Abnormality detection method and device for weak and small signals of on-orbit spacecraft
By combining the Meta-KAN network model with meta-learning training based on ground experiments and on-orbit telemetry data, the problem of weak signal identification in on-orbit spacecraft was solved, enabling accurate detection and early anomaly identification in complex noise environments, and improving the status monitoring capability of on-orbit spacecraft.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING INST OF TECH
- Filing Date
- 2026-01-28
- Publication Date
- 2026-05-15
AI Technical Summary
Existing technologies struggle to effectively identify weak signals in orbital spacecraft, especially in complex noise environments where it is difficult to extract effective features. They also exhibit weak model transferability, lack interpretability and adaptability to small samples, leading to biases in system state assessment and on-orbit risks.
Meta-KAN network model is used for meta-learning training. Combining large-scale ground experimental data and on-orbit telemetry data, signal feature extraction and anomaly detection are achieved through interpretable operator network structure and hierarchical adaptive feature encoding. Parameter transfer training is carried out using support set and query set to improve the generalization ability and robustness of the model for on-orbit application.
It significantly improves the recognition accuracy and anti-interference capability of weak signals from spacecraft in orbit, enhances the generalization ability and robustness of the model, realizes early anomaly identification and status monitoring, and reduces the dependence on the number of in-orbit samples.
Smart Images
Figure CN122045928A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of weak signal processing technology, and more specifically to a method and apparatus for detecting anomalies in weak signals of on-orbit spacecraft. Background Technology
[0002] During operation, spacecraft in orbit generate a large number of real-time monitoring signals from their attitude control, energy management, and communication payload systems. However, due to the complex space environment, long signal transmission distances, and strong noise interference, some key monitoring signals exhibit typical weak signal characteristics such as low energy, low signal-to-noise ratio, and blurred features. These signals often conceal early signs of equipment failure or performance degradation. If they cannot be identified in a timely and accurate manner, it may lead to deviations in system status assessment or even trigger on-orbit risks.
[0003] Existing methods for identifying weak signals mostly rely on traditional filtering, threshold judgment, or single-model identification based on deep learning. These methods generally suffer from the following problems: First, the model is sensitive to noise and has difficulty extracting effective features from complex backgrounds; Second, the ground test environment differs significantly from the on-orbit environment, resulting in weak model transferability. Third, while deep learning methods have strong representational capabilities, they lack interpretability and small-sample adaptability, making it difficult to cope with the reality of scarce on-orbit data.
[0004] Therefore, how to provide a detection method that can combine large-scale ground-based experimental data with a small amount of on-orbit data for adaptive learning, while taking into account detection accuracy, generalization ability and interpretability, so as to improve the identification and analysis capabilities of weak signals of on-orbit spacecraft, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] In view of the above problems, the present invention is proposed to provide an anomaly detection method and apparatus for weak signals of on-orbit spacecraft that overcomes or at least partially solves the above problems.
[0006] To achieve the above objectives, the present invention adopts the following technical solution: An anomaly detection method for weak signals from an on-orbit spacecraft includes the following steps: S1. Acquire large-scale ground test data of the spacecraft and preprocess it to construct the support set and query set for training in the meta-learning dataset; S2. Based on the interpretable operator network structure, a Meta-KAN network model for weak signal anomaly detection is constructed, and meta-learning training is performed using the support set and query set to obtain a pre-trained Meta-KAN network model. S3. Collect on-orbit telemetry data of the spacecraft, fine-tune the parameters of the pre-trained Meta-KAN network model, realize the knowledge transfer from the ground environment to the on-orbit application scenario, and obtain the trained Meta-KAN network model. S4. Input the weak signal data of the spacecraft to be detected into the trained Meta-KAN network model and output the corresponding signal state.
[0007] Preferably, both ground test data and telemetry data include the spacecraft's current, voltage, and temperature signals.
[0008] Preferably, step S2 includes the following: S21. Embed the basic interpretable operator network KAN unit into the meta-learning framework to form a transferable hierarchical network, and build a Meta-KAN network structure for weak signal anomaly detection. S22. Based on the support set and query set constructed in step S1 for training, the task-level support set and query set are divided in the meta-learning framework. The support set is used to train the Meta-KAN network model to extract general features of weak signals from spacecraft, and the query set is used to improve the Meta-KAN network model to process weak signals from different types of spacecraft. S23. Calculate the gradient through the meta-task total loss function and perform backpropagation to update it. Use a task-batch-based meta-optimization algorithm to iteratively train the Meta-KAN network parameters. The meta-task total loss function includes the task fitting loss of the support set and the generalization evaluation loss of the query set.
[0009] Preferably, the Meta-KAN network model includes a cascaded signal feature encoding network and a state discrimination network. The signal feature encoding network is used to extract and encode the time-domain and frequency-domain signal features of the spacecraft under different operating conditions, and the state discrimination network is used to aggregate and classify the encoded features to realize the anomaly detection of weak signals. Preferably, both the signal feature coding network and the state discrimination network adopt a hierarchical adaptive structure, specifically including: an input normalization layer, a kernel function adaptation layer, a feature mapping layer, and a nonlinear combination layer; The input normalization layer is used to input the support set and the query set, the kernel function adaptation layer is used to automatically adjust the local response range according to the input task, the feature mapping layer is used to express the correlation of signal features at different scales, and the nonlinear combination layer is used to generate high-dimensional distinguishable feature representations.
[0010] The preferred method is to calculate the gradient using the total loss function of the meta-task and then perform backpropagation updates as follows:
[0011] in, Let be the parameter vector of the Meta-KAN network in the i-th task iteration. For learning rate, The total loss function for the meta-task. For parameters The gradient; Meta-task total loss function for:
[0012] in, , These are the loss weight coefficients for the support set and the query set, respectively. To support the collection of task losses, To inquire about the losses incurred during the task.
[0013] Preferably, step S3 includes the following: S31. Collect on-orbit telemetry data from spacecraft in orbit, and perform noise reduction, outlier removal, and time synchronization processing; S32. Align the processed on-orbit telemetry data with the ground test data in terms of feature distribution, and use the pre-trained Meta-KAN network model parameters as the initial weights to achieve transfer fine-tuning; S33. With the goal of minimizing the joint loss of classification task loss and feature distribution alignment loss, a low-learning-rate iterative training method is adopted to train the pre-trained Meta-KAN network model using a small sample meta-task approach, gradually improving the anomaly detection performance of the Meta-KAN network for weak signals in spacecraft on orbit. After completing the predetermined number of iterations and passing the verification, the trained Meta-KAN network model is solidified.
[0014] An anomaly detection system for weak signals of an on-orbit spacecraft, based on the aforementioned anomaly detection method for weak signals of an on-orbit spacecraft, includes: a data acquisition and processing module, a feature extraction and learning module, an adaptive fine-tuning module, and an anomaly detection module; The data acquisition and processing module is used to acquire and preprocess large-scale ground test data of spacecraft, construct support sets and query sets for training in the meta-learning dataset, and also to acquire on-orbit telemetry data of spacecraft. The feature extraction and learning module is used to construct a Meta-KAN network model for weak signal anomaly detection based on an interpretable operator network structure, and to perform meta-learning training using the support set and query set to obtain a pre-trained Meta-KAN network model. The adaptive fine-tuning module is used to fine-tune the parameters of the pre-trained Meta-KAN network model based on the spacecraft's on-orbit telemetry data, so as to realize the knowledge transfer from the ground environment to the on-orbit application scenario and obtain the trained Meta-KAN network model. The anomaly detection module is used to input the weak signal data of the spacecraft to be detected into the trained Meta-KAN network model and output the corresponding signal status.
[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for detecting anomalies in weak signals of an on-orbit spacecraft.
[0016] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the aforementioned method for detecting anomalies in weak signals from an on-orbit spacecraft.
[0017] As can be seen from the above technical solutions, compared with the prior art, this invention discloses a method and device for detecting weak signals in on-orbit spacecraft anomalies. By introducing large-scale ground test data to construct support sets and query sets, and combining them with a meta-learning framework for parameter transfer training, the dependence on the number of on-orbit samples is significantly reduced, achieving accurate identification of weak signals under small sample conditions. Through the hierarchical adaptive structure and feature encoding mechanism of the Meta-KAN network, the model can extract key features in a strong noise background, improving the accuracy and anti-interference capability of signal detection. Through an adaptive fine-tuning strategy, ground knowledge is efficiently transferred to on-orbit application scenarios, enhancing the model's generalization ability and robustness under different mission environments and electromagnetic conditions, and effectively improving the early anomaly identification and status monitoring capabilities of on-orbit spacecraft. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.
[0019] Figure 1 This is a schematic diagram of an anomaly detection method for weak signals in an on-orbit spacecraft provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the meta-KAN network model meta-learning training provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the detection results of weak signals in a spacecraft provided in an embodiment of the present invention. Detailed Implementation
[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] This invention discloses an anomaly detection method for weak signals in an on-orbit spacecraft, comprising the following steps: S1. Acquire large-scale ground test data of the spacecraft and preprocess it to construct the support set and query set for training in the meta-learning dataset; S2. Based on the interpretable operator network structure, a Meta-KAN network model for weak signal anomaly detection is constructed, and meta-learning training is performed using the support set and query set to obtain a pre-trained Meta-KAN network model. S3. Collect on-orbit telemetry data of the spacecraft, fine-tune the parameters of the pre-trained Meta-KAN network model, realize the knowledge transfer from the ground environment to the on-orbit application scenario, and obtain the trained Meta-KAN network model. S4. Input the weak signal data of the spacecraft to be detected into the trained Meta-KAN network model and output the corresponding signal state.
[0022] To further implement the above technical solutions, both ground test data and telemetry data include the spacecraft's current, voltage, and temperature signals.
[0023] In this embodiment, the preprocessing in step S1 includes standardizing the support set and the query set respectively:
[0024] in, The input is the raw, large-scale ground test data. , These are the mean and standard deviation of the original data, respectively. The data has been standardized. The length of the raw data vector for a single signal can be 1024 points, and a single sample can be in matrix format (number of tasks, number of channels, number of sampling points), for example (20, 4, 1024).
[0025] To further implement the above technical solutions, such as Figure 2 The specific content of step S2 includes: S21. Embed the basic interpretable operator network KAN unit into the meta-learning framework to form a transferable hierarchical network, and build a Meta-KAN network structure for weak signal anomaly detection. S22. Based on the support set and query set constructed in step S1 for training, the task-level support set and query set are divided in the meta-learning framework. The support set is used to train the Meta-KAN network model to extract general features of weak signals from spacecraft, and the query set is used to improve the Meta-KAN network model to process weak signals from different types of spacecraft. S23. Calculate the gradient through the meta-task total loss function and perform backpropagation to update it. Use a task-batch-based meta-optimization algorithm to iteratively train the Meta-KAN network parameters. The meta-task total loss function includes the task fitting loss of the support set and the generalization evaluation loss of the query set.
[0026] To further implement the above technical solution, the Meta-KAN network model includes a cascaded signal feature encoding network and a state discrimination network. The signal feature encoding network is used to extract and encode the time-domain and frequency-domain signal features of the spacecraft under different operating conditions, and the state discrimination network is used to aggregate and classify the encoded features to realize the anomaly detection of weak signals. In this embodiment, the final layer of the state discrimination network is a binary classifier composed of a nonlinear combination structure, which outputs... ∈{0,1} represents the current state of the signal, where, =0 indicates the normal state. =1 indicates an abnormal state.
[0027] To further implement the above technical solution, both the signal feature coding network and the state discrimination network adopt a hierarchical adaptive structure, which specifically includes: an input normalization layer, a kernel function adaptation layer, a feature mapping layer, and a nonlinear combination layer. The input normalization layer is used to input the support set and the query set, the kernel function adaptation layer is used to automatically adjust the local response range according to the input task, the feature mapping layer is used to express the correlation of signal features at different scales, and the nonlinear combination layer is used to generate high-dimensional distinguishable feature representations to improve the detection accuracy of weak signals in noisy environments.
[0028] In this embodiment, for any input sample The kernel function adaptation layer adapts to the task parameters. Generate a local response kernel; the kernel function adaptation layer is represented as follows:
[0029] in, As the core, The scale parameter is automatically adjusted by the current task and is used to characterize the sensitivity of local features under different working conditions; The feature mapping layer maps multi-scale time-domain and frequency-domain features to a unified high-dimensional representation space, as follows:
[0030] in, Represents the kernel function response vector. This represents a cross-scale correlation mapping matrix used to fuse the time-frequency feature patterns of weak signals; The nonlinear combination layer utilizes additivity neural operator structures to nonlinearly reconstruct the mapped features, forming a more discriminative high-dimensional representation:
[0031] in, As a learnable B-spline base, For task-adaptive combination coefficients; The nonlinear combination layer enhances the nonlinear interaction between features, enabling weak anomalous signals to be effectively amplified and distinguished in a background of strong noise. For the state discrimination network, the nonlinear combined output h is used to obtain the final state label via logistic regression. This process can be expressed as: .
[0032] To further implement the above technical solution, the method of calculating the gradient and updating it through backpropagation using the total loss function of the meta-task is as follows:
[0033] in, Let be the parameter vector of the Meta-KAN network in the i-th task iteration. For learning rate, The total loss function for the meta-task. For parameters The gradient; Meta-task total loss function for:
[0034] in, , These are the loss weight coefficients for the support set and the query set, respectively. To support the collection of task losses, To inquire about the losses incurred during the task.
[0035] To further implement the above technical solution, step S3 includes the following: S31. Collect on-orbit telemetry data from spacecraft in orbit, and perform noise reduction, outlier removal, and time synchronization processing; S32. Align the processed on-orbit telemetry data with the ground test data in terms of feature distribution, and use the pre-trained Meta-KAN network model parameters as the initial weights to achieve transfer fine-tuning; S33. With the goal of minimizing the joint loss of classification task loss and feature distribution alignment loss, a low-learning-rate iterative training method is adopted to train the pre-trained Meta-KAN network model using a small sample meta-task approach, gradually improving the anomaly detection performance of the Meta-KAN network for weak signals in spacecraft on orbit. After completing the predetermined number of iterations and passing the verification, the trained Meta-KAN network model is solidified.
[0036] In this embodiment, the specific details of the iterative training method for the few-sample meta-task are as follows: Several small-sample meta-tasks are constructed using the processed on-orbit telemetry data. Each meta-task contains a small number of support set samples and corresponding query set samples. For each meta-task, the pre-trained Meta-KAN network model is iteratively updated using a low learning rate. Specifically, the model parameters are first fine-tuned on the support set to capture weak signal features under the current task. Then, the model performance is evaluated on the query set and the joint loss of classification task loss and feature distribution alignment loss is calculated. The model weights are updated through backpropagation. Through iterative training on all meta-tasks, the model gradually accumulates the ability to identify weak signals under different working conditions. The loss function for feature distribution alignment is expressed as:
[0037] in, The distribution of encoding features for ground test data. The distribution of encoding features for on-orbit telemetry data, For remapping to the kernel mapping of RKHS, This indicates the coding characteristics of on-orbit telemetry data. Indicates the coding characteristics of ground test data; The migration fine-tuning employs the minimization of the joint loss with an alignment term, expressed as:
[0038] in, This represents the parameters of the Meta-KAN network model. For binary cross-entropy, To obtain the fine-tuned parameters, the pre-trained parameters of the Meta-KAN network model are used as the starting point for the alignment term weight hyperparameters. Several rounds of iterative optimization are performed with a small learning rate. After completing the predetermined number of iterations and passing the validation set evaluation, the trained Meta-KAN network model parameters are solidified for subsequent on-orbit anomaly detection.
[0039] In another embodiment, a set of weak aerospace signal data to be detected is input into the trained Meta-KAN network model. When the signal is in a normal state, the Meta-KAN network outputs signal state 0; when the signal is in an abnormal state, the Meta-KAN network outputs signal state 1, such as... Figure 3 As shown.
[0040] An anomaly detection system for weak signals of on-orbit spacecraft, based on an anomaly detection method for weak signals of on-orbit spacecraft, includes: a data acquisition and processing module, a feature extraction and learning module, an adaptive fine-tuning module, and an anomaly detection module; The data acquisition and processing module is used to acquire and preprocess large-scale ground test data of spacecraft, construct support sets and query sets for training in the meta-learning dataset, and also to acquire on-orbit telemetry data of spacecraft. The feature extraction and learning module is used to construct a Meta-KAN network model for weak signal anomaly detection based on an interpretable operator network structure, and to perform meta-learning training using the support set and query set to obtain a pre-trained Meta-KAN network model. The adaptive fine-tuning module is used to fine-tune the parameters of the pre-trained Meta-KAN network model based on the spacecraft's on-orbit telemetry data, so as to realize the knowledge transfer from the ground environment to the on-orbit application scenario and obtain the trained Meta-KAN network model. The anomaly detection module is used to input the weak signal data of the spacecraft to be detected into the trained Meta-KAN network model and output the corresponding signal status.
[0041] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for detecting anomalies in weak signals from an on-orbit spacecraft.
[0042] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a method for detecting anomalies in weak signals of an on-orbit spacecraft.
[0043] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.
[0044] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for detecting anomalies in weak signals from an on-orbit spacecraft, characterized in that, Includes the following steps: S1. Acquire large-scale ground test data of the spacecraft and preprocess it to construct the support set and query set for training in the meta-learning dataset; S2. Based on the interpretable operator network structure, a Meta-KAN network model for weak signal anomaly detection is constructed, and meta-learning training is performed using the support set and query set to obtain a pre-trained Meta-KAN network model. S3. Collect on-orbit telemetry data of the spacecraft, fine-tune the parameters of the pre-trained Meta-KAN network model, realize the knowledge transfer from the ground environment to the on-orbit application scenario, and obtain the trained Meta-KAN network model. S4. Input the weak signal data of the spacecraft to be detected into the trained Meta-KAN network model and output the corresponding signal state.
2. The method for detecting anomalies in weak signals of an on-orbit spacecraft as described in claim 1, characterized in that, Both ground test data and telemetry data include the spacecraft's current, voltage, and temperature signals.
3. The method for detecting anomalies in weak signals of an on-orbit spacecraft as described in claim 1, characterized in that, The specific content of step S2 includes: S21. Embed the basic interpretable operator network KAN unit into the meta-learning framework to form a transferable hierarchical network, and build a Meta-KAN network structure for weak signal anomaly detection. S22. Based on the support set and query set constructed in step S1 for training, the task-level support set and query set are divided in the meta-learning framework. The support set is used to train the Meta-KAN network model to extract general features of weak signals from spacecraft, and the query set is used to improve the Meta-KAN network model to process weak signals from different types of spacecraft. S23. Calculate the gradient through the meta-task total loss function and perform backpropagation to update it. Use a task-batch-based meta-optimization algorithm to iteratively train the Meta-KAN network parameters. The meta-task total loss function includes the task fitting loss of the support set and the generalization evaluation loss of the query set.
4. The method for detecting anomalies in weak signals of an on-orbit spacecraft as described in claim 3, characterized in that, The Meta-KAN network model includes a cascaded signal feature encoding network and a state discrimination network. The signal feature encoding network is used to extract and encode the time-domain and frequency-domain signal features of the spacecraft under different operating conditions, while the state discrimination network is used to aggregate and classify the encoded features to achieve anomaly detection of weak signals.
5. The method for detecting anomalies in weak signals of an on-orbit spacecraft as described in claim 4, characterized in that, Both the signal feature coding network and the state discrimination network adopt a hierarchical adaptive structure, which specifically includes: an input normalization layer, a kernel function adaptation layer, a feature mapping layer, and a nonlinear combination layer. The input normalization layer is used to input the support set and the query set, the kernel function adaptation layer is used to automatically adjust the local response range according to the input task, the feature mapping layer is used to express the correlation of signal features at different scales, and the nonlinear combination layer is used to generate high-dimensional distinguishable feature representations.
6. The method for detecting anomalies in weak signals of an on-orbit spacecraft as described in claim 3, characterized in that, The method for calculating the gradient and updating it via backpropagation using the total loss function of the meta-task is as follows: in, Let be the parameter vector of the Meta-KAN network in the i-th task iteration. For learning rate, The total loss function for the meta-task. For parameters The gradient; Meta-task total loss function for: in, , These are the loss weight coefficients for the support set and the query set, respectively. To support the collection of task losses, To inquire about the losses incurred during the task.
7. The method for detecting anomalies in weak signals of an on-orbit spacecraft as described in claim 1, characterized in that, The specific content of step S3 includes: S31. Collect on-orbit telemetry data from spacecraft in orbit, and perform noise reduction, outlier removal, and time synchronization processing; S32. Align the processed on-orbit telemetry data with the ground test data in terms of feature distribution, and use the pre-trained Meta-KAN network model parameters as the initial weights to achieve transfer fine-tuning; S33. With the goal of minimizing the joint loss of classification task loss and feature distribution alignment loss, a low-learning-rate iterative training method is adopted to train the pre-trained Meta-KAN network model using a small sample meta-task approach, gradually improving the anomaly detection performance of the Meta-KAN network for weak signals in spacecraft on orbit. After completing the predetermined number of iterations and passing the verification, the trained Meta-KAN network model is solidified.
8. An anomaly detection system for weak signals in an on-orbit spacecraft, characterized in that, An anomaly detection method for weak signals in an on-orbit spacecraft according to any one of claims 1-7 includes: a data acquisition and processing module, a feature extraction and learning module, an adaptive fine-tuning module, and an anomaly detection module. The data acquisition and processing module is used to acquire and preprocess large-scale ground test data of spacecraft, construct support sets and query sets for training in the meta-learning dataset, and also to acquire on-orbit telemetry data of spacecraft. The feature extraction and learning module is used to construct a Meta-KAN network model for weak signal anomaly detection based on an interpretable operator network structure, and to perform meta-learning training using the support set and query set to obtain a pre-trained Meta-KAN network model. The adaptive fine-tuning module is used to fine-tune the parameters of the pre-trained Meta-KAN network model based on the spacecraft's on-orbit telemetry data, so as to realize the knowledge transfer from the ground environment to the on-orbit application scenario and obtain the trained Meta-KAN network model. The anomaly detection module is used to input the weak signal data of the spacecraft to be detected into the trained Meta-KAN network model and output the corresponding signal status.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements an anomaly detection method for weak signals in an on-orbit spacecraft as described in any one of claims 1-7.
10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements an anomaly detection method for weak signals of an on-orbit spacecraft as described in any one of claims 1-7.