WO2020181962A1 - 一种对抗网络自编码器的卫星异常检测方法及系统 - Google Patents

一种对抗网络自编码器的卫星异常检测方法及系统 Download PDF

Info

Publication number
WO2020181962A1
WO2020181962A1 PCT/CN2020/075658 CN2020075658W WO2020181962A1 WO 2020181962 A1 WO2020181962 A1 WO 2020181962A1 CN 2020075658 W CN2020075658 W CN 2020075658W WO 2020181962 A1 WO2020181962 A1 WO 2020181962A1
Authority
WO
WIPO (PCT)
Prior art keywords
satellite
optimized
variational autoencoder
data
operating state
Prior art date
Application number
PCT/CN2020/075658
Other languages
English (en)
French (fr)
Inventor
皮德常
陈俊夫
吴致远
Original Assignee
南京航空航天大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南京航空航天大学 filed Critical 南京航空航天大学
Priority to US17/255,267 priority Critical patent/US12093804B2/en
Publication of WO2020181962A1 publication Critical patent/WO2020181962A1/zh

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B7/00Radio transmission systems, i.e. using radiation field
    • H04B7/14Relay systems
    • H04B7/15Active relay systems
    • H04B7/185Space-based or airborne stations; Stations for satellite systems
    • H04B7/18578Satellite systems for providing broadband data service to individual earth stations
    • H04B7/18582Arrangements for data linking, i.e. for data framing, for error recovery, for multiple access
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/18Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/044Recurrent networks, e.g. Hopfield networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/084Backpropagation, e.g. using gradient descent
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • G06N3/088Non-supervised learning, e.g. competitive learning

Definitions

  • the present invention relates to the field of engineering application and information science, in particular to a satellite anomaly detection method and system against network autoencoders.
  • the abnormal detection of telemetry data is of great significance in the fields of satellite troubleshooting and real-time health detection. Taking into account the satellite's complex design structure and harsh working environment, it is impossible to directly perform anomaly detection in the outer space environment.
  • NASA National Aeronautics and Space Administration
  • NASA National Aeronautics and Space Administration
  • a satellite anomaly detection and fault diagnosis method based on simple condition monitoring was proposed.
  • an algorithm-based satellite anomaly detection and fault diagnosis method was proposed.
  • a knowledge-based satellite anomaly detection and fault diagnosis method was proposed.
  • a satellite anomaly detection and fault diagnosis method based on agent and network is proposed. Since entering the 21st century, in order to improve the stability and safety of satellites in orbit, NASA has proposed satellite integrated health management technology. This technology integrates the performance evaluation, anomaly detection, fault prediction and other modules of each satellite subsystem, as well as its corresponding processing measures and logistical support arrangements, into a comprehensive satellite health management system.
  • DaweiPan et al. proposed an anomaly detection algorithm based on kernel principal component analysis (KPCA) and association rules, using KPCA to construct a feature matrix, and using closed patterns to mine association rules between telemetry attributes in a time series to identify anomalies. This method is not effective for high-dimensional telemetry parameters and cannot solve the problem of high data latitude.
  • Qin et al. designed an algorithm to distinguish between periodic data range and monotonicity, and early warning of abnormalities by detecting the lower limit of telemetry data.
  • This method has a higher detection accuracy for telemetry data with strong periodicity, but for telemetry data with insignificant or no periodicity, it cannot accurately determine the frequency change, so it cannot detect anomalies.
  • Wang et al. first analyzed the illumination period and shadow period of the satellite in orbit, then used the local linear mapping algorithm to reduce dimensionality and extract features, map the high-dimensional data space to the low-dimensional feature space, and finally calculate the statistics T2 and SPE are used to detect faults, but automatic detection cannot be achieved, resulting in high anomaly detection errors, and the accuracy of automatic satellite anomaly detection cannot be guaranteed.
  • the object of the present invention is to provide a satellite anomaly detection method and system against network autoencoders, and solve the problem of low accuracy of satellite anomaly automatic detection in the prior art.
  • a satellite anomaly detection method against network autoencoders including:
  • variational autoencoder takes satellite telemetry data as input and reconstructed data as output;
  • variational autoencoder includes an encoder and a decoder;
  • generator The adversarial network includes generator and discriminator;
  • the optimized variational autoencoder includes the optimized encoder and the decoder; so The optimized encoder is used to encode the satellite telemetry data to be detected into hidden variables; the decoder is used to decode the hidden variables into the reconstructed data;
  • the optimized variational autoencoder is used to determine the current operating state of the satellite; the current operating state includes a normal operating state or an abnormal operating state.
  • the use of the optimized variational autoencoder to determine the current operating state of the satellite according to the satellite telemetry data to be detected specifically includes:
  • the Mahalanobis distance is used to determine the reconstruction error
  • the reconstruction error is not less than the error threshold, it is determined that the current operating state of the satellite is an abnormal operating state.
  • the acquiring error threshold specifically includes:
  • the wavelet variance is used to analyze the satellite telemetry sample data to determine the error threshold at each moment.
  • the adding the generative confrontation network to the variational autoencoder to determine the optimized variational autoencoder specifically includes:
  • the optimized variational autoencoder is determined according to the optimized latent variable and the decoder in the variational autoencoder.
  • a satellite anomaly detection system against network autoencoders including:
  • the first acquisition module is used to acquire a variational autoencoder and a generative countermeasure network;
  • the variational autoencoder takes satellite telemetry data as input and reconstructed data as output;
  • the variational autoencoder includes an encoder And a decoder;
  • the generative confrontation network includes a generator and a discriminator;
  • the optimized variational autoencoder determination module is used to add the generative confrontation network to the variational autoencoder to determine the optimized variational autoencoder; the optimized variational autoencoder It includes an optimized encoder and the decoder; the optimized encoder is used to encode the satellite telemetry data to be detected into a hidden variable; the decoder is used to decode the hidden variable into the reconstruction data;
  • the second acquisition module is used to acquire satellite telemetry data to be detected
  • the satellite's current operating state determining module is used to determine the current operating state of the satellite by using the optimized variational autoencoder according to the satellite telemetry data to be detected; the current operating state includes normal operating state or abnormal Operating status.
  • the module for determining the current operating state of the satellite specifically includes:
  • a reconstruction data determination unit configured to determine the reconstruction data by using the optimized variational autoencoder according to the satellite telemetry data to be detected;
  • a reconstruction error determination unit configured to determine the reconstruction error by using the Mahalanobis distance according to the satellite telemetry data to be detected and the reconstruction data;
  • An error threshold value acquisition unit for acquiring an error threshold value
  • a judging unit configured to judge whether the reconstruction error is less than the error threshold
  • a normal operating state determining unit configured to determine that the current operating state of the satellite is a normal operating state if the reconstruction error is less than the error threshold;
  • An abnormal operating state determining unit configured to determine that the current operating state of the satellite is an abnormal operating state if the reconstruction error is not less than the error threshold.
  • the error threshold obtaining unit specifically includes:
  • Satellite telemetry sample data acquisition subunit for acquiring satellite telemetry sample data in a satellite period in the database
  • the error threshold determination subunit at each time is used to analyze the satellite remote measurement sample data by using wavelet variance according to the satellite remote measurement sample data to determine the error threshold at each time.
  • the optimized variational autoencoder determining module specifically includes:
  • the generator determining unit is configured to determine the generator of the generative confrontation network according to the encoder in the variational autoencoder; the generator is used to optimize the encoder in the variational autoencoder and determine the optimization After the encoder;
  • a discriminator determining unit which determines the discriminator of the generative confrontation network according to the two-way long and short-term memory network
  • the optimized variational autoencoder determining unit is configured to determine the optimized variational autoencoder according to the optimized latent variable and the decoder in the variational autoencoder.
  • the present invention Compared with the prior art, the present invention has the advantage that: the present invention provides a satellite anomaly detection method and system against network autoencoders, in which an optimized variational encoder is obtained by introducing a confrontation network in a variational encoder , Through the optimized variational encoder to realize the detection of satellite telemetry data, the present invention uses pure data drive, does not require expert experience, and can be applied to a variety of occasions; combines the respective advantages of the variational autoencoder and the generative confrontation network, fast and accurate Detecting abnormal data, that is, detecting the abnormal state of the satellite, solves the problem of the inability to achieve automatic detection and high detection error in the prior art, and ensures the accuracy of automatic detection of satellite anomalies.
  • Figure 1 is a schematic flow diagram of a satellite anomaly detection method against network autoencoders provided by the present invention
  • Figure 2 is a schematic diagram of the cyclic neural network structure provided by the present invention.
  • Figure 3 is a schematic structural diagram of a satellite anomaly detection system against network autoencoders provided by the present invention.
  • the basic idea of the present invention is to use an autoencoder to reconstruct data, and determine abnormalities based on reconstruction errors.
  • Introducing the countermeasure network combining the advantages of both the variational autoencoder and the generative countermeasure network, makes the reconstruction model more accurate, reduces the error of the reconstruction model, and reduces the error of anomaly detection.
  • Mahalanobis distance is used to calculate the reconstruction error. Perform periodic analysis on the data to further improve the accuracy of reconstructed data.
  • Fig. 1 is a schematic flow chart of a satellite anomaly detection method against a network autoencoder provided by the present invention.
  • the satellite anomaly detection method against a network autoencoder provided by the present invention includes:
  • variational autoencoder takes satellite telemetry data as input and reconstructed data as output;
  • variational autoencoder includes an encoder and a decoder;
  • the generative confrontation network includes a generator and a discriminator.
  • the loss function of the variational autoencoder is Among them, L VEA is the error of the variational autoencoder , L R is the reconstruction error of the sample, Is the KL divergence of the posterior probability and the true posterior probability.
  • Figure 2 is a schematic diagram of the unidirectional recurrent neural network structure, and the unidirectional recurrent neural network can only process data in one direction.
  • the present invention uses a long- and short-term memory network as the discriminator in the generative confrontation network.
  • Use formula Determine the superposition of the two recurrent neural network layers and enter the hidden layer.
  • P data and P z are the distribution of real data and generated data, respectively, DIS is the generator, GEN is the discriminator, x is the input data, and z is the hidden variable generated by the encoder.
  • the generator it is hoped that the generated data can fool the discriminator and make the discriminant value of the generated data close to 1.
  • the discriminator model's discriminatory results are in a state of swinging up and down; then, repeat the fixed generator, optimize the discriminator, and fix the discriminator, and train the generator. A process, until the discriminator cannot distinguish the true and false of the generated data.
  • the purpose of the generator is to keep P z getting closer to P data through continuous training of the generative model, and to construct a discriminator to distinguish the generated data from the real data.
  • the optimized variational autoencoder includes the optimized encoder and the decoder
  • the optimized encoder is used to encode the satellite telemetry data to be detected into a hidden variable
  • the decoder is used to decode the hidden variable into the reconstructed data.
  • the generator of the generative confrontation network is determined according to the encoder in the variational autoencoder; the generator is used to optimize the encoder in the variational autoencoder and determine the optimized encoder.
  • the discriminator of the generative confrontation network is determined according to the two-way long and short-term memory network.
  • the hidden variables encoded by the optimized encoder are discriminated according to the discriminator, and the optimized hidden variables are determined.
  • the optimized variational autoencoder is determined according to the optimized latent variable and the decoder in the variational autoencoder.
  • S103 Obtain satellite telemetry data to be detected.
  • the satellite telemetry data changes as the satellite periodically changes.
  • the optimized variational autoencoder is used to determine the current operating state of the satellite; the current operating state includes a normal operating state or an abnormal operating state.
  • the optimized variational autoencoder is used to determine the reconstruction data.
  • the Mahalanobis distance is used to determine the reconstruction error.
  • the score of the reconstruction error is determined on the basis of the Mahalanobis distance to determine the reconstruction error.
  • x(i) is the i-th satellite telemetry data, It is the i-th satellite telemetry data reconstructed by the autoencoder.
  • RE score (i) is a relatively robust reconstruction error metric in high-dimensional space, which can reflect the trend of satellite telemetry data reconstruction errors. In order to monitor this trend and detect abnormal changes, an error threshold must be set. That is, the accuracy of reconstruction error is provided by setting the error threshold.
  • the reconstruction error is less than the error threshold, it is determined that the current operating state of the satellite is a normal operating state.
  • the reconstruction error is not less than the error threshold, it is determined that the current operating state of the satellite is an abnormal operating state.
  • Satellite telemetry data reflects the current state of satellites in orbit. Because satellite operations are periodic, satellite telemetry data will also show periodic changes with the satellite period. The satellite telemetry data change period is accurately analyzed for abnormal detection. Research is of great significance. Due to the high dimensionality and large amount of satellite telemetry data, it is impossible to directly observe the period of satellite telemetry data.
  • the error threshold obtained through the period of satellite telemetry data specifically includes:
  • the wavelet variance is used to analyze the satellite telemetry sample data, and the specific process of determining the error threshold at each moment is:
  • the variance of the satellite telemetry data is determined.
  • the recent historical data constitutes a new prediction sample, which is called the validation set, and the dynamic error threshold interval of the validation set is dynamically constructed through the time window.
  • Fig. 3 is a schematic structural diagram of a satellite anomaly detection system against a network autoencoder provided by the present invention.
  • a satellite anomaly detection system against a network autoencoder provided by the present invention includes: An acquisition module 301, an optimized variational autoencoder determination module 302, a second acquisition module 303, and a satellite current operating state determination module 304.
  • the first acquisition module 301 is used to acquire a variational autoencoder and a generative countermeasure network; the variational autoencoder takes satellite telemetry data as input and reconstructed data as an output; the variational autoencoder includes an encoder And a decoder; the generative confrontation network includes a generator and a discriminator.
  • the optimized variational autoencoder determining module 302 is used to add the generative confrontation network to the variational autoencoder to determine the optimized variational autoencoder; the optimized variational autoencoder It includes an optimized encoder and the decoder; the optimized encoder is used to encode the satellite telemetry data to be detected into a hidden variable; the decoder is used to decode the hidden variable into the reconstruction data.
  • the second acquisition module 303 is used to acquire satellite telemetry data to be detected.
  • the satellite current operating state determining module 304 is used to determine the current operating state of the satellite by using the optimized variational autoencoder according to the satellite telemetry data to be detected; the current operating state includes normal operating state or abnormal Operating status.
  • the current operating state determination module 304 of the satellite specifically includes: a reconstruction data determination unit, a reconstruction error determination unit, an error threshold acquisition unit determination unit, a normal operation state determination unit, and an abnormal operation state determination unit.
  • the reconstruction data determining unit is used for determining the reconstruction data by using the optimized variational autoencoder according to the satellite telemetry data to be detected.
  • the reconstruction error determining unit is used to determine the reconstruction error by using the Mahalanobis distance according to the satellite telemetry data to be detected and the reconstruction data.
  • the error threshold acquisition unit is used to acquire the error threshold.
  • the judging unit is used to judge whether the reconstruction error is less than the error threshold.
  • the normal operating state determining unit is configured to determine that the current operating state of the satellite is a normal operating state if the reconstruction error is less than the error threshold.
  • the abnormal operating state determining unit is configured to determine that the current operating state of the satellite is an abnormal operating state if the reconstruction error is not less than the error threshold.
  • the error threshold obtaining unit specifically includes: a satellite telemetry sample data obtaining subunit and an error threshold determining subunit at each moment.
  • the satellite telemetry sample data acquisition subunit is used to acquire satellite telemetry sample data in a satellite cycle in the database.
  • the error threshold determination subunit at each moment is used to analyze the satellite telemetry sample data by using wavelet variance based on the satellite telemetry sample data to determine the error threshold at each moment.
  • the optimized variational autoencoder determining module 302 specifically includes: a generator determining unit, a discriminator determining unit, an optimized latent variable determining unit, and an optimized variational autoencoder determining unit.
  • the generator determining unit is used to determine the generator of the generative confrontation network according to the encoder in the variational autoencoder; the generator is used to optimize the encoder in the variational autoencoder and determine the optimized Encoder.
  • the discriminator determining unit determines the discriminator of the generative confrontation network according to the two-way long and short-term memory network.
  • the optimized hidden variable determining unit is used for discriminating the hidden variable encoded by the optimized encoder according to the discriminator, and determining the optimized hidden variable.
  • the optimized variational autoencoder determining unit is used to determine the hidden variable
  • the decoder in the variational autoencoder determines the optimized variational autoencoder.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Aviation & Aerospace Engineering (AREA)
  • Astronomy & Astrophysics (AREA)
  • Mathematical Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Biology (AREA)
  • Operations Research (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Radio Relay Systems (AREA)
  • Testing Or Calibration Of Command Recording Devices (AREA)

Abstract

一种对抗网络自编码器的卫星异常检测方法及系统。该方法包括获取变分自编码器和生成式对抗网络(S101);在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器(S102);获取待检测的卫星遥测数据(S103);根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态(S104);所述当前运行状态包括正常运行状态或异常运行状态。所述一种对抗网络自编码器的卫星异常检测方法及系统,能够解决现有技术中卫星异常自动检测准确度低的问题。

Description

一种对抗网络自编码器的卫星异常检测方法及系统
本申请要求于2019年03月13日提交中国专利局、申请号为201910195659.X、发明名称为“一种对抗网络自编码器的卫星异常检测方法”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
技术领域
本发明涉及工程应用与信息科学领域,特别是涉及一种对抗网络自编码器的卫星异常检测方法及系统。
背景技术
由于卫星长期处在太阳辐射等恶劣的外太空环境下,其在轨运行期间可能会发生无法预料的异常或故障,提前采取措施及时发现这些无法预料的异常或故障,对保障卫星长期稳定运行至关重要。因此,遥测数据的异常检测在卫星故障排查和实时健康检测等领域具有重要的意义。考虑到卫星复杂的设计结构和恶劣的工作环境,无法直接在外太空的环境下进行异常检测。目前通过在卫星各部件上安装多个传感器直接采集卫星各部件的在轨运行数据,将在轨运行数据传输到地面遥测中心,存储为时序遥测数据,然后对卫星时序遥测数据进行分析,进而实现对卫星在轨状态的异常检测。
美国国家航空航天局(National Aeronautics and Space Administration,NASA)在卫星异常检测和故障诊断方面开展了大量的工作。60年代提出了基于简单的状态监测的卫星异常检测和故障诊断方法,70年代提出了基于算法的卫星异常检测和故障诊断方法,80年代提出了基于知识的卫星异常检测和故障诊断方法,90年代提出了基于agent以及网络的卫星异常检测和故障诊断方法。进入21世纪以来,为提高卫星在轨运行的平稳性和安全性,NASA提出了卫星综合健康管理技术。该技术将卫星各子 系统的性能评估、异常检测、故障预测等模块,以及其相应的处理措施和后勤保障安排等,融合为一个卫星健康状况综合管理系统。
从20世纪80年代开始,数据挖掘就在统计分析领域初露锋芒。其中,基于统计的假设检验首先在异常检测领域崭露头角,通过对小概率事件的统计,来判别数据样本是否发生异常。此后,Grubbs等提出在数据集满足标准的t分布基础上对每个数据计算其Grubbs测试统计值,若数据的统计值超过某个阈值,则被认为是一个异常数据。Knorr等在异常检测领域首先提出了距离的概念,计算不同数据点之间的距离作为判断是否发生异常的依据,但是当数据点来自于不同密度分布的数据集时,该方法的异常检测效果不佳。为解决这一问题,Breunig等将密度计算加入到异常检测算法中,该算法避免了密度分布不均的数据集异常检测结果差的问题,但是,该算法复杂度过高,不能处理海量数据。DaweiPan等提出了一种基于核主成分分析(KPCA)和关联规则的异常检测算法,采用KPCA构造特征矩阵,在时间序列中使用闭合模式挖掘遥测属性之间的关联规则,以此来识别异常。该方法对于遥测参数高维的情况效果不好,不能解决数据纬度高的问题。此外,Qin等人设计了一种区分周期性数据范围和单调性的算法,通过检测遥测数据的下界限来提前预警异常。这种方法对于具有较强周期性的遥测数据检测准确率较高,但是针对周期性不明显或者没有周期性的遥测数据,无法准确判断其频率的变化,故无法检测异常。Wang等针对TX-1星的故障检测,自先分析出卫星在轨光照期和阴影期,然后使用局部线性映射算法降维和提取特征,将高维数据空间映射到低维特征空间,最后计算统计量T2和SPE来检测故障,但是无法实现自动检 测,并造成异常检测误差高,无法保证卫星异常自动检测的准确度。
发明内容
本发明的目的是提供一种对抗网络自编码器的卫星异常检测方法及系统,解决现有技术中卫星异常自动检测准确度低的问题。
为达到上述目的,本发明的技术方案为:
一种对抗网络自编码器的卫星异常检测方法,包括:
获取变分自编码器和生成式对抗网络;所述变分自编码器以卫星遥测数据为输入,以重构数据为输出;所述变分自编码器包括编码器和解码器;所述生成式对抗网络包括生成器和判别器;
在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器;所述优化后的变分自编码器包括优化后的编码器和所述解码器;所述优化后的编码器用于将所述待检测的卫星遥测数据编码为隐变量;所述解码器用于将所述隐变量解码为所述重构数据;
获取待检测的卫星遥测数据;
根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态;所述当前运行状态包括正常运行状态或异常运行状态。
可选的,所述根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态,具体包括:
根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定所述重构数据;
根据所述待检测的卫星遥测数据和所述重构数据,采用马氏距离,确 定重构误差;
获取误差阈值;
判断所述重构误差是否小于所述误差阈值;
若所述重构误差小于所述误差阈值,确定所述卫星的当前运行状态为正常运行状态;
若所述重构误差不小于所述误差阈值,确定所述卫星的当前运行状态为异常运行状态。
可选的,所述获取误差阈值,具体包括:
获取数据库中的一个卫星周期内的卫星遥测样本数据;
根据所述卫星遥测样本数据,采用小波方差,对所述卫星遥测样本数据进行分析,确定每一时刻的误差阈值。
可选的,所述在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器,具体包括:
根据所述变分自编码器中的编码器确定所述生成式对抗网络的生成器;所述生成器用于优化所述变分自编码器中的编码器,确定优化后的编码器;
根据双向长短时记忆网络确定所述生成式对抗网络的判别器;
根据所述判别器对所述优化后的编码器编码后的所述隐变量进行判别,确定优化后的隐变量;
根据所述优化后的隐变量以及所述变分自编码器中的解码器确定优化后的变分自编码器。
一种对抗网络自编码器的卫星异常检测系统,包括:
第一获取模块,用于获取变分自编码器和生成式对抗网络;所述变分自编码器以卫星遥测数据为输入,以重构数据为输出;所述变分自编码器包括编码器和解码器;所述生成式对抗网络包括生成器和判别器;
优化后的变分自编码器确定模块,用于在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器;所述优化后的变分自编码器包括优化后的编码器和所述解码器;所述优化后的编码器用于将所述待检测的卫星遥测数据编码为隐变量;所述解码器用于将所述隐变量解码为所述重构数据;
第二获取模块,用于获取待检测的卫星遥测数据;
卫星的当前运行状态确定模块,用于根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态;所述当前运行状态包括正常运行状态或异常运行状态。
可选的,所述卫星的当前运行状态确定模块具体包括:
重构数据确定单元,用于根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定所述重构数据;
重构误差确定单元,用于根据所述待检测的卫星遥测数据和所述重构数据,采用马氏距离,确定重构误差;
误差阈值获取单元,用于获取误差阈值;
判断单元,用于判断所述重构误差是否小于所述误差阈值;
正常运行状态确定单元,用于若所述重构误差小于所述误差阈值,确定所述卫星的当前运行状态为正常运行状态;
异常运行状态确定单元,用于若所述重构误差不小于所述误差阈值, 确定所述卫星的当前运行状态为异常运行状态。
可选的,所述误差阈值获取单元具体包括:
卫星遥测样本数据获取子单元,用于获取数据库中的一个卫星周期内的卫星遥测样本数据;
每一时刻的误差阈值确定子单元,用于根据所述卫星遥测样本数据,采用小波方差,对所述卫星遥测样本数据进行分析,确定每一时刻的误差阈值。
可选的,所述优化后的变分自编码器确定模块具体包括:
生成器确定单元,用于根据所述变分自编码器中的编码器确定所述生成式对抗网络的生成器;所述生成器用于优化所述变分自编码器中的编码器,确定优化后的编码器;
判别器确定单元,根据双向长短时记忆网络确定所述生成式对抗网络的判别器;
优化后的变分自编码器确定单元,用于根据所述优化后的隐变量以及所述变分自编码器中的解码器确定优化后的变分自编码器。
本发明与现有技术相比的优点在于:本发明提供一种对抗网络自编码器的卫星异常检测方法及系统,其中通过在变分编码器中引入对抗网络,得到优化后的变分编码器,通过优化后的变分编码器实现对卫星遥测数据的检测,本发明采用纯数据驱动,无需专家经验,能够适用多种场合;结合变分自编码器和生成对抗网络各自的优点,快速准确的检测出异常数据,即检测出卫星的异常状态,解决现有技术中无法实现自动检测和检测误差高的问题,保证了卫星异常自动检测的准确度。
说明书附图
下面结合附图对本发明作进一步说明:
图1为本发明所提供的一种对抗网络自编码器的卫星异常检测方法流程示意图;
图2为本发明所提供的循环神经网络结构示意图;
图3为本发明所提供的一种对抗网络自编码器的卫星异常检测系统结构示意图。
具体实施方式
本发明的基本思想是:采用自编码器重构数据,依据重构误差来判定异常。引入对抗网络,结合变分自编码器和生成对抗网络两者的优点,使得重构模型更加精确,降低重构模型的误差,降低了异常检测的误差。对于异常判断,考虑到传感器冗余而且不依靠专家经验,采用马氏距离进行计算重构误差。对数据进行周期分析,进一步的提高重构数据的准确性。
基于所述的一种对抗网络自编码器的卫星异常检测方法及系统,本发明结合具体实施例,对本发明作进一步详细说明。
图1为本发明所提供的一种对抗网络自编码器的卫星异常检测方法流程示意图,如图1所示,本发明所提供的一种对抗网络自编码器的卫星异常检测方法,包括:
S101,获取变分自编码器和生成式对抗网络;所述变分自编码器以卫星遥测数据为输入,以重构数据为输出;所述变分自编码器包括编码器和解码器;所述生成式对抗网络包括生成器和判别器。
所述变分自编码器的损失函数为
Figure PCTCN2020075658-appb-000001
其中,L VEA为变分自编码器的误差,L R为样本的重构误差,
Figure PCTCN2020075658-appb-000002
为后验概率和真实后验概率的KL散度。
图2为单向的循环神经网络结构示意图,而单向的循环神经网络只能在一个方向处理数据。为了在两个方向上处理数据,本发明采用长短时记忆网络作为生成式对抗网络中的判别器。
其中,利用公式
Figure PCTCN2020075658-appb-000003
确定自前向后循环神经网络层的更新。
利用公式
Figure PCTCN2020075658-appb-000004
确定字后向前循环神经网络层的更新。
利用公式
Figure PCTCN2020075658-appb-000005
确定两层循环神经网络层的叠加,并输入隐藏层。
利用公式
Figure PCTCN2020075658-appb-000006
确定生成式对抗网络的目标函数。
P data和P z分别是真实数据和生成数据的分布,DIS为生成器,GEN为判别器,x为输入数据,z为编码器生成的隐变量。
在训练判别器时,真实数据判别值越接近1越好,对生成数据的判别值越接近0越好。在训练生成器时,希望生成数据能够欺骗判别器,让生成数据的判别值接近1。生成器训练刚开始时,生成数据和真实数据的分布有很大差异,判别器模型的判别结果处于上下摆动状态;然后,重复固定生成器,优化判别器,以及固定判别器,训练生成器这一过程,直到判别器无法判别生成数据的真假。生成器的目的就是通过不断训练生成模型,使P z越来越接近P data,并且构建一个判别器来区分生成数据 和真实数据。
S102,在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器;所述优化后的变分自编码器包括优化后的编码器和所述解码器;所述优化后的编码器用于将所述待检测的卫星遥测数据编码为隐变量;所述解码器用于将所述隐变量解码为所述重构数据。
根据所述变分自编码器中的编码器确定所述生成式对抗网络的生成器;所述生成器用于优化所述变分自编码器中的编码器,确定优化后的编码器。
根据双向长短时记忆网络确定所述生成式对抗网络的判别器。
根据所述判别器对所述优化后的编码器编码后的所述隐变量进行判别,确定优化后的隐变量。
根据所述优化后的隐变量以及所述变分自编码器中的解码器确定优化后的变分自编码器。
S103,获取待检测的卫星遥测数据。所述卫星遥测数据随着卫星周期性的变化而变化。
S104,根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态;所述当前运行状态包括正常运行状态或异常运行状态。
根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定所述重构数据。
根据所述待检测的卫星遥测数据和所述重构数据,采用马氏距离,确定重构误差。
利用公式
Figure PCTCN2020075658-appb-000007
确定重构误差。其中x为待检测的卫星遥测数据,μ为卫星遥测数据均值,∑ -1表示卫星遥测数据空间的协方差矩阵的逆矩阵。
为了进一步的提高重构误差的精确度,在马氏距离确定重构误差的基础上,确定重构误差的得分。
即利用公式
Figure PCTCN2020075658-appb-000008
确定重构误差的得分。x(i)为第i个卫星遥测数据,
Figure PCTCN2020075658-appb-000009
为经过自编码器重构的第i个卫星遥测数据。
RE score(i)在高维空间中是一种相对鲁棒性的重构误差度量值,可以反应卫星遥测数据重构误差变化的趋势。为了监测这一趋势并且检测异常变动需要设定误差阈值。即通过误差阈值的设置提供重构误差的准确性。
获取误差阈值。
判断所述重构误差是否小于所述误差阈值。
若所述重构误差小于所述误差阈值,确定所述卫星的当前运行状态为正常运行状态。
若所述重构误差不小于所述误差阈值,确定所述卫星的当前运行状态为异常运行状态。
卫星遥测数据反映了卫星在轨运行的当前状态,由于卫星运行具有周期性,因此,卫星遥测数据也会随着卫星周期而呈现周期性变化,准确地分析出卫星遥测数据变化周期对其异常检测研究具有重要意义。由于卫星遥测数据维度高,数据量大,无法直接观察出卫星遥测数据的周期。通过卫星遥测数据的周期获取误差阈值具体包括:
获取数据库中的一个卫星周期内的卫星遥测样本数据;根据所述卫星遥测样本数据,采用小波方差,对所述卫星遥测样本数据进行分析,确定每一时刻的误差阈值。
根据所述卫星遥测样本数据,采用小波方差,对所述卫星遥测样本数据进行分析,确定每一时刻的误差阈值的具体过程为:
(1)小波函数满足
Figure PCTCN2020075658-appb-000010
利用公式
Figure PCTCN2020075658-appb-000011
确定一簇小波函数系。
(2)利用公式
Figure PCTCN2020075658-appb-000012
对小波函数进行离散变换。
(3)利用公式
Figure PCTCN2020075658-appb-000013
确定小波方差。
根据小波方差放映的卫星遥测数据波动的能量分布,确定卫星遥测数据的方差。
根据卫星轨道运转的特点,对一个周期内的卫星遥测数据进行分析。先将一个前向周期窗口内收集到的数据统计得到97%上分位点,再将该点放大α倍,利用U i=y i(1+α)确定第i个输入量对应的上界。其中,y i是第i个输入量前向一个周期内97%上分位点,α是未知的比例系数。
为了获得最优的动态误差阈值区间,将近期历史数据构成一个新的预测样本,该预测样本称为验证集,通过时间窗口动态构建验证集的动态误差阈值区间。
图3为本发明所提供的一种对抗网络自编码器的卫星异常检测系统结构示意图,如图3所示,本发明所提供的一种对抗网络自编码器的卫星异常检测系统,包括:第一获取模块301、优化后的变分自编码器确定 模块302、第二获取模块303和卫星的当前运行状态确定模块304。
第一获取模块301用于获取变分自编码器和生成式对抗网络;所述变分自编码器以卫星遥测数据为输入,以重构数据为输出;所述变分自编码器包括编码器和解码器;所述生成式对抗网络包括生成器和判别器。
优化后的变分自编码器确定模块302用于在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器;所述优化后的变分自编码器包括优化后的编码器和所述解码器;所述优化后的编码器用于将所述待检测的卫星遥测数据编码为隐变量;所述解码器用于将所述隐变量解码为所述重构数据。
第二获取模块303用于获取待检测的卫星遥测数据。
卫星的当前运行状态确定模块304用于根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态;所述当前运行状态包括正常运行状态或异常运行状态。
所述卫星的当前运行状态确定模块304具体包括:重构数据确定单元、重构误差确定单元、误差阈值获取单元判断单元、正常运行状态确定单元和异常运行状态确定单元。
重构数据确定单元用于根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定所述重构数据。
重构误差确定单元用于根据所述待检测的卫星遥测数据和所述重构数据,采用马氏距离,确定重构误差。
误差阈值获取单元用于获取误差阈值。
判断单元用于判断所述重构误差是否小于所述误差阈值。
正常运行状态确定单元用于若所述重构误差小于所述误差阈值,确定所述卫星的当前运行状态为正常运行状态。
异常运行状态确定单元用于若所述重构误差不小于所述误差阈值,确定所述卫星的当前运行状态为异常运行状态。
所述误差阈值获取单元具体包括:卫星遥测样本数据获取子单元和每一时刻的误差阈值确定子单元。
卫星遥测样本数据获取子单元用于获取数据库中的一个卫星周期内的卫星遥测样本数据。
每一时刻的误差阈值确定子单元用于根据所述卫星遥测样本数据,采用小波方差,对所述卫星遥测样本数据进行分析,确定每一时刻的误差阈值。
所述优化后的变分自编码器确定模块302具体包括:生成器确定单元、判别器确定单元、优化后的隐变量确定单元和优化后的变分自编码器确定单元。
生成器确定单元用于根据所述变分自编码器中的编码器确定所述生成式对抗网络的生成器;所述生成器用于优化所述变分自编码器中的编码器,确定优化后的编码器。
判别器确定单元根据双向长短时记忆网络确定所述生成式对抗网络的判别器。
优化后的隐变量确定单元用于根据所述判别器对所述优化后的编码器编码后的所述隐变量进行判别,确定优化后的隐变量。
优化后的变分自编码器确定单元用于根据所述优化后的隐变量以及
所述变分自编码器中的解码器确定优化后的变分自编码器。
提供以上实施例仅仅是为了描述本发明的目的,而并非要限制本发明的范围。本发明的范围由所附权利要求限定。不脱离本发明的精神和原理而做出的各种等同替换和修改,均应涵盖在本发明的范围之内。

Claims (8)

  1. 一种对抗网络自编码器的卫星异常检测方法,其特征在于,包括:
    获取变分自编码器和生成式对抗网络;所述变分自编码器以卫星遥测数据为输入,以重构数据为输出;所述变分自编码器包括编码器和解码器;所述生成式对抗网络包括生成器和判别器;
    在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器;所述优化后的变分自编码器包括优化后的编码器和所述解码器;所述优化后的编码器用于将所述待检测的卫星遥测数据编码为隐变量;所述解码器用于将所述隐变量解码为所述重构数据;
    获取待检测的卫星遥测数据;
    根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态;所述当前运行状态包括正常运行状态或异常运行状态。
  2. 根据权利要求1所述的一种对抗网络自编码器的卫星异常检测方法,其特征在于,所述根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态,具体包括:
    根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定所述重构数据;
    根据所述待检测的卫星遥测数据和所述重构数据,采用马氏距离,确定重构误差;
    获取误差阈值;
    判断所述重构误差是否小于所述误差阈值;
    若所述重构误差小于所述误差阈值,确定所述卫星的当前运行状态为 正常运行状态;
    若所述重构误差不小于所述误差阈值,确定所述卫星的当前运行状态为异常运行状态。
  3. 根据权利要求2所述的一种对抗网络自编码器的卫星异常检测方法,其特征在于,所述获取误差阈值,具体包括:
    获取数据库中的一个卫星周期内的卫星遥测样本数据;
    根据所述卫星遥测样本数据,采用小波方差,对所述卫星遥测样本数据进行分析,确定每一时刻的误差阈值。
  4. 根据权利要求1所述的一种对抗网络自编码器的卫星异常检测方法,其特征在于,所述在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器,具体包括:
    根据所述变分自编码器中的编码器确定所述生成式对抗网络的生成器;所述生成器用于优化所述变分自编码器中的编码器,确定优化后的编码器;
    根据双向长短时记忆网络确定所述生成式对抗网络的判别器;
    根据所述判别器对所述优化后的编码器编码后的所述隐变量进行判别,确定优化后的隐变量;
    根据所述优化后的隐变量以及所述变分自编码器中的解码器确定优化后的变分自编码器。
  5. 一种对抗网络自编码器的卫星异常检测系统,其特征在于,包括:
    第一获取模块,用于获取变分自编码器和生成式对抗网络;所述变分自编码器以卫星遥测数据为输入,以重构数据为输出;所述变分自编码器 包括编码器和解码器;所述生成式对抗网络包括生成器和判别器;
    优化后的变分自编码器确定模块,用于在所述变分自编码器中加入所述生成式对抗网络,确定优化后的变分自编码器;所述优化后的变分自编码器包括优化后的编码器和所述解码器;所述优化后的编码器用于将所述待检测的卫星遥测数据编码为隐变量;所述解码器用于将所述隐变量解码为所述重构数据;
    第二获取模块,用于获取待检测的卫星遥测数据;
    卫星的当前运行状态确定模块,用于根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定卫星的当前运行状态;所述当前运行状态包括正常运行状态或异常运行状态。
  6. 根据权利要求5所述的一种对抗网络自编码器的卫星异常检测系统,其特征在于,所述卫星的当前运行状态确定模块具体包括:
    重构数据确定单元,用于根据所述待检测的卫星遥测数据,采用所述优化后的变分自编码器,确定所述重构数据;
    重构误差确定单元,用于根据所述待检测的卫星遥测数据和所述重构数据,采用马氏距离,确定重构误差;
    误差阈值获取单元,用于获取误差阈值;
    判断单元,用于判断所述重构误差是否小于所述误差阈值;
    正常运行状态确定单元,用于若所述重构误差小于所述误差阈值,确定所述卫星的当前运行状态为正常运行状态;
    异常运行状态确定单元,用于若所述重构误差不小于所述误差阈值,确定所述卫星的当前运行状态为异常运行状态。
  7. 根据权利要求6所述的一种对抗网络自编码器的卫星异常检测系统,其特征在于,所述误差阈值获取单元具体包括:
    卫星遥测样本数据获取子单元,用于获取数据库中的一个卫星周期内的卫星遥测样本数据;
    每一时刻的误差阈值确定子单元,用于根据所述卫星遥测样本数据,采用小波方差,对所述卫星遥测样本数据进行分析,确定每一时刻的误差阈值。
  8. 根据权利要求5所述的一种对抗网络自编码器的卫星异常检测系统,其特征在于,所述优化后的变分自编码器确定模块具体包括:
    生成器确定单元,用于根据所述变分自编码器中的编码器确定所述生成式对抗网络的生成器;所述生成器用于优化所述变分自编码器中的编码器,确定优化后的编码器;
    判别器确定单元,根据双向长短时记忆网络确定所述生成式对抗网络的判别器;
    优化后的变分自编码器确定单元,用于根据所述优化后的隐变量以及所述变分自编码器中的解码器确定优化后的变分自编码器。
PCT/CN2020/075658 2019-03-13 2020-02-18 一种对抗网络自编码器的卫星异常检测方法及系统 WO2020181962A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/255,267 US12093804B2 (en) 2019-03-13 2020-02-18 Satellite anomaly detection method and system for adversarial network auto-encoder

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910195659.X 2019-03-13
CN201910195659.XA CN109948117B (zh) 2019-03-13 2019-03-13 一种对抗网络自编码器的卫星异常检测方法

Publications (1)

Publication Number Publication Date
WO2020181962A1 true WO2020181962A1 (zh) 2020-09-17

Family

ID=67010161

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/075658 WO2020181962A1 (zh) 2019-03-13 2020-02-18 一种对抗网络自编码器的卫星异常检测方法及系统

Country Status (3)

Country Link
US (1) US12093804B2 (zh)
CN (1) CN109948117B (zh)
WO (1) WO2020181962A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113835106A (zh) * 2020-12-03 2021-12-24 北京航空航天大学 一种大数据条件下的关联性健康基线无监督自主构建方法
US20220060235A1 (en) * 2020-08-18 2022-02-24 Qualcomm Incorporated Federated learning for client-specific neural network parameter generation for wireless communication
CN116151123A (zh) * 2023-03-02 2023-05-23 中国科学院上海天文台 磁测卫星异常状态智能检测与识别方法
US20230182927A1 (en) * 2021-12-10 2023-06-15 Mitsubishi Electric Research Laboratories, Inc. System and Method for Controlling a Motion of a Spacecraft in a Multi-Object Celestial System

Families Citing this family (55)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109948117B (zh) 2019-03-13 2023-04-07 南京航空航天大学 一种对抗网络自编码器的卫星异常检测方法
CN110572696B (zh) * 2019-08-12 2021-04-20 浙江大学 一种变分自编码器与生成对抗网络结合的视频生成方法
CN110598851A (zh) * 2019-08-29 2019-12-20 北京航空航天大学合肥创新研究院 一种融合lstm和gan的时间序列数据异常检测方法
CN110704508B (zh) * 2019-09-30 2023-04-07 佛山科学技术学院 一种智能生产线异常数据的处理方法及装置
CN111132209B (zh) * 2019-12-04 2023-05-05 东南大学 基于变分自编码器估计无线局域网络接入点吞吐量的方法
CN110992354B (zh) * 2019-12-13 2022-04-12 华中科技大学 基于引入自动记忆机制对抗自编码器的异常区域检测方法
CN111031051B (zh) * 2019-12-17 2021-03-16 清华大学 一种网络流量异常检测方法及装置、介质
US11545255B2 (en) * 2019-12-20 2023-01-03 Shanghai United Imaging Intelligence Co., Ltd. Systems and methods for classifying an anomaly medical image using variational autoencoder
CN111136659B (zh) * 2020-01-15 2022-06-21 南京大学 基于第三人称模仿学习的机械臂动作学习方法及系统
CN111277603B (zh) * 2020-02-03 2021-11-19 杭州迪普科技股份有限公司 无监督异常检测系统和方法
CN110991625B (zh) * 2020-03-02 2020-06-16 南京邮电大学 基于循环神经网络的地表异常现象遥感监测方法、装置
US11301724B2 (en) * 2020-04-30 2022-04-12 Robert Bosch Gmbh Semantic adversarial generation based function testing method in autonomous driving
CN111695598B (zh) * 2020-05-11 2022-04-29 东南大学 一种基于生成对抗网络的监测数据异常诊断方法
CN113837351B (zh) * 2020-06-08 2024-04-23 爱思开海力士有限公司 新异检测器
CN111815561B (zh) * 2020-06-09 2024-04-16 中海石油(中国)有限公司 一种基于深度时空特征的管道缺陷及管道组件的检测方法
CN111696066B (zh) * 2020-06-13 2022-04-19 中北大学 基于改进wgan-gp的多波段图像同步融合与增强方法
CN112179691B (zh) * 2020-09-04 2021-07-13 西安交通大学 基于对抗学习策略的机械装备运行状态异常检测系统和方法
CN112461537B (zh) * 2020-10-16 2022-06-17 浙江工业大学 基于长短时神经网络与自动编码机的风电齿轮箱状态监测方法
CN112149757B (zh) * 2020-10-23 2022-08-19 新华三大数据技术有限公司 一种异常检测方法、装置、电子设备及存储介质
CN112327331A (zh) * 2020-11-02 2021-02-05 中山大学 一种gnss欺骗干扰检测方法、装置、设备和存储介质
CN112597831A (zh) * 2021-02-22 2021-04-02 杭州安脉盛智能技术有限公司 一种基于变分自编码器和对抗网络的信号异常检测方法
CN113156473B (zh) * 2021-03-04 2023-05-02 中国北方车辆研究所 信息融合定位系统卫星信号环境的自适应判别方法
CN113011476B (zh) * 2021-03-05 2022-11-11 桂林电子科技大学 基于自适应滑动窗口gan的用户行为安全检测方法
US11843623B2 (en) * 2021-03-16 2023-12-12 Mitsubishi Electric Research Laboratories, Inc. Apparatus and method for anomaly detection
CN113159947A (zh) * 2021-03-17 2021-07-23 同济大学 一种基于生成对抗网络的困难异常样本检测框架
US20220303158A1 (en) * 2021-03-19 2022-09-22 NEC Laboratories Europe GmbH End-to-end channel estimation in communication networks
CN112949753B (zh) * 2021-03-26 2023-10-24 西安交通大学 一种基于二元关系的卫星遥测时序数据异常检测方法
CN113127705B (zh) * 2021-04-02 2022-08-05 西华大学 一种异构双向生成对抗网络模型及时间序列异常检测方法
CN113312809B (zh) * 2021-04-06 2022-12-13 北京航空航天大学 一种基于相关团划分的航天器遥测数据多参数异常检测方法
CN113077005B (zh) * 2021-04-13 2024-04-05 西安交通大学 一种基于lstm自编码器和正常信号数据的异常检测系统及方法
CN114075979B (zh) * 2021-04-21 2024-09-17 卢靖 盾构掘进环境变化实时辨识系统及方法
CN113407425B (zh) * 2021-05-13 2022-09-23 桂林电子科技大学 基于BiGAN与OTSU的内部用户行为检测方法
CN113240011B (zh) * 2021-05-14 2023-04-07 烟台海颐软件股份有限公司 一种深度学习驱动的异常识别与修复方法及智能化系统
CN113360694B (zh) * 2021-06-03 2022-09-27 安徽理工大学 一种基于自编码器的恶意图像查询样本检测过滤方法
CN113485863B (zh) * 2021-07-14 2023-05-05 北京航空航天大学 基于改进生成对抗网络生成异构不平衡故障样本的方法
CN114118224B (zh) * 2021-11-02 2024-04-12 中国运载火箭技术研究院 一种基于神经网络的全系统遥测参数异常检测系统
CN114118460A (zh) * 2021-11-05 2022-03-01 国网江苏省电力有限公司南京供电分公司 基于变分自编码器的低压台区线损率异常检测方法及装置
CN114065862B (zh) * 2021-11-18 2024-02-13 南京航空航天大学 一种多维时序数据异常检测方法和系统
CN114091600B (zh) * 2021-11-18 2024-01-12 南京航空航天大学 一种数据驱动的卫星关联故障传播路径辨识方法及系统
CN114297936A (zh) * 2021-12-31 2022-04-08 深圳前海微众银行股份有限公司 一种数据异常检测方法及装置
CN114549930B (zh) * 2022-02-21 2023-01-10 合肥工业大学 一种基于轨迹数据的快速路短时车头间距预测方法
CN114745187B (zh) * 2022-04-19 2022-11-01 中国人民解放军战略支援部队航天工程大学 一种基于pop流量矩阵的内部网络异常检测方法及系统
CN115314287B (zh) * 2022-08-08 2024-08-20 大连理工大学 一种基于深度聚类的对抗异常检测系统
CN115859202B (zh) * 2022-11-24 2023-10-10 浙江邦盛科技股份有限公司 一种非平稳时序数据流场景下的异常检测方法及装置
CN115543589A (zh) * 2022-12-05 2022-12-30 成都国星宇航科技股份有限公司 星上分系统备份的调用方法、系统、设备及存储介质
CN115622888B (zh) * 2022-12-19 2023-03-07 中国人民解放军国防科技大学 基于多学科协作逆向优化的跨域融合星座设计方法
CN115964636B (zh) * 2022-12-23 2023-11-07 浙江苍南仪表集团股份有限公司 基于机器学习和动态阈值的燃气流量异常检测方法及系统
CN117113165A (zh) * 2023-07-07 2023-11-24 上海蓝箭鸿擎科技有限公司 一种卫星报警数据的处理方法及计算机存储介质
CN116702083B (zh) * 2023-08-10 2023-12-26 武汉能钠智能装备技术股份有限公司四川省成都市分公司 一种卫星遥测数据异常检测方法及系统
CN116743646B (zh) * 2023-08-15 2023-12-19 云南省交通规划设计研究院股份有限公司 一种基于域自适应深度自编码器隧道网络异常检测方法
CN117134818B (zh) * 2023-10-27 2024-02-02 亚太卫星宽带通信(深圳)有限公司 一种高低双轨道卫星
CN117235655B (zh) * 2023-11-15 2024-02-02 北明天时能源科技(北京)有限公司 基于联邦学习的智慧供热异常工况识别方法及系统
CN117692346A (zh) * 2024-01-31 2024-03-12 浙商银行股份有限公司 基于谱正则化变分自编码器的消息阻塞预测方法及装置
CN117706360A (zh) * 2024-02-02 2024-03-15 深圳市昱森机电有限公司 电机运行状态的监测方法、装置、设备及存储介质
CN117914629B (zh) * 2024-03-18 2024-05-28 台州市大数据发展有限公司 一种网络安全检测方法及系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107123151A (zh) * 2017-04-28 2017-09-01 深圳市唯特视科技有限公司 一种基于变分自动编码器和生成对抗网络的图像转化方法
CN107358195A (zh) * 2017-07-11 2017-11-17 成都考拉悠然科技有限公司 基于重建误差的非特定异常事件检测及定位方法、计算机
CN107991876A (zh) * 2017-12-14 2018-05-04 南京航空航天大学 基于生成式对抗网络的航空发动机状态监测数据生成方法
CN109447263A (zh) * 2018-11-07 2019-03-08 任元 一种基于生成对抗网络的航天异常事件检测方法
CN109948117A (zh) * 2019-03-13 2019-06-28 南京航空航天大学 一种对抗网络自编码器的卫星异常检测方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR101364831B1 (ko) * 2010-11-08 2014-02-20 한국전자통신연구원 텔레메트리 데이터의 통계적 분석을 이용하여 위성중계기의 상태를 감시하는 장치 및 방법
CN106650297B (zh) * 2017-01-06 2019-04-19 南京航空航天大学 一种无领域知识的卫星分系统异常检测方法
CN108537742B (zh) * 2018-03-09 2021-07-09 天津大学 一种基于生成对抗网络的遥感图像全色锐化方法
US11710033B2 (en) * 2018-06-12 2023-07-25 Bank Of America Corporation Unsupervised machine learning system to automate functions on a graph structure

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107123151A (zh) * 2017-04-28 2017-09-01 深圳市唯特视科技有限公司 一种基于变分自动编码器和生成对抗网络的图像转化方法
CN107358195A (zh) * 2017-07-11 2017-11-17 成都考拉悠然科技有限公司 基于重建误差的非特定异常事件检测及定位方法、计算机
CN107991876A (zh) * 2017-12-14 2018-05-04 南京航空航天大学 基于生成式对抗网络的航空发动机状态监测数据生成方法
CN109447263A (zh) * 2018-11-07 2019-03-08 任元 一种基于生成对抗网络的航天异常事件检测方法
CN109948117A (zh) * 2019-03-13 2019-06-28 南京航空航天大学 一种对抗网络自编码器的卫星异常检测方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
KIRAN, B.R. ET AL.: "An Overview of Deep Learning Based Methods for Unsupervised and Semi-Supervised Anomaly Detection in Videos", JOURNAL OF IMAGING, vol. 4, no. 2, 7 February 2018 (2018-02-07), XP080851755, DOI: 20200506143947A *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220060235A1 (en) * 2020-08-18 2022-02-24 Qualcomm Incorporated Federated learning for client-specific neural network parameter generation for wireless communication
US11909482B2 (en) * 2020-08-18 2024-02-20 Qualcomm Incorporated Federated learning for client-specific neural network parameter generation for wireless communication
CN113835106A (zh) * 2020-12-03 2021-12-24 北京航空航天大学 一种大数据条件下的关联性健康基线无监督自主构建方法
CN113835106B (zh) * 2020-12-03 2023-08-01 北京航空航天大学 一种大数据条件下的关联性健康基线无监督自主构建方法
US20230182927A1 (en) * 2021-12-10 2023-06-15 Mitsubishi Electric Research Laboratories, Inc. System and Method for Controlling a Motion of a Spacecraft in a Multi-Object Celestial System
CN116151123A (zh) * 2023-03-02 2023-05-23 中国科学院上海天文台 磁测卫星异常状态智能检测与识别方法
CN116151123B (zh) * 2023-03-02 2024-10-15 中国科学院上海天文台 磁测卫星异常状态智能检测与识别方法

Also Published As

Publication number Publication date
US20210174171A1 (en) 2021-06-10
CN109948117B (zh) 2023-04-07
US12093804B2 (en) 2024-09-17
CN109948117A (zh) 2019-06-28

Similar Documents

Publication Publication Date Title
WO2020181962A1 (zh) 一种对抗网络自编码器的卫星异常检测方法及系统
CN103996077B (zh) 一种基于多维时间序列的电气设备故障预测方法
JP4308437B2 (ja) センサの性能確認装置および方法
CN109522948A (zh) 一种基于正交局部保持投影的故障检测方法
CN110751199A (zh) 一种基于贝叶斯神经网络的卫星异常检测方法
US12119114B2 (en) Missing medical diagnosis data imputation method and apparatus, electronic device and medium
US11480956B2 (en) Computing an explainable event horizon estimate
CN111401471A (zh) 一种航天器姿态异常检测方法及系统
CN103631145B (zh) 基于监控指标切换的多工况过程监控方法和系统
CN110348150A (zh) 一种基于相关概率模型的故障检测方法
CN104156615A (zh) 基于ls-svm的传感器检测数据点异常检测方法
CN111368428A (zh) 一种基于监控二阶统计量的传感器精度下降故障检测方法
Ou et al. A deep sequence multi-distribution adversarial model for bearing abnormal condition detection
CN113283113B (zh) 太阳电池阵发电电流预测模型训练方法、异常检测方法、设备及介质
CN117907631A (zh) 一种基于风速传感器的风速矫正方法及系统
CN110455370B (zh) 防汛抗旱远程遥测显示系统
CN117310829A (zh) 一种基于地磁异常数据的优化识别方法
Liu et al. Graph attention Network-Based model for multiple fault detection and identification of sensors in nuclear power plant
Chen et al. A deep auto-encoder satellite anomaly advance warning framework
Baraldi et al. A fuzzy similarity based method for signal reconstruction during plant transients
Lin et al. Fault detection and isolation for multi-type sensors in nuclear power plants via a knowledge-guided spatial–temporal model
Shao et al. Data-model-linked remaining useful life prediction method with small sample data: A case of subsea valve
Al Rasyid et al. Anomaly detection in wireless body area network using Mahalanobis distance and sequential minimal optimization regression
Huang et al. Condition monitoring and fault diagnosis of hydropower generator based on LSTM correction model
CN118152993B (zh) 一种基于物联网的智能水利资源感知系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20770634

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20770634

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 22/04/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20770634

Country of ref document: EP

Kind code of ref document: A1