM basis models are set as shown in the following table:
TABLE 1 LS-SVM base model parameter settings
And 4, step 4: referring to fig. 3, the spacecraft solar cell array on-orbit telemetry data in the verification set is substituted into each trained LS-SVM base model, the error value of each trained LS-SVM base model on the verification set is used for distributing weight to each trained LS-SVM base model, and the n trained LS-SVM base models are output and weighted to obtain an ILS-SVM prediction model which is used as a finally trained solar cell array power generation current prediction model.
Step 4.1: substituting the on-orbit telemetering data of the spacecraft solar cell array with the verification concentration into each trained LS-SVM base model, calculating the prediction error of each trained LS-SVM base model, and obtaining the prediction errors of n trained LS-SVM base models, wherein the prediction errors are respectively e1，e2，...，en。
The prediction error index used here is Mean Square Error (MSE), and its calculation formula is:
where m denotes the number of samples, yiAnd y'iRespectively representing the actual value of the sample and the expected value of the model at the ith time.
Step 4.2: taking the reciprocal of the prediction error of the n trained LS-SVM base models as the prediction accuracy corresponding to each model, namely the prediction accuracy of the n trained LS-SVM base models is respectively
Distributing weight to each LS-SVM base model according to the prediction precision of n trained LS-SVM base models, wherein the weight alpha of the ith LS-SVM base modeliThe calculation formula of (2) is as follows:
step 4.3: after the weight of each LS-SVM base model is obtained, the output of each LS-SVM base model is weighted to obtain the prediction output of the ILS-SVM prediction model, namely:
wherein
Represents the output of the ILS-SVM prediction model at the ith moment, n represents the number of LS-SVM base models, and alphajRepresents the weight of the jth LS-SVM basis model,
and represents the output of the jth LS-SVM base model at the ith moment.
In this embodiment, the verification set (the last 20% of data from 8 months in 2019 to 8 months in 2020) is substituted into each LS-SVM base model, and the prediction error, prediction precision, and weight of each LS-SVM base model are calculated, as shown in the following table:
TABLE 2 prediction result information of the base model
And obtaining an ILS-SVM prediction model through the output weighting of the basic models, wherein the prediction error of the ILS-SVM prediction model obtained through calculation is 0.0770, and is effectively reduced compared with the prediction error of each basic model. The prediction result of the ILS-SVM prediction model on the telemetry data of the solar cell array power generation current in the part of 9-month-1-day 2020 is plotted in FIG. 4.
And 5: substituting the test set into the trained ILS-SVM prediction model to obtain an expected value of the solar cell array generating current predicted by the prediction model; a deviation threshold value for abnormality determination is calculated from a predicted deviation sequence between the expected value and the actual value, and an abnormality determination criterion is set.
In this embodiment, data from 9/month 1 to 9/month 15 in 2020 is substituted as a test set into the trained ILS-SVM prediction model to obtain an expected value of the generated current of the solar cell array predicted by the prediction model, and a prediction deviation calculation formula is used to calculate the expected value of the generated current of the solar cell array
And calculating to obtain a prediction deviation sequence e. Considering that the predicted deviation sequence e of the model obeys or approximately obeys a normal distribution if the test set data size is large enough, the deviation threshold for anomaly detection is calculated using the Lauda criterion for this: calculating to obtain the mean value of the prediction deviation sequence e as mu and the standard deviation as sigma, and then the deviation threshold value for anomaly detection is ethμ +3 σ, deviation threshold e calculated in this exampleth＝0.65。
The conventional abnormality determination method is to exceed the deviation threshold ethThe samples (a) are all judged to be abnormal, and the result of abnormal judgment on the telemetry data values in the period from 9/month 1 in 2020 to 9/month 15 in 2020 by using the abnormal judgment criterion is shown in fig. 5(a), although no abnormality occurs in the satellite solar cell array in this period, the deviation values of 134 data distortion points (accounting for 1.6% of the total data amount) exceed the threshold value ethIt is determined to be abnormal. In order to avoid determining the data distortion point as abnormal and effectively reduce the false alarm rate, the present embodiment makes a certain correction on the abnormal determination criterion. Under normal conditions, data distortion phenomena caused by noise interference and other factors do not continuously occur, and the prediction deviation is set to exceed the deviation threshold e when continuous 3 samples occurthThen, it is determined that an abnormality occurs in the solar cell array in the time period, and the result of detecting the abnormality after the abnormality determination criterion is replaced is shown in fig. 5 (b).
Step 6: inputting the on-orbit telemetry data of the satellite solar cell array to be detected into an ILS-SVM prediction model, predicting to obtain an expected value of the power generation current of the solar cell array, comparing the deviation between the expected value and the actual value with a set deviation threshold, and judging whether the satellite solar cell array is abnormal or not by combining an abnormality judgment criterion.
In this embodiment, in 2020, after 9, 15 and 9 months, the in-orbit telemetry data of the satellite solar cell array is input into the ILS-SVM prediction model, and abnormal conditions of the in-orbit telemetry data of the satellite solar cell array in 18 and 9 months are observed and found, as shown in fig. 6. As can be seen from the observation images, on the first half of the 9 th and 18 th days, the on-orbit telemetry data of the satellite solar cell array has no abnormal condition, and from the 259 th sample point (corresponding to the time of 10: 02 on the 9 th and 18 th days), the on-orbit telemetry data of the satellite solar cell array starts to periodically generate abnormal data. Fig. 7(a) and 7(b) plot the expected value versus the actual value of the model between samples 200 to 390 and samples 390 to 580, respectively, and mark the outliers in the actual value variation curve. Observing the image, the solar cell array power generation current data are intermittently abnormal in 18 days in 9 months, the telemetering data in the second and the third periods are abnormal in fig. 7(a), and the telemetering data in the first, the second, the third and the five periods are abnormal in the next fig. 7 (b). Fig. 8 is a graph showing expected values of the model for ten cycles before 9 and 19 days, compared with actual values, in which fig. 8(a) is a graph showing expected values of the model for the first to fifth cycles among ten cycles before 9 and 19 days, compared with actual values, and fig. 8(b) is a graph showing expected values of the model for the sixth to tenth cycles among ten cycles before 9 and 19 days, compared with actual values. And observing the image information, and finding that the solar cell array generating current continuously generates abnormity from the first period of 9 months and 19 days. The following conclusions can be made through analysis: the satellite solar battery array is 10 at 9 months, 18 days and 2020: 02 no abnormality occurred before; at 9 months, 18 days 10: 18 days 02 to 9, 21: in the period of 30, the solar cell array has not formally appeared the disconnection abnormality of the cell string, but the intermittent abnormality of the generating current of the solar cell array is appeared due to partial electronic components and the edge which is endangered to be damaged; and at 9 months, 18 days 21: after 30, the solar cell array has an irreversible cell string disconnection abnormality, and the generated current is reduced by about 1A compared with the conventional state.
The invention accurately identifies the satellite solar cell array in the 21 st 18 th 9 th 2020: the battery string circuit breaking abnormality after 30 days is five days earlier than that found by ground detection personnel in 24 days at 9 months. In addition, the invention also detects that the satellite solar cell array has certain symptoms before the abnormal disconnection of the cell string occurs, and can help ground monitoring personnel to adjust the solar cell array in advance to reduce the economic loss caused by the subsequent abnormality of the solar cell array and even avoid the subsequent abnormality of the solar cell array.
In summary, although the present invention has been described with reference to the preferred embodiments, it should be understood that various changes and modifications can be made by those skilled in the art without departing from the spirit and scope of the invention.
Claims (10)
1. The method for training the generation current prediction model of the spacecraft solar cell array is characterized by comprising the following steps of:
constructing n LS-SVM base models;
acquiring on-orbit telemetry data of a spacecraft solar cell array as an off-line data set for training n LS-SVM base models, wherein the on-orbit telemetry data of the spacecraft solar cell array is solar cell array generating current and on-orbit telemetry parameter data related to the solar cell array generating current;
taking out a part of spacecraft solar cell array on-orbit telemetry data from an off-line training data set as a training set, randomly extracting n groups of sub-training sets from the training set, respectively using the sub-training sets for training n LS-SVM base models, and constructing a state evolution rule between solar cell array generating current and on-orbit telemetry parameters related to the solar cell array generating current by adopting an LS-SVM base model to obtain n trained LS-SVM base models;
and taking out a part of the spacecraft solar cell array on-orbit telemetering data from the off-line training data set as a verification set, substituting the spacecraft solar cell array on-orbit telemetering data in the verification set into each trained LS-SVM base model, distributing weights to each trained LS-SVM base model by utilizing the error values of each trained LS-SVM base model on the verification set, and outputting and weighting the n trained LS-SVM base models to obtain an ILS-SVM prediction model which is used as a finally trained solar cell array power generation current prediction model.
2. The method for training the generated current prediction model of the spacecraft solar array of claim 1, wherein the on-orbit telemetry data of the on-orbit early spacecraft solar array is obtained as an off-line data set for training n LS-SVM base models.
3. The spacecraft solar array power generation current prediction model training method according to claim 1, wherein the on-orbit telemetry parameters related to the solar array power generation current refer to on-orbit telemetry parameter data with strong correlation to the solar array power generation current, and the on-orbit telemetry parameters with strong correlation to the solar array power generation current are selected by using a Pearson correlation coefficient analysis method or a maximum mutual information coefficient analysis method.
4. The spacecraft solar array power generation current prediction model training method of claim 1, wherein the on-orbit telemetry parameters related to the solar array power generation current comprise solar array temperature, solar radiation intensity, orbital plane angle.
5. The anomaly detection method for the spacecraft solar cell array is characterized by comprising the following steps:
acquiring on-orbit telemetering data of a spacecraft solar cell array;
obtaining an expected value of the solar cell array power generation current corresponding to the spacecraft solar cell array on-orbit telemetry data to be detected by using the trained solar cell array power generation current prediction model obtained by the solar cell array power generation current prediction model training method according to any one of claims 1 to 4;
calculating the deviation between the actual value of the solar cell array generating current of the spacecraft and the expected value predicted by the solar cell array generating current prediction model;
and setting an abnormity judgment criterion about the deviation, and detecting whether the spacecraft solar cell array is abnormal or not based on the deviation.
6. The method according to claim 5, wherein the anomaly determination criterion is: and if the acquired deviations corresponding to the on-orbit telemetry data of the m continuous spacecraft solar cell arrays exceed a set deviation threshold value, judging that the spacecraft solar cell arrays are abnormal, wherein m is larger than 1.
7. The method for detecting the abnormality of the spacecraft solar cell array according to claim 6, wherein m is 3.
8. The method for detecting an abnormality of a spacecraft solar cell array according to claim 6 or 7, wherein the method for determining the deviation threshold value is: taking out a part of spacecraft solar array on-orbit remote measurement data from an off-line training data set as a test set, inputting the test set into a trained solar array power generation current prediction model to obtain a solar array power generation current expected value predicted by the solar array power generation current prediction model, and comparing the actual value of the solar array power generation current of the test set with the solar array power generation current expected value predicted by the solar array power generation current prediction model to obtain a prediction deviation sequence e; and (3) predicting that the deviation sequence e follows or approximately follows normal distribution, and calculating a deviation threshold value for anomaly detection by adopting a Laplace criterion, wherein the method comprises the following steps: assuming that the mean of the predicted deviation series e is μ and the standard deviation is σ, the deviation threshold for anomaly detection is eth＝μ+3σ。
9. An electronic device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, wherein the computer program, when executed by the processor, implements a solar array power generation current prediction model training method according to any one of claims 1 to 4 or implements an anomaly detection method for a spacecraft solar array according to claim 5.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements a solar array generation current prediction model training method according to any one of claims 1 to 4, or implements an anomaly detection method for a spacecraft solar array according to claim 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110656835.2A CN113283113B (en) | 2021-06-11 | 2021-06-11 | Solar cell array power generation current prediction model training method, abnormality detection method, device and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110656835.2A CN113283113B (en) | 2021-06-11 | 2021-06-11 | Solar cell array power generation current prediction model training method, abnormality detection method, device and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113283113A CN113283113A (en) | 2021-08-20 |
| CN113283113B true CN113283113B (en) | 2022-04-15 |
Family
ID=77284621
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110656835.2A Active CN113283113B (en) | 2021-06-11 | 2021-06-11 | Solar cell array power generation current prediction model training method, abnormality detection method, device and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113283113B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115081552B (en) * | 2022-07-28 | 2022-11-11 | 一道新能源科技(衢州)有限公司 | Solar cell data exception handling method and system based on cloud platform |
| CN116091457B (en) * | 2023-01-16 | 2023-10-27 | 北京航科星云科技有限公司 | Satellite solar cell array anomaly prediction method and device and computer equipment |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104182630A (en) * | 2014-08-20 | 2014-12-03 | 国家电网公司 | Residual battery capacity detection method based on simplified least square support vector machine |
| JP2017184355A (en) * | 2016-03-29 | 2017-10-05 | 能美防災株式会社 | Solar power generation system and temperature abnormality detection method of solar panel |
| CN111190113A (en) * | 2020-04-15 | 2020-05-22 | 中国人民解放军国防科技大学 | Spacecraft storage battery performance degradation abnormity detection method |
| CN112632850A (en) * | 2020-12-14 | 2021-04-09 | 华中科技大学 | Method and system for detecting abnormal battery in lithium battery pack |
-
2021
- 2021-06-11 CN CN202110656835.2A patent/CN113283113B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104182630A (en) * | 2014-08-20 | 2014-12-03 | 国家电网公司 | Residual battery capacity detection method based on simplified least square support vector machine |
| JP2017184355A (en) * | 2016-03-29 | 2017-10-05 | 能美防災株式会社 | Solar power generation system and temperature abnormality detection method of solar panel |
| CN111190113A (en) * | 2020-04-15 | 2020-05-22 | 中国人民解放军国防科技大学 | Spacecraft storage battery performance degradation abnormity detection method |
| CN112632850A (en) * | 2020-12-14 | 2021-04-09 | 华中科技大学 | Method and system for detecting abnormal battery in lithium battery pack |
Non-Patent Citations (3)
| Title |
|---|
| LS-SVM算法在光伏短期功率预测中的应用;杨育刚等;《国网技术学院学报》;20141030(第05期);全文 * |
| 基于LS-SVM的混合动力镍氢电池组SOC预测;陈健美等;《中南大学学报(自然科学版)》;20130126(第01期);全文 * |
| 最小二乘支持向量机在航空蓄电池剩余容量预测中的应用;刘勇智等;《蓄电池》;20130620(第03期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113283113A (en) | 2021-08-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108734360B (en) | Correction-based ELM prediction model multidimensional telemetering data intelligent interpretation method | |
| Sharma et al. | Sensor faults: Detection methods and prevalence in real-world datasets | |
| CN113283113B (en) | Solar cell array power generation current prediction model training method, abnormality detection method, device and medium | |
| US20160369777A1 (en) | System and method for detecting anomaly conditions of sensor attached devices | |
| CN114371409B (en) | Training method of battery state prediction model, battery state prediction method and device | |
| WO2023070960A1 (en) | Wind power prediction method based on convolutional transformer architecture, and system and device | |
| CN110222991B (en) | Metering device fault diagnosis method based on RF-GBDT | |
| CN107807860B (en) | Power failure analysis method and system based on matrix decomposition | |
| CN109522948A (en) | A kind of fault detection method based on orthogonal locality preserving projections | |
| CN105303331A (en) | Transformer repair risk decision-making method | |
| CN108510147B (en) | Electric energy quality comprehensive evaluation method based on residual error fluctuation model | |
| US20220113354A1 (en) | Data processing apparatus, data processing method and computer readable medium | |
| CN116341272A (en) | Construction safety risk management and control system for digital distribution network engineering | |
| Amiri et al. | Faults detection and diagnosis of PV systems based on machine learning approach using random forest classifier | |
| CN115908051A (en) | Method for determining energy storage capacity of power system | |
| Livera et al. | Optimal development of location and technology independent machine learning photovoltaic performance predictive models | |
| CN117554821A (en) | Energy storage battery on-line monitoring management method and system based on deep learning network | |
| Dong et al. | Prognostics 102: efficient Bayesian-based prognostics algorithm in Matlab | |
| Dong et al. | Fault diagnosis and classification in photovoltaic systems using scada data | |
| CN117195451A (en) | Bridge monitoring data restoration method based on graph theory | |
| Wang et al. | An anomaly detection method for spacecraft solar arrays based on the ILS-SVM model | |
| Sridevi et al. | Predictive Maintenance of Lead-Acid Batteries Using Machine Learning Algorithms | |
| Bhattacharyya et al. | Adaptive, online models to detect and estimate gross error in SPNDs | |
| Bashkari et al. | Distribution power system outage diagnosis based on root cause analysis | |
| JP7177238B1 (en) | FAILURE DETERMINATION SYSTEM, FAILURE DETERMINATION PROGRAM AND FAILURE DETERMINATION METHOD |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |