A satellite telemetry data outlier detection method, device and electronic equipment

Abstract

This application provides a method, apparatus, and electronic device for detecting outliers in satellite telemetry data. The outlier detection model for telemetry data is trained using a kernel extreme learning machine, which can effectively detect outliers in satellite telemetry data, greatly improving the efficiency and accuracy of outlier detection.

Description

Technical Field

[0001] This application relates to the field of satellite telemetry data processing technology, and more specifically, to a method, apparatus, and electronic device for detecting outliers in satellite telemetry data. Background Technology

[0002] Satellites are constantly exposed to harsh space radiation environments. High-energy particles can easily induce single-event latch-up (SEL) in complementary metal-oxide-semiconductor (CMOS) systems, resulting in extremely high or low values ​​in telemetry data, known as outliers. Furthermore, factors such as mechanical component damage, electronic circuit aging, sensor performance degradation, and environmental interference can all contribute to outlier data in telemetry. Therefore, designing effective on-orbit outlier detection algorithms is crucial.

[0003] Outlier detection methods for telemetry data primarily rely on expert experience. However, outliers are generally complex and diverse in harsh environments, making them difficult to simulate in advance, resulting in low accuracy of outlier detection methods for telemetry data. Summary of the Invention

[0004] To address the aforementioned problems, the purpose of this application is to provide a method, apparatus, and electronic device for detecting outliers in satellite telemetry data.

[0005] In a first aspect, embodiments of this application provide a method for detecting outliers in satellite telemetry data, including:

[0006] Obtain historical telemetry data of the kth telemetry type, wherein the historical telemetry data of the kth telemetry type includes: a first part of historical data, a second part of historical data, and a third part of historical data; the first part of historical data, the second part of historical data, and the third part of historical data all include: normal values ​​and outliers;

[0007] From the telemetry data of the k-1 telemetry types that have been detected, determine the telemetry type associated with the k-th telemetry type, and input the normal values ​​in the historical telemetry data of the telemetry type associated with the k-th telemetry type and the first part of the historical data into the kernel extreme learning machine to train the kernel extreme learning machine and obtain the outlier detection model of the telemetry data of the k-th telemetry type.

[0008] The second part of historical data is input into the outlier detection model of the telemetry data of the k-th telemetry type. The outlier detection model of the telemetry data of the k-th telemetry type processes the second part of historical data to obtain multiple predicted values ​​of the k-th telemetry type. Each historical telemetry data in the third part of historical data corresponds one-to-one with each of the multiple predicted values ​​of the k-th telemetry type. The third part of historical data that corresponds one-to-one with each of the multiple predicted values ​​is called the true value of historical data.

[0009] Obtain the first normal value correlation coefficient and the second normal value correlation coefficient. Based on each predicted value among the multiple predicted values, the first normal value correlation coefficient, and the second normal value correlation coefficient, determine the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the k-th telemetry type. Based on the normal value range, optimize the first normal value correlation coefficient and the second normal value correlation coefficient to obtain the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient.

[0010] When the telemetry data to be detected is obtained, it is determined whether the telemetry data to be detected is an outlier based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient.

[0011] Secondly, embodiments of this application also provide a satellite telemetry data outlier detection device, comprising:

[0012] The acquisition module is used to acquire historical telemetry data of the kth telemetry type. The historical telemetry data of the kth telemetry type includes: a first part of historical data, a second part of historical data, and a third part of historical data. The first part of historical data, the second part of historical data, and the third part of historical data all include: normal values ​​and outliers.

[0013] The training module is used to determine the telemetry type associated with the k-th telemetry type from the telemetry data of the k-1 telemetry types that have been detected, and input the normal values ​​in the historical telemetry data of the telemetry type associated with the k-th telemetry type and the first part of historical data into the kernel extreme learning machine to train the kernel extreme learning machine and obtain the outlier detection model of the telemetry data of the k-th telemetry type.

[0014] The first processing module is used to input the second part of historical data into the outlier detection model of the telemetry data of the k-th telemetry type, and process the second part of historical data through the outlier detection model of the telemetry data of the k-th telemetry type to obtain multiple predicted values ​​of the k-th telemetry type; wherein, each historical telemetry data in the third part of historical data corresponds one-to-one with the multiple predicted values ​​of the k-th telemetry type; the third part of historical data corresponding one-to-one with each of the multiple predicted values ​​is called the true value of historical data;

[0015] The second processing module is used to obtain the first normal value correlation coefficient and the second normal value correlation coefficient, and based on each predicted value among the multiple predicted values, the first normal value correlation coefficient and the second normal value correlation coefficient, determine the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the k-th telemetry type, and optimize the first normal value correlation coefficient and the second normal value correlation coefficient according to the normal value range to obtain the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient.

[0016] The third processing module is used to determine whether the telemetry data to be detected is an outlier based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient when the telemetry data to be detected is acquired.

[0017] Thirdly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method described in the first aspect above.

[0018] Fourthly, embodiments of this application also provide an electronic device, the electronic device including a memory, a processor and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor using the steps of the method described in the first aspect above.

[0019] In the solutions provided by the first to fourth aspects of this application, the normal values ​​in the telemetry data of the telemetry type associated with the k-th telemetry type and the first part of the historical data in the historical telemetry data of the k-th telemetry type are input into a kernel extreme learning machine to train the kernel extreme learning machine, thereby obtaining an outlier detection model for the telemetry data of the k-th telemetry type. This outlier detection model is then used to determine outliers in the telemetry data of the k-th telemetry type to be detected. Compared with related technologies that rely on expert experience for outlier detection, the outlier detection model obtained by training the telemetry data using a kernel extreme learning machine can effectively detect outliers in the satellite's telemetry data, greatly improving the efficiency and accuracy of outlier detection.

[0020] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 A flowchart of a satellite telemetry data outlier detection method provided in Embodiment 1 of this application is shown;

[0023] Figure 2 This paper shows a schematic diagram of the structure of a satellite telemetry data outlier detection device provided in Embodiment 2 of this application;

[0024] Figure 3 A schematic diagram of the structure of an electronic device provided in Embodiment 3 of this application is shown. Detailed Implementation

[0025] In the description of this application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0026] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0027] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0028] Satellites are constantly exposed to harsh space radiation environments. High-energy particles can easily induce single-event latch-up (SEL) in complementary metal-oxide-semiconductor (CMOS) systems, resulting in extremely high or low values ​​in telemetry data, known as outliers. Furthermore, factors such as mechanical component damage, electronic circuit aging, sensor performance degradation, and environmental interference can all contribute to outlier data in telemetry. Therefore, designing effective on-orbit outlier detection algorithms is crucial.

[0029] Currently, outlier detection methods for telemetry data are mainly divided into three types: threshold-based, model-based, and data-driven. Threshold-based and model-based methods rely on existing expert experience. However, outliers are generally complex and diverse in harsh environments, making them difficult to simulate in advance. Furthermore, it is difficult to establish physical models for every satellite subsystem, especially given the high complexity and strong coupling characteristics of satellites. Therefore, the detection performance of these two methods is very limited. With the advent of the information age and the rapid development of intelligent technologies such as data mining and machine learning, data-driven outlier detection methods have gradually become a research hotspot in the field of telemetry outlier detection in recent years. This invention, based on kernel extreme learning machines, designs a method and system for outlier detection in satellite telemetry data, which can effectively detect outliers in satellite telemetry data.

[0030] Based on this, this embodiment proposes a method, apparatus, and electronic device for detecting outliers in satellite telemetry data. The method involves inputting normal values ​​from telemetry data associated with the k-th telemetry type and a first portion of historical data from the historical telemetry data of the k-th telemetry type into a kernel extreme learning machine (KELM). The KELM is then trained to obtain an outlier detection model for the k-th telemetry type. This model is then used to determine outliers in the telemetry data of the k-th telemetry type to be detected. The outlier detection model obtained by training the KELM effectively detects outliers in satellite telemetry data, significantly improving the efficiency and accuracy of outlier detection.

[0031] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0032] Example 1

[0033] The execution entity of the satellite telemetry data outlier detection method proposed in this embodiment is a server, which can be a computing device set on the satellite, so that it operates in space along with the satellite.

[0034] See Figure 1 The flowchart shown illustrates a method for detecting outliers in satellite telemetry data. The method proposed in this embodiment includes the following specific steps:

[0035] Step 100: Obtain historical telemetry data of the kth telemetry type. The historical telemetry data of the kth telemetry type includes: a first part of historical data, a second part of historical data, and a third part of historical data. The first part of historical data, the second part of historical data, and the third part of historical data all include: normal values ​​and outliers.

[0036] In step 100 above, the historical telemetry data is pre-cached in the server.

[0037] The first part of the historical data, the second part of the historical data, and the third part of the historical telemetry data each have a different number of historical telemetry data.

[0038] The historical telemetry data includes: normal values, outliers, and abnormal values.

[0039] The normal, outlier, and abnormal values ​​in the historical telemetry data were all manually labeled.

[0040] The telemetry types include, but are not limited to: battery temperature type, battery voltage type, bus voltage type, load current type, solar panel temperature type, solar array current type, and power controller temperature type.

[0041] In one implementation, when the kth telemetry type is a battery temperature type, the historical telemetry data of the kth telemetry type is the historical battery temperature; the telemetry data to be detected of the kth telemetry type is the battery temperature to be detected.

[0042] The server also stores the association relationships of telemetry types, linking telemetry types that may have mutual influence.

[0043] In one embodiment, the battery temperature type, the battery voltage type, the bus voltage type, the load current type, the solar panel temperature type, the solar array current type, and the power controller temperature type are interrelated, meaning that these three types are interconnected and mutually influential telemetry types.

[0044] Step 102: From the telemetry data of the k-1 telemetry types that have been detected, determine the telemetry type associated with the k-th telemetry type, and input the normal values ​​in the historical telemetry data of the telemetry type associated with the k-th telemetry type and the first part of the historical data into the kernel extreme learning machine to train the kernel extreme learning machine and obtain the outlier detection model of the telemetry data of the k-th telemetry type.

[0045] In step 102 above, the normal values ​​from the historical telemetry data associated with the k-th telemetry type and the first part of the historical data from the historical telemetry data of the k-th telemetry type are input into the kernel extreme learning machine. This means that the normal values ​​from the historical telemetry data associated with the k-th telemetry type are used to assist the training of the kernel extreme learning machine with the historical telemetry data of the k-th telemetry type, resulting in an outlier detection model for the telemetry data of the k-th telemetry type. This method achieves higher outlier detection accuracy than training the kernel extreme learning machine solely with the historical telemetry data of the k-th telemetry type.

[0046] The specific process of inputting the normal values ​​from the historical telemetry data of the telemetry type associated with the k-th telemetry type and the first part of the historical data into the kernel extreme learning machine, and training the kernel extreme learning machine to obtain the outlier detection model of the telemetry data of the k-th telemetry type is existing technology and will not be described in detail here.

[0047] The parameters of the outlier detection model for the telemetry data of the k-th telemetry type obtained through training include, but are not limited to, the output weight matrix.

[0048] Step 104: Input the second part of historical data into the outlier detection model of the telemetry data of the kth telemetry type, and process the second part of historical data through the outlier detection model of the telemetry data of the kth telemetry type to obtain multiple predicted values ​​of the kth telemetry type; wherein, each historical telemetry data in the third part of historical data corresponds one-to-one with the multiple predicted values ​​of the kth telemetry type; the third part of historical data that corresponds one-to-one with each of the multiple predicted values ​​is called the true value of historical data.

[0049] In step 104 above, the outlier detection model of the telemetry data of the kth telemetry type processes the second part of historical data to obtain multiple predicted values ​​of the kth telemetry type. The specific process is existing technology and will not be described in detail here.

[0050] For example, if the number of telemetry data to be detected is 100, and 90 of the telemetry data are input into the outlier detection model of the k-th telemetry type, then the true values ​​in the outlier detection model that are not input into the k-th telemetry type are 10.

[0051] Accordingly, after the outlier detection model of the k-th telemetry type processes the partial telemetry data, it can obtain 10 predicted values. Each of these 10 predicted values ​​corresponds one-to-one with the 10 predicted values ​​that were not input into the outlier detection model of the k-th telemetry type.

[0052] In the example above, the outlier detection model for the telemetry data of the kth telemetry type can predict any number of values; 10 predicted values ​​are just an example.

[0053] After obtaining multiple predicted values ​​for the kth telemetry type through step 104 above, the following steps (1) to (2) can be continued:

[0054] (1) In order for the error between the predicted value and the corresponding true value to approach 0, the output weight matrix satisfies the following formula:

[0055]

[0056]

[0057] Where, β ★ This represents the optimal output weight matrix where the error between the predicted value and the corresponding true value approaches 0; C represents the regularization parameter; H represents the optimal output weight matrix where the error between the predicted value and the true value approaches 0. T H represents the transpose of the feature mapping matrix from the input layer to the hidden layer in the outlier detection model for the k-th telemetry type; I represents the identity matrix; Y represents the sequence of historical data true values ​​composed of multiple historical data true values; X represents the second part of historical data input into the outlier detection model for the k-th telemetry type; X i This represents the i-th telemetry data point in the second part of the historical data; η represents the offset degree, which is a constant. It represents a length scale and is a constant.

[0058] (2) Using the obtained β ★ The outlier detection model for the telemetry data of the k-th telemetry type is optimized.

[0059] In step (1) above, in order to make the error between the predicted value and the corresponding true value approach 0, the following Lagrange function is first constructed:

[0060]

[0061] Where L represents the number of hidden nodes; K represents the number of telemetry types; N represents the amount of telemetry data for each telemetry type; ζ k α represents the error between the predicted value of the k-th telemetry type and the corresponding historical data value; ki h represents the Lagrange multiplier of the i-th telemetry data of the k-th telemetry type; jk X represents the j-th row of the feature mapping matrix from the input layer to the hidden layer of the outlier detection model for the k-th telemetry type; k This represents the second part of the historical data for the k-th telemetry type; β jk Y represents the j-th output weight value of the outlier detection model for the k-th telemetry type; ki This represents the actual historical data value corresponding to the i-th predicted value of the k-th telemetry type; ζ ki This represents the error between the i-th predicted value of the k-th telemetry type and the corresponding i-th historical data value. This represents the Lagrange function.

[0062] From the above formula, we can see that due to C, η and Included in β ★ In the middle, C, η and Is with β ★ The relevant parameters; then, after obtaining β ★ Based on this, using β ★ Replacing β in the above Lagrange function allows us to adjust the regularization parameters C, η, and ... Optimize.

[0063] Using mean squared error (MSE) as the evaluation metric for learning performance, a new convex optimization problem is constructed, as shown in the following formula:

[0064]

[0065] Where L represents the number of hidden nodes; g(x) represents the activation function; w (k)l and b (k)l X represents the input weight and bias of the l-th hidden node in the outlier detection model for the k-th telemetry type, respectively; (k)i Y represents the sequence of historical telemetry data from the second historical data used to obtain the i-th prediction value of the k-th telemetry type; (k)i This represents the true value of the i-th historical data of the k-th telemetry type.

[0066] Here, the historical remote sensing data sequence includes at least two historical telemetry data from the second historical data.

[0067] After going through steps (1) to (2) above, using the obtained β ★ C, η and After optimizing the outlier detection model for the telemetry data of the kth telemetry type, the following step 106 can be performed to determine the normal value range of the telemetry data of the kth telemetry type.

[0068] To ensure the accuracy of the normal value range determined in step 106, the normal value range can be optimized first, and then the determination of whether the telemetry data to be detected are outliers can be made. Therefore, a value range optimization index S is constructed in step 106. 2 To evaluate the interval construction performance. And through step 106 and S 2 The relevant formulas are used to optimize the correlation coefficients of the first normal value and the second normal value, respectively.

[0069] Step 106: Obtain the first normal value correlation coefficient and the second normal value correlation coefficient. Based on each predicted value among the multiple predicted values, the first normal value correlation coefficient, and the second normal value correlation coefficient, determine the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the k-th telemetry type. Based on the normal value range, optimize the first normal value correlation coefficient and the second normal value correlation coefficient to obtain the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient.

[0070] In step 106 above, the correlation coefficients of the first normal value before optimization and the correlation coefficients of the second normal value before optimization are pre-cached in the server.

[0071] In one implementation, the correlation coefficients of the first normal value before optimization and the second normal value before optimization can be set to 1.

[0072] Specifically, obtaining the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient includes:

[0073] The upper and lower limits of the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the kth telemetry type are calculated using the following formulas.

[0074]

[0075] in, This represents the upper limit of the normal value range for the i-th telemetry data of the k-th telemetry type; This represents the lower limit of the normal value range for the i-th telemetry data of the k-th telemetry type; u (k) ω represents the correlation coefficient of the first normal value; (k) This represents the correlation coefficient of the second normal value; This represents the i-th predicted value among multiple predicted values;

[0076] To ensure that the outlier detection rate of the predicted values ​​approaches 1 and the false alarm rate of the predicted values ​​approaches 0, the correlation coefficients of the first and second normal values ​​are optimized using the following formula:

[0077] S 2 = (1+σ)·W·(1+γ(p)·e -τ(p-μ) )

[0078]

[0079]

[0080]

[0081] Among them, S 2 The value range optimization index is represented by γ(p); the Heaviside function is represented by τ; the first constant is represented by μ; the second constant is represented by N; and N represents the number of telemetry data points in the outlier detection model of the k-th telemetry type. Then c i =1, Then c i =0; y (k)i Represents the i-th true value of the k-th telemetry type; R represents the difference between the maximum and minimum values ​​in the partial telemetry data of the outlier detection model input to the k-th telemetry type; if Then E i =0; if but

[0082] Step 108: When the telemetry data to be detected is obtained, determine whether the telemetry data to be detected is an outlier based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient.

[0083] In step 108 above, in order to determine whether the telemetry data to be detected is an outlier based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient, step 108 can specifically perform the following steps (1) to (3):

[0084] (1) When the telemetry data to be detected is obtained, a preset number of telemetry data of the kth telemetry type obtained before the telemetry data to be detected are input into the outlier detection model of the telemetry data of the kth telemetry type to obtain the prediction value corresponding to the telemetry data to be detected.

[0085] (2) Based on the predicted value, the optimized first normal value correlation coefficient, and the optimized second normal value correlation coefficient, the normal value range of the telemetry data to be detected is obtained;

[0086] (3) When the telemetry data to be detected is not within the normal value range of the telemetry data to be detected, the acquired telemetry data is determined to be an outlier.

[0087] In step (1) above, the telemetry data to be detected is the telemetry data obtained by the server in real time.

[0088] A preset number of telemetry data of the kth telemetry type acquired before the telemetry data to be detected are all cached in the server.

[0089] The telemetry data of the kth telemetry type mentioned above are cached in the server according to the order in which they were acquired by the server.

[0090] Therefore, when acquiring a preset number of telemetry data of the kth telemetry type, the server, according to the time when acquiring the telemetry data to be detected, inputs a preset number of telemetry data of the kth telemetry type with an acquisition time close to the time when acquiring the telemetry data to be detected into the outlier detection model of the telemetry data of the kth telemetry type. The outlier detection model of the kth telemetry type processes the preset number of telemetry data of the kth telemetry type to obtain the predicted value corresponding to the telemetry data to be detected.

[0091] In step (2) above, the specific process of obtaining the normal value range of the telemetry data to be detected based on the predicted value, the optimized first normal value correlation coefficient, and the optimized second normal value correlation coefficient is similar to the process in step 106 above of calculating the upper limit and lower limit of the normal value range of the historical data true value corresponding to each predicted value in the telemetry data of the kth telemetry type. It will not be described again here.

[0092] For example, the detection performance of the satellite telemetry data outlier detection method proposed in this embodiment was tested on telemetry data from the actual power supply and distribution system of a smart communication satellite launched by the applicant. Seven typical variables were selected based on their actual physical meaning: battery temperature, battery voltage, bus voltage, load current, solar panel temperature, solar array current, and power controller temperature. The dataset was collected from January 1, 2021 to January 31, 2021, with 60% of the data used for network training, 20% for validation, and the remaining 20% ​​used to validate the algorithm's performance. The iteration ended when the error between two iterations was less than 0.001, with a maximum of 500 iterations. Since BP neural networks are widely used in various scenarios and exhibit good performance, they were used as a traditional data-driven detection method for comparison. Some model-based detection methods were also used for comparison. The BP neural network contained a fully connected hidden layer with 50 neurons, and the activation function was set to the Tanh function. The learning rate was set to 0.5, and the momentum factor was set to 0.6. The iteration termination condition was the same as that of the kernel extreme learning machine.

[0093] A comparison of the predicted and true values ​​of the seven typical telemetry variables obtained through the satellite telemetry data outlier detection method proposed in this embodiment reveals that, except for a few discontinuities, the kernel extreme learning machine can track the original telemetry data in most cases. The prediction results are listed in Table 1, where RMSEt, RMSEv, and RMSEr represent the RMSE during the training, validation, and testing phases, respectively. It is readily apparent that the proposed method exhibits greater advantages in modeling and generalization capabilities.

[0094] Table 1. Short-term prediction results of satellite power supply and distribution system telemetry data using different data-driven methods.

[0095]

[0096] For most telemetry variables, outlier detection using traditional model-based methods is constrained by the true positive rate (TPR) and false positive rate (FPR). For example, setting relaxed boundary conditions can improve the TPR in outlier detection, resulting in better compression performance, but the FPR is unsatisfactory, as shown in Table 2. Here, TPR refers to the proportion of actual health data in the predicted health data; FPR refers to the proportion of outliers in the predicted health data. In fact, improving compression performance and maintaining a low FPR are equally important. This is because unhealthy telemetry data (non-outliers) carries critical information needed by ground stations to monitor satellite operations. In our algorithm, the dynamic nature of the intervals ensures that outliers for almost all telemetry variables can be effectively detected, while maintaining a low FPR. In particular, for data near discontinuities, forward dynamic interval construction combined with inverse verification avoids treating these telemetry data as outliers, further reducing the FPR of outlier detection.

[0097] Table 2. Results of Dynamic Range Construction for Telemetry Data of Satellite Power Supply and Distribution System

[0098]

[0099] In summary, this embodiment proposes a method for detecting outliers in satellite telemetry data. The method involves inputting normal values ​​from telemetry data associated with the k-th telemetry type and historical telemetry data of the k-th telemetry type into a kernel extreme learning machine (KELM). The KELM is then trained to obtain an outlier detection model for the k-th telemetry type. This model is then used to determine outliers in the telemetry data of the k-th telemetry type to be detected. Compared to related technologies that rely on expert experience for outlier detection, the outlier detection model trained using a KELM effectively detects outliers in satellite telemetry data, significantly improving both the efficiency and accuracy of outlier detection.

[0100] Example 2

[0101] This embodiment proposes a satellite telemetry data outlier detection device, which is used to perform the satellite telemetry data outlier detection method proposed in Embodiment 1 above.

[0102] See Figure 2 The diagram shows a structural schematic of a satellite telemetry data outlier detection device. This embodiment proposes a satellite telemetry data outlier detection device, comprising:

[0103] The acquisition module 200 is used to acquire historical telemetry data of the kth telemetry type. The historical telemetry data of the kth telemetry type includes: a first part of historical data, a second part of historical data, and a third part of historical data. The first part of historical data, the second part of historical data, and the third part of historical data all include: normal values ​​and outliers.

[0104] Training module 202 is used to determine the telemetry type associated with the k-th telemetry type from the telemetry data of the detected k-1 telemetry types, and input the normal values ​​in the historical telemetry data of the telemetry type associated with the k-th telemetry type and the first part of historical data into the kernel extreme learning machine to train the kernel extreme learning machine and obtain the outlier detection model of the telemetry data of the k-th telemetry type.

[0105] The first processing module 204 is used to input the second part of historical data into the outlier detection model of the telemetry data of the k-th telemetry type, and process the second part of historical data through the outlier detection model of the telemetry data of the k-th telemetry type to obtain multiple predicted values ​​of the k-th telemetry type; wherein, each historical telemetry data in the third part of historical data corresponds one-to-one with the multiple predicted values ​​of the k-th telemetry type; the third part of historical data that corresponds one-to-one with each of the multiple predicted values ​​is called the true value of historical data;

[0106] The second processing module 206 is used to obtain the first normal value correlation coefficient and the second normal value correlation coefficient, and based on each predicted value among the multiple predicted values, the first normal value correlation coefficient and the second normal value correlation coefficient, determine the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the k-th telemetry type, and optimize the first normal value correlation coefficient and the second normal value correlation coefficient according to the normal value range to obtain the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient.

[0107] The third processing module 208 is used to determine whether the telemetry data to be detected is an outlier based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient when the telemetry data to be detected is acquired.

[0108] Specifically, the second processing module is used to obtain a first normal value correlation coefficient and a second normal value correlation coefficient; determine the normal value range of the historical data true values ​​corresponding to each of the predicted values ​​in the telemetry data of the k-th telemetry type, based on each predicted value, the first normal value correlation coefficient, and the second normal value correlation coefficient; and optimize the first normal value correlation coefficient and the second normal value correlation coefficient based on the normal value range, including:

[0109] The upper and lower limits of the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the kth telemetry type are calculated using the following formulas.

[0110]

[0111] in, This represents the upper limit of the normal value range for the i-th telemetry data of the k-th telemetry type; This represents the lower limit of the normal value range for the i-th telemetry data of the k-th telemetry type; u (k) ω represents the correlation coefficient of the first normal value; (k) This represents the correlation coefficient of the second normal value; This represents the i-th predicted value among multiple predicted values;

[0112] To ensure that the outlier detection rate of the predicted values ​​approaches 1 and the false alarm rate of the predicted values ​​approaches 0, the correlation coefficients of the first and second normal values ​​are optimized using the following formula:

[0113] S 2 = (1+σ)·W·(1+γ(p)·e -τ(p-μ) )

[0114]

[0115]

[0116]

[0117] Among them, S 2 The value range optimization index is represented by γ(p); the Heaviside function is represented by τ; the first constant is represented by μ; the second constant is represented by N; and N represents the number of telemetry data points in the outlier detection model of the k-th telemetry type. Then c i =1, Then c i =0; y (k)i Represents the i-th true value of the k-th telemetry type; R represents the difference between the maximum and minimum values ​​in the partial telemetry data of the outlier detection model input to the k-th telemetry type; if Then E i =0; if but

[0118] The parameters of the outlier detection model for the telemetry data of the kth telemetry type include: the output weight matrix;

[0119] Furthermore, the training module is specifically used for:

[0120] To ensure that the error between the predicted value and the corresponding true value approaches zero, the regularization parameter and the output weight matrix satisfy the following formula:

[0121]

[0122]

[0123] Where, β ★ This represents the optimal output weight matrix where the error between the predicted value and the corresponding true value approaches 0; C represents the regularization parameter; H represents the optimal output weight matrix where the error between the predicted value and the true value approaches 0. T I represents the transpose of the feature mapping matrix from the input layer to the hidden layer in the outlier detection model of the k-th telemetry type; Y represents the sequence of historical data true values ​​composed of multiple historical data true values; X represents the second part of historical data input into the outlier detection model of the k-th telemetry type; X i This represents the i-th telemetry data point in the second part of the historical data; η represents the offset degree, which is a constant. It represents a length scale and is a constant.

[0124] Using the obtained β ★ The outlier detection model for the telemetry data of the k-th telemetry type is optimized.

[0125] Specifically, the third processing module is used for:

[0126] When the telemetry data to be detected is obtained, a preset number of telemetry data of the kth telemetry type obtained before the telemetry data to be detected are input into the outlier detection model of the telemetry data of the kth telemetry type to obtain the predicted value corresponding to the telemetry data to be detected.

[0127] Based on the predicted value, the optimized first normal value correlation coefficient, and the optimized second normal value correlation coefficient, the normal value range of the telemetry data to be detected is obtained;

[0128] When the telemetry data to be detected is not within the normal range of telemetry data values, the acquired telemetry data is determined to be an outlier.

[0129] In summary, this embodiment proposes a satellite telemetry data outlier detection device. It involves inputting normal values ​​from telemetry data associated with the k-th telemetry type and historical telemetry data of the k-th telemetry type into a kernel extreme learning machine (KELM). The KELM is then trained to obtain an outlier detection model for the k-th telemetry type. This model is then used to process a portion of the telemetry data to be detected for the k-th telemetry type, yielding multiple predicted values ​​for that type. Each of these predicted values ​​is then judged as an outlier. Compared to related technologies that rely on expert experience for outlier detection, the outlier detection model trained using a KELM effectively detects outliers in satellite telemetry data, significantly improving both the efficiency and accuracy of outlier detection.

[0130] Example 3

[0131] This embodiment proposes a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it executes the steps of the satellite telemetry data outlier detection method described in Embodiment 1 above. For specific implementation details, please refer to Method Embodiment 1, which will not be repeated here.

[0132] In addition, see Figure 3The diagram shows the structure of an electronic device. This embodiment also proposes an electronic device, which includes a bus 51, a processor 52, a transceiver 53, a bus interface 54, a memory 55, and a user interface 56. The electronic device includes a memory 55.

[0133] In this embodiment, the electronic device further includes: one or more programs stored in the memory 55 and executable on the processor 52, configured to be executed by the processor to perform the one or more programs for steps (1) to (5):

[0134] (1) Obtain historical telemetry data of the kth telemetry type, wherein the historical telemetry data of the kth telemetry type includes: a first part of historical data, a second part of historical data and a third part of historical data; the first part of historical data, the second part of historical data and the third part of historical data each include: normal values ​​and outliers;

[0135] (2) From the telemetry data of the k-1 telemetry types that have been detected, determine the telemetry type associated with the k-th telemetry type, and input the normal value in the historical telemetry data of the telemetry type associated with the k-th telemetry type and the first part of the historical data into the kernel extreme learning machine, train the kernel extreme learning machine, and obtain the outlier detection model of the telemetry data of the k-th telemetry type.

[0136] (3) Input the second part of historical data into the outlier detection model of the telemetry data of the kth telemetry type, and process the second part of historical data through the outlier detection model of the telemetry data of the kth telemetry type to obtain multiple predicted values ​​of the kth telemetry type; wherein, each historical telemetry data in the third part of historical data corresponds one-to-one with the multiple predicted values ​​of the kth telemetry type; the third part of historical data that corresponds one-to-one with each of the multiple predicted values ​​is called the true value of historical data;

[0137] (4) Obtain the first normal value correlation coefficient and the second normal value correlation coefficient. Based on each predicted value among the multiple predicted values, the first normal value correlation coefficient and the second normal value correlation coefficient, determine the normal value range of the historical data true value corresponding to each predicted value in the telemetry data of the kth telemetry type. Based on the normal value range, optimize the first normal value correlation coefficient and the second normal value correlation coefficient to obtain the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient.

[0138] (5) When the telemetry data to be detected is obtained, the telemetry data to be detected is judged as an outlier based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient.

[0139] Transceiver 53 is used to receive and send data under the control of processor 52.

[0140] The bus architecture (represented by bus 51) can include any number of interconnected buses and bridges, linking various circuits including one or more processors represented by processor 52 and memory represented by memory 55. Bus 51 can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be further described in this embodiment. Bus interface 54 provides an interface between bus 51 and transceiver 53. Transceiver 53 can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. For example, transceiver 53 receives external data from other devices. Transceiver 53 is used to transmit data processed by processor 52 to other devices. Depending on the nature of the computing system, a user interface 56 may also be provided, such as a keypad, display, speaker, microphone, or joystick.

[0141] Processor 52 is responsible for managing bus 51 and general processing, such as running general-purpose operating system 551 as described above. Memory 55 can be used to store data used by processor 52 during operation.

[0142] Optionally, the processor 52 may be, but is not limited to, a central processing unit, a microcontroller, a microprocessor, or a programmable logic device.

[0143] It is understood that the memory 55 in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM). The memory 55 of the systems and methods described in this embodiment is intended to include, but is not limited to, these and any other suitable types of memory.

[0144] In some implementations, memory 55 stores elements such as executable modules or data structures, or subsets thereof, or extended sets thereof: operating system 551 and application programs 552.

[0145] The operating system 551 includes various system programs, such as the framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. The application program 552 includes various applications, such as a media player and a browser, used to implement various application functions. Programs implementing the methods of the embodiments of this application can be included in the application program 552.

[0146] In summary, this embodiment proposes a computer-readable storage medium and electronic device. By inputting normal values ​​from telemetry data associated with the k-th telemetry type and historical telemetry data of the k-th telemetry type into a kernel extreme learning machine (KELM), the KELM is trained to obtain an outlier detection model for the k-th telemetry type's telemetry data. Using this outlier detection model, a portion of the telemetry data to be detected in the k-th telemetry type is processed to obtain multiple predicted values ​​for the k-th telemetry type. Outlier judgment is then performed on each of these predicted values. Compared to related technologies that rely on expert experience for outlier detection, the outlier detection model trained using a KELM can effectively detect outliers in satellite telemetry data, significantly improving the efficiency and accuracy of outlier detection.

[0147] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting outliers in satellite telemetry data, characterized in that, include: Obtain historical telemetry data of the kth telemetry type, wherein the historical telemetry data of the kth telemetry type includes: a first part of historical data, a second part of historical data, and a third part of historical data; the first part of historical data, the second part of historical data, and the third part of historical data all include: normal values ​​and outliers; The second part of historical data is input into the outlier detection model of the telemetry data of the k-th telemetry type. The outlier detection model of the telemetry data of the k-th telemetry type processes the second part of historical data to obtain multiple predicted values ​​of the k-th telemetry type. Each historical telemetry data in the third part of historical data corresponds one-to-one with each of the multiple predicted values ​​of the k-th telemetry type. The third part of historical data that corresponds one-to-one with each of the multiple predicted values ​​is called the true value of historical data. Obtain the first normal value correlation coefficient and the second normal value correlation coefficient. Based on each predicted value among the multiple predicted values, the first normal value correlation coefficient, and the second normal value correlation coefficient, determine the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the k-th telemetry type. Based on the normal value range, optimize the first normal value correlation coefficient and the second normal value correlation coefficient to obtain the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient. When the telemetry data to be detected is obtained, it is determined whether the telemetry data to be detected is an outlier based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient. The process of obtaining the first normal value correlation coefficient and the second normal value correlation coefficient, and determining the normal value range of the historical data true values ​​corresponding to each of the predicted values ​​in the telemetry data of the k-th telemetry type based on each predicted value, the first normal value correlation coefficient, and the second normal value correlation coefficient, and optimizing the first normal value correlation coefficient and the second normal value correlation coefficient based on the normal value range, includes: The upper and lower limits of the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the kth telemetry type are calculated using the following formulas. ； in, Represents the k-th telemetry type. The upper limit of the normal value range for each telemetry data point; Represents the k-th telemetry type. The lower limit of the normal value range for each telemetry data point; This represents the correlation coefficient of the first normal value; This represents the correlation coefficient of the second normal value; Represents the first of multiple predicted values One predicted value; To ensure that the outlier detection rate of the predicted values ​​approaches 1 and the false alarm rate of the predicted values ​​approaches 0, the correlation coefficients of the first and second normal values ​​are optimized using the following formula: ； ； ； ； in, Indicates the optimization index for the range of values; This represents the Heaviside function; Denotes the first constant; Indicates the second constant; This represents the number of telemetry data points in the outlier detection model of the k-th telemetry type; if ,but , ; Indicates the k-th telemetry type. One true value; This represents the difference between the maximum and minimum values ​​in a subset of telemetry data input into the outlier detection model for the k-th telemetry type; if ,but ;like ,but .

2. The method according to claim 1, characterized in that, The parameters of the outlier detection model for the telemetry data of the kth telemetry type include: the output weight matrix; Before the steps of obtaining the first normal value correlation coefficient and the second normal value correlation coefficient, determining the normal value range of the historical data true values ​​corresponding to each of the predicted values ​​in the telemetry data of the k-th telemetry type based on each predicted value, the first normal value correlation coefficient, and the second normal value correlation coefficient, and optimizing the first normal value correlation coefficient and the second normal value correlation coefficient based on the normal value range to obtain the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient, the method further includes: To ensure that the error between the predicted value and the corresponding true value approaches zero, the output weight matrix satisfies the following formula: ； ； in, This represents the optimal output weight matrix when the error between the predicted value and the corresponding true value approaches 0. Represents the regularization parameter; This represents the transpose of the feature mapping matrix from the input layer to the hidden layer in the outlier detection model for the k-th telemetry type. Represents the identity matrix; This represents a sequence of historical data true values, composed of multiple historical data true values. This represents the second part of historical data in the outlier detection model input into the telemetry data of the kth telemetry type; This indicates the first in the second part of the historical data. One telemetry data; Indicates the degree of offset, which is a constant; It represents a length scale and is a constant. Utilize The outlier detection model for the telemetry data of the k-th telemetry type is optimized.

3. The method according to claim 2, characterized in that, When the telemetry data to be detected is acquired, the determination of whether the telemetry data to be detected is an outlier is made based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient, including: When the telemetry data to be detected is obtained, a preset number of telemetry data of the kth telemetry type obtained before the telemetry data to be detected are input into the outlier detection model of the telemetry data of the kth telemetry type to obtain the predicted value corresponding to the telemetry data to be detected. Based on the predicted value, the optimized first normal value correlation coefficient, and the optimized second normal value correlation coefficient, the normal value range of the telemetry data to be detected is obtained; When the telemetry data to be detected is not within the normal range of telemetry data values, the acquired telemetry data is determined to be an outlier.

4. The method according to claim 1, characterized in that, Also includes: From the telemetry data of the k-1 telemetry types that have been detected, the telemetry type associated with the k-th telemetry type is determined, and the normal values ​​in the historical telemetry data of the telemetry type associated with the k-th telemetry type and the first part of the historical data are input into the kernel extreme learning machine to train the kernel extreme learning machine and obtain the outlier detection model of the telemetry data of the k-th telemetry type.

5. A satellite telemetry data outlier detection device, characterized in that, include: The acquisition module is used to acquire historical telemetry data of the kth telemetry type. The historical telemetry data of the kth telemetry type includes: a first part of historical data, a second part of historical data, and a third part of historical data. The first part of historical data, the second part of historical data, and the third part of historical data all include: normal values ​​and outliers. The first processing module is used to input the second part of historical data into the outlier detection model of the telemetry data of the k-th telemetry type, and process the second part of historical data through the outlier detection model of the telemetry data of the k-th telemetry type to obtain multiple predicted values ​​of the k-th telemetry type; wherein, each historical telemetry data in the third part of historical data corresponds one-to-one with the multiple predicted values ​​of the k-th telemetry type; the third part of historical data corresponding one-to-one with each of the multiple predicted values ​​is called the true value of historical data; The second processing module is used to obtain the first normal value correlation coefficient and the second normal value correlation coefficient, and based on each predicted value among the multiple predicted values, the first normal value correlation coefficient and the second normal value correlation coefficient, determine the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the k-th telemetry type, and optimize the first normal value correlation coefficient and the second normal value correlation coefficient according to the normal value range to obtain the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient. The third processing module is used to determine whether the telemetry data to be detected is an outlier based on the optimized first normal value correlation coefficient and the optimized second normal value correlation coefficient when the telemetry data to be detected is acquired. The second processing module is used to obtain a first normal value correlation coefficient and a second normal value correlation coefficient, determine the normal value range of the historical data true values ​​corresponding to each of the predicted values ​​in the telemetry data of the k-th telemetry type based on each predicted value, the first normal value correlation coefficient, and the second normal value correlation coefficient, and optimize the first normal value correlation coefficient and the second normal value correlation coefficient based on the normal value range, including: The upper and lower limits of the normal value range of the historical data true values ​​corresponding to each predicted value in the telemetry data of the kth telemetry type are calculated using the following formulas. ； in, Represents the k-th telemetry type. The upper limit of the normal value range for each telemetry data point; Represents the k-th telemetry type. The lower limit of the normal value range for each telemetry data point; This represents the correlation coefficient of the first normal value; This represents the correlation coefficient of the second normal value; Represents the first of multiple predicted values One predicted value; To ensure that the outlier detection rate of the predicted values ​​approaches 1 and the false alarm rate of the predicted values ​​approaches 0, the correlation coefficients of the first and second normal values ​​are optimized using the following formula: ； ； ； in, Indicates the optimization index for the range of values; This represents the Heaviside function; Denotes the first constant; Indicates the second constant; This represents the number of telemetry data points in the outlier detection model of the k-th telemetry type; if ,but , ; Indicates the k-th telemetry type. One true value; This represents the difference between the maximum and minimum values ​​in a subset of telemetry data input into the outlier detection model for the k-th telemetry type; if ,but ;like ,but .

6. The apparatus according to claim 5, characterized in that, The parameters of the outlier detection model for the telemetry data of the kth telemetry type include: the output weight matrix; The training module is also specifically used for: To ensure that the error between the predicted value and the corresponding true value approaches zero, the regularization parameter and the output weight matrix satisfy the following formula: ； ； in, This represents the optimal output weight matrix when the error between the predicted value and the corresponding true value approaches 0. Represents the regularization parameter; This represents the transpose of the feature mapping matrix from the input layer to the hidden layer in the outlier detection model for the k-th telemetry type. Represents the identity matrix; This represents a sequence of historical data true values, composed of multiple historical data true values. This represents the second part of historical data in the outlier detection model input into the telemetry data of the kth telemetry type; This indicates the first in the second part of the historical data. One telemetry data; Indicates the degree of offset, which is a constant; It represents a length scale and is a constant. Utilize The outlier detection model for the telemetry data of the k-th telemetry type is optimized.

7. The apparatus according to claim 6, characterized in that, The third processing module is specifically used for: When the telemetry data to be detected is obtained, a preset number of telemetry data of the kth telemetry type obtained before the telemetry data to be detected are input into the outlier detection model of the telemetry data of the kth telemetry type to obtain the predicted value corresponding to the telemetry data to be detected. Based on the predicted value, the optimized first normal value correlation coefficient, and the optimized second normal value correlation coefficient, the normal value range of the telemetry data to be detected is obtained; When the telemetry data to be detected is not within the normal range of telemetry data values, the acquired telemetry data is determined to be an outlier.

8. The apparatus according to claim 5, characterized in that, Also includes: The training module is used to determine the telemetry type associated with the k-th telemetry type from the telemetry data of the k-1 telemetry types that have been detected, and input the normal values ​​in the historical telemetry data of the telemetry type associated with the k-th telemetry type and the first part of historical data into the kernel extreme learning machine to train the kernel extreme learning machine and obtain the outlier detection model of the telemetry data of the k-th telemetry type.

9. A computer-readable storage medium storing a computer program thereon, characterized in that, When the computer program is run by the processor, it performs the steps of the method described in any one of claims 1-4.

10. An electronic device, characterized in that, The electronic device includes a memory, a processor, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor of the steps of the method according to any one of claims 1-4.

Images