A method, system, device and storage medium for detecting device anomaly

Abstract

The embodiment of the present invention provides a method, system, device and storage medium for detecting device abnormalities. The method includes: within a preset time length, using a three-light acquisition device to collect data from a target device, obtaining a three-light fusion image sequence data, inputting it into an abnormal feature extraction model for feature extraction, obtaining multivariate time series sequence data, performing subsequence division on part or all of the data in the multivariate time series sequence data, obtaining multivariate subsequence segment data, inputting it into a preset abnormal detection model for error judgment, and obtaining the abnormal state judgment result of the target device. By subsequencing the multivariate time series sequence data after abnormal feature extraction, multivariate subsequence segment data reflecting device abnormalities is obtained on the basis of ensuring data accuracy, so that the present invention is improved in both detection accuracy and detection efficiency compared with the prior art. It can be seen that the present invention improves the efficiency of abnormal detection of power equipment.

Description

Technical Field

[0001] The present invention relates to the field of equipment detection, and in particular to a method, system, equipment and storage medium for equipment abnormality detection. Background Art

[0002] At present, when abnormality detection is performed on power equipment, it is usually performed by operation and maintenance personnel based on the operation and inspection parameters of the power equipment, or by light wave observation equipment to detect the abnormal state of the power equipment. Among them, when the operation and maintenance personnel use light wave observation equipment to detect the abnormal state, they use the light wave observation equipment to detect infrared light, ultraviolet light or visible light images of the power equipment.

[0003] However, the existing technology requires operation and maintenance personnel to carry the above three types of optical wave observation equipment respectively, collect image information of the three optical wave spectrum segments of the power equipment respectively, and then manually perform fault analysis on the three types of image information to determine whether the power equipment is currently in an abnormal state. When the features in the image information are not obvious, manual judgment is prone to misjudgment. This leads to a decrease in the efficiency of the existing technology for abnormal detection of power equipment. Therefore, how to improve the efficiency of abnormal detection of power equipment is a problem that needs to be solved urgently. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a method, system, device and storage medium for detecting abnormalities of equipment, so as to improve the efficiency of detecting abnormalities of power equipment. The specific technical solution is as follows:

[0005] A method for detecting anomalies in equipment, the method comprising:

[0006] Within a preset time period, a three-light acquisition device is used to acquire data from a target device to obtain three-light fusion image sequence data, wherein the three-light acquisition device is a device that can simultaneously acquire image data of three light spectrum segments of the target device, and each three-light fusion image in the three-light fusion image sequence data carries image features of three light spectrum segments, and the three light spectrum segments include an infrared light spectrum segment, an ultraviolet light spectrum segment, and a visible light spectrum segment.

[0007] The three-light fusion image sequence data is input into an abnormal feature extraction model for feature extraction to obtain multivariate time series data.

[0008] Part or all of the multivariate time series data is divided into subsequences according to a preset window and a preset step size to obtain multivariate subsequence segment data.

[0009] The multivariable subsequence segment data is input into a preset anomaly detection model, so that the preset anomaly detection model performs error judgment on the multivariable subsequence segment data to obtain an abnormal state judgment result of the target device.

[0010] Optionally, the preset anomaly detection model is composed of a preset generation model, a preset discrimination model and a result fusion model, and the multivariate subsequence segment data is input into the preset anomaly detection model so that the preset anomaly detection model performs error judgment on the multivariate subsequence segment data to obtain the abnormal state judgment result of the target device, including:

[0011] The multivariate subsequence segment data is input into the preset generation model so that the preset generation model generates simulated samples according to the multivariate subsequence segment data, and calculates the reconstruction error between the simulated samples and the real samples of the normal operation of the target device.

[0012] The multivariate subsequence segment data is input into the preset discriminant model so that the preset discriminant model generates a discriminant error according to the multivariate subsequence segment data, wherein the discriminant error refers to whether the target device is normal or the target device is abnormal.

[0013] The reconstruction error and the discrimination error are input into the result fusion model for weighted result fusion to obtain a judgment result, wherein the judgment result is that the target device operates normally or the target device operates abnormally.

[0014] Optionally, the method further includes:

[0015] According to the characteristics of the time series data stored in the preset database, the preset window and the step size corresponding to part of the data or all of the data in the multivariate time series data are determined.

[0016] Optionally, the training process of the preset generation model includes:

[0017] The generation model is trained using the data of the normal operation of the target device as input to obtain the preset generation model, wherein the output of the preset generation model is the reconstruction error between the simulated sample and the real sample of the normal operation of the target device.

[0018] Optionally, the training process of the preset discriminant model includes:

[0019] The discriminant model is trained using the data of the normal operation of the target device and the data of the abnormal operation of the target device to obtain the preset discriminant model, wherein the output of the preset discriminant model is that the target device is normal or the target device is abnormal.

[0020] Optionally, the step of dividing part or all of the multivariate time series data into subsequences according to a preset window and a preset step size to obtain multivariate subsequence segment data includes:

[0021] The preset missing value processing algorithm and data normalization algorithm are used to remove missing data points and erroneous data points from the multivariate time series data to obtain first multivariate time series data.

[0022] The first multivariate time series data is subjected to dimensionality reduction and interference dimension removal processing using a feature selection algorithm to obtain the second multivariate time series data, wherein the number of data in the second multivariate time series data is not greater than the number of data in the multivariate time series data.

[0023] A system for detecting abnormality of equipment, the system comprising:

[0024] The data acquisition module is used to use a three-light acquisition device to collect data from a target device within a preset time period to obtain three-light fusion image sequence data, wherein the three-light acquisition device is a device that can simultaneously collect image data of three light spectrum segments of the target device, and each three-light fusion image in the three-light fusion image sequence data carries image features of three light spectrum segments, and the three light spectrum segments include an infrared light spectrum segment, an ultraviolet light spectrum segment, and a visible light spectrum segment.

[0025] The feature extraction module is used to input the three-light fusion image sequence data into the abnormal feature extraction model for feature extraction to obtain multivariate time series data.

[0026] The subsequence division module is used to divide part or all of the data in the multivariate time series data into subsequences according to a preset window and a preset step size to obtain multivariate subsequence segment data.

[0027] The result output module is used to input the multivariable subsequence segment data into a preset anomaly detection model so that the preset anomaly detection model performs error judgment on the multivariable subsequence segment data to obtain an abnormal state judgment result of the target device.

[0028] Optionally, the result output module is configured as follows:

[0029] The first error generation submodule is used to input the multivariate subsequence segment data into the preset generation model so that the preset generation model generates simulated samples according to the multivariate subsequence segment data, and calculates the reconstruction error between the simulated samples and the real samples of the normal operation of the target device.

[0030] The second error generation submodule is used to input the multivariate subsequence segment data into the preset discriminant model so that the preset discriminant model generates a discriminant error according to the multivariate subsequence segment data, wherein the discriminant error refers to whether the target device is normal or the target device is abnormal.

[0031] The result fusion submodule is used to input the reconstruction error and the discrimination error into the result fusion model for weighted result fusion to obtain a judgment result, wherein the judgment result is that the target device operates normally or the target device operates abnormally.

[0032] Optionally, the system further includes:

[0033] The data calling module is used to determine the preset window and the step size corresponding to part of or all of the data in the multivariate time series data according to the time series data characteristics stored in the preset database.

[0034] Optionally, the first error generation submodule includes: a first training submodule,

[0035] The first training submodule uses the data of normal operation of the target device as input to train the generation model to obtain the preset generation model, wherein the output of the preset generation model is the reconstruction error between the simulated sample and the real sample of normal operation of the target device.

[0036] Optionally, the second error generation submodule includes: a second training submodule,

[0037] The second training submodule trains the discrimination model using the data of the normal operation of the target device and the data of the abnormal operation of the target device to obtain the preset discrimination model, wherein the output of the preset discrimination model is that the target device is normal or the target device is abnormal.

[0038] Optionally, the subsequence division module includes:

[0039] The first preprocessing submodule is used to use a preset missing value processing algorithm and a data normalization algorithm to remove missing data points and erroneous data points from the multivariate time series data to obtain first multivariate time series data.

[0040] The second preprocessing submodule is used to use a feature selection algorithm to perform dimensionality reduction and interference dimension removal processing on the first multivariate time series data to obtain the second multivariate time series data, wherein the number of data in the second multivariate time series data is not greater than the number of data in the multivariate time series data.

[0041] A device for detecting device anomalies, comprising:

[0042] processor;

[0043] A memory for storing instructions executable by the processor.

[0044] The processor is configured to execute the instructions to implement the method for device abnormality detection as described in any one of the above items.

[0045] A computer-readable storage medium, when the instructions in the computer-readable storage medium are executed by the processor of the device, enables the device for device anomaly detection to perform any of the device anomaly detection methods described above.

[0046] The embodiments of the present invention provide a method, system, device and storage medium for detecting device anomalies.

[0047] By sub-sequencing the multivariate time series data after abnormal feature extraction according to the preset window and preset step size, the multivariate sub-sequence segment data that can reflect the abnormal state of the target device is obtained on the basis of ensuring data accuracy, so that the present invention is improved in detection accuracy and detection efficiency compared with the prior art method that requires manual identification of abnormal states in images. At the same time, the present invention uses a preset abnormal detection model to make error judgments on the multivariate sub-sequence segment data, so that the present invention further improves the detection efficiency compared with the manual judgment method in the prior art. Finally, the present invention uses a three-light acquisition device to obtain a three-light fusion image sequence data, replacing the prior art method that requires operation and maintenance personnel to carry three types of light wave observation equipment for image acquisition, so that the present invention simplifies the detection data acquisition process and improves the detection efficiency compared with the prior art. It can be seen that the present invention improves the efficiency of abnormal detection of power equipment.

[0048] Of course, it is not necessary to achieve all of the advantages described above at the same time to implement any product or method of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A flowchart of a method for detecting device abnormality provided by an embodiment of the present invention;

[0051] Figure 2A block diagram of a system for detecting device anomalies provided by an embodiment of the present invention;

[0052] Figure 3 A block diagram of a device for detecting device abnormalities provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] The embodiment of the present invention provides a method for detecting device abnormality. Figure 1 As shown, the method includes:

[0055] S101. Within a preset time period, use a three-light acquisition device to collect data from a target device to obtain three-light fusion image sequence data, wherein the three-light acquisition device is a device that can simultaneously collect image data of three light spectrum segments of the target device, and each three-light fusion image in the three-light fusion image sequence data carries image features of three light spectrum segments, and the three light spectrum segments include an infrared light spectrum segment, an ultraviolet light spectrum segment, and a visible light spectrum segment.

[0056] Optionally, the above-mentioned three-light acquisition device is an image acquisition device that integrates three light wave channels of infrared light, ultraviolet light and visible light. The three-light acquisition device can output three-light fusion image sequence data, and fuse the images of the three light spectrum segments through a preset registration algorithm and a preset time domain frame alignment algorithm to obtain three-light fusion image sequence data. The present invention uses a three-light acquisition device to collect data from the target device, so that compared with the prior art, the present invention requires the operation and maintenance personnel to carry three light wave observation devices to obtain the target device image, and manually analyze the three light wave images, thereby improving work efficiency. The above-mentioned features are used to reflect abnormal conditions of the target device.

[0057] S102, inputting the three-light fusion image sequence data into an abnormal feature extraction model for feature extraction to obtain multivariate time series data.

[0058] Optionally, in an optional embodiment of the present invention, the multivariate time series data may be sequence data corresponding to multiple abnormal features at the same time point. The abnormal features include but are not limited to: roughness, contrast, directionality, hot spot temperature difference between the previous and next frames, device temperature mean, device temperature range, discharge photon number and spot area. Roughness, contrast and directionality are visible light spectrum characteristics, hot spot temperature difference between the previous and next frames, device temperature mean and device temperature range are infrared light spectrum characteristics, and discharge photon number and spot area are ultraviolet light spectrum characteristics.

[0059] Optionally, in another optional embodiment of the present invention, the above-mentioned abnormal feature extraction model includes: a visible light abnormal feature extraction model, an infrared light abnormal feature extraction model and an ultraviolet light abnormal feature extraction model. The present invention extracts the features reflecting the abnormal conditions of the target device in the three-light fusion image sequence data by establishing an abnormal feature extraction model, so that compared with the prior art, the present invention does not require a manual judgment method for the image data, and improves the detection efficiency while ensuring the accuracy of the abnormal feature extraction of the target device.

[0060] It is understandable that the visible light feature extraction model can be extracted based on the texture feature algorithm proposed by Tamura; the infrared light abnormal feature extraction model can first segment the target device and the background of the infrared image based on the maximum inter-class variance method, and then calculate the feature value reflecting the abnormal condition of the target device based on the maximum value, mean and range. The abnormal feature extraction model can extract features based on binarization, morphological filtering, removal of the minimum area and accumulation of time series data.

[0061] S103, dividing part or all of the multivariate time series data into subsequences according to a preset window and a preset step size to obtain multivariate subsequence segment data.

[0062] Optionally, in an optional embodiment of the present invention, before performing the above-mentioned sub-sequence division, the multivariate time series sequence data may be preprocessed to remove erroneous and missing data points and non-existent invalid feature dimensions in the multivariate time series sequence data, thereby reducing the amount of data that needs to be sub-sequenced and improving the overall detection efficiency while ensuring data accuracy.

[0063] Among them, the above-mentioned preprocessing methods include but are not limited to: missing value processing, data standardization and feature selection, etc. It is understandable that the above-mentioned preprocessing can be implemented by a preprocessing algorithm, and the types of preprocessing algorithms include but are not limited to: singular value decomposition algorithm (Singular Value Decomposition, SVD), principal component analysis algorithm (Principal Component Analysis, PCA) and linear discriminant analysis algorithm (Linear Discriminant Analysis, LDA).

[0064] Optionally, in another optional embodiment of the present invention, the above-mentioned preprocessed multivariate time series data is divided into subsequences according to preset windows and preset step sizes, in order to eliminate the influence of noise caused by the preprocessing operation on the final detection results, and the multivariate subsequence segment data obtained after subsequence division is divided into multiple time periods according to chronological order, which is conducive to capturing the time correlation of the time series distribution, making the detection results more accurate. At the same time, the multivariate time series data located in a specific window and step size can accurately reflect the abnormal condition of the target device. Among them, the above-mentioned preset window and preset step size can be obtained based on historical operation and inspection data and target device performance parameters, and the present invention does not impose too many restrictions on this.

[0065] S104: Input the multivariable subsequence segment data into a preset anomaly detection model, so that the preset anomaly detection model performs error judgment on the multivariable subsequence segment data to obtain an abnormal state judgment result of the target device.

[0066] Optionally, the present invention utilizes a preset anomaly detection model to perform result determination on the abnormal state of the target device, so that the detection efficiency of the present invention is improved compared to the prior art, which manually determines the abnormal state of the target device.

[0067] The present invention performs subsequence division on the multivariate time series sequence data after abnormal feature extraction according to a preset window and a preset step size, and obtains multivariate subsequence segment data that can reflect the abnormal state of the target device on the basis of ensuring data accuracy, so that the present invention improves both detection accuracy and detection efficiency compared to the prior art method that requires manual identification of abnormal states in images. At the same time, the present invention uses a preset abnormality detection model to perform error judgment on the multivariate subsequence segment data, so that the present invention further improves the detection efficiency compared to the manual judgment method in the prior art. Finally, the present invention uses a three-light acquisition device to obtain a three-light fusion image sequence data, replacing the prior art method that requires operation and maintenance personnel to carry three types of light wave observation equipment for image acquisition, so that the present invention simplifies the detection data acquisition process and improves the detection efficiency compared to the prior art. It can be seen that the present invention improves the efficiency of abnormal detection of power equipment.

[0068] Optionally, the preset anomaly detection model is composed of a preset generation model, a preset discrimination model and a result fusion model, and the multivariate subsequence segment data is input into the preset anomaly detection model so that the preset anomaly detection model performs error judgment on the multivariate subsequence segment data to obtain an abnormal state judgment result of the target device, including:

[0069] The multivariate subsequence segment data is input into a preset generation model so that the preset generation model generates simulated samples according to the multivariate subsequence segment data, and calculates the reconstruction error between the simulated samples and the real samples of the normal operation of the target device.

[0070] The multivariate subsequence segment data is input into a preset discriminant model so that the preset discriminant model generates a discriminant error according to the multivariate subsequence segment data, wherein the discriminant error refers to whether the target device is normal or the target device is abnormal.

[0071] The reconstruction error and the discrimination error are input into the result fusion model for weighted result fusion to obtain a discrimination result, wherein the discrimination result is that the target device operates normally or the target device operates abnormally.

[0072] Those skilled in the art will appreciate that, in other optional embodiments of the present invention, the preset generation model and the preset discrimination model are trained using data from the normal operation of the target device as a training set, and the preset result fusion model is trained using data from the normal operation and abnormal operation of the target device as a training set.

[0073] Optionally, in other optional embodiments of the present invention, the basic network structure used by the above-mentioned preset generation model includes but is not limited to: Generative Adversarial Networks (GAN), Variational Auto-Encoder (VAE) and Hidden Markov Chain, etc. The basic network structure used by the above-mentioned preset discrimination model includes but is not limited to: Linear Regression Network (LR), Logistic Regression Network (LR), Support Vector Machine (SVM), etc. The basic network structure of the above-mentioned preset result fusion model includes but is not limited to: selection voting and weighted averaging, etc.

[0074] Optionally, the reconstruction error reflects the degree of difference between the estimator and the estimated quantity. In an optional embodiment of the present invention, the reconstruction error reflects the degree of difference between the simulated sample and the real sample of the normal operation of the target device. The reconstruction error can be obtained using the mean square error (MSE).

[0075] Optionally, in an optional embodiment of the present invention, the preset discrimination model mentioned above can be directly used as the detection result of the abnormal state of the target device after being trained.

[0076] Optionally, the above method further includes:

[0077] According to the characteristics of the time series data stored in the preset database, the preset window and step size corresponding to part of the data or all of the data in the multivariate time series data are determined.

[0078] Optionally, in an optional embodiment of the present invention, the sources of the preset window and step size include but are not limited to: target device operation cycle, target device historical operation and inspection data, target device performance parameters and frame rate of collected data. The preset window and step size can be fine-tuned based on experimental results. Optionally, the training process of the preset generation model includes:

[0079] The data of normal operation of the target device is used as input to train the generation model to obtain a preset generation model, wherein the output of the preset generation model is the reconstruction error between the simulated sample and the real sample of the normal operation of the target device.

[0080] Optionally, the training process of the preset discriminant model includes:

[0081] The discriminant model is trained using the data of the normal operation of the target device and the data of the abnormal operation of the target device to obtain a preset discriminant model, wherein the output of the preset discriminant model is that the target device is normal or the target device is abnormal.

[0082] Optionally, part or all of the data in the multivariate time series data is divided into subsequences according to a preset window and a preset step size to obtain multivariate subsequence segment data, including:

[0083] Using a preset missing value processing algorithm and a data normalization algorithm, the multivariate time series data is subjected to an operation of removing missing data points and erroneous data points to obtain first multivariate time series data;

[0084] A feature selection algorithm is used to perform dimensionality reduction and interference dimension removal processing on the first multivariate time series data to obtain second multivariate time series data, wherein the number of data in the second multivariate time series data is not greater than the number of data in the multivariate time series data.

[0085] The present invention performs subsequence division on the multivariate time series sequence data after abnormal feature extraction according to a preset window and a preset step size, and obtains multivariate subsequence segment data that can reflect the abnormal state of the target device on the basis of ensuring data accuracy, so that the present invention improves both detection accuracy and detection efficiency compared to the prior art method that requires manual identification of abnormal states in images. At the same time, the present invention uses a preset abnormality detection model to perform error judgment on the multivariate subsequence segment data, so that the present invention further improves the detection efficiency compared to the manual judgment method in the prior art. Finally, the present invention uses a three-light acquisition device to obtain a three-light fusion image sequence data, replacing the prior art method that requires operation and maintenance personnel to carry three types of light wave observation equipment for image acquisition, so that the present invention simplifies the detection data acquisition process and improves the detection efficiency compared to the prior art. It can be seen that the present invention improves the efficiency of abnormal detection of power equipment.

[0086] Corresponding to the above-mentioned method embodiment for detecting abnormality of equipment, the present invention also provides a system for detecting abnormality of equipment, such as Figure 2 As shown, the system includes:

[0087] The data acquisition module 201 is used to use a three-light acquisition device to collect data from a target device within a preset time period to obtain three-light fusion image sequence data, wherein the three-light acquisition device is a device that can simultaneously collect image data of three light spectrum segments of the target device, and each three-light fusion image in the three-light fusion image sequence data carries image features of three light spectrum segments, and the three light spectrum segments include an infrared light spectrum segment, an ultraviolet light spectrum segment, and a visible light spectrum segment;

[0088] The feature extraction module 202 is used to input the three-light fusion image sequence data into the abnormal feature extraction model to extract features and obtain multivariate time series data;

[0089] The subsequence division module 203 is used to divide part or all of the data in the multivariate time series data into subsequences according to a preset window and a preset step size to obtain multivariate subsequence segment data;

[0090] The result output module 204 is used to input the multivariate subsequence segment data into a preset anomaly detection model, so that the preset anomaly detection model performs error determination on the multivariate subsequence segment data to obtain an abnormal state determination result of the target device.

[0091] Optionally, the result output module 204 includes:

[0092] The first error generation submodule is used to input the multivariate subsequence segment data into the preset generation model so that the preset generation model generates simulation samples according to the multivariate subsequence segment data and calculates the reconstruction error between the simulation samples and the real samples of the normal operation of the target device.

[0093] The second error generation submodule is used to input the multivariate subsequence segment data into a preset discriminant model so that the preset discriminant model generates a discriminant error according to the multivariate subsequence segment data, wherein the discriminant error refers to whether the target device is normal or abnormal.

[0094] The result fusion submodule is used to input the reconstruction error and the discrimination error into the result fusion model for weighted result fusion to obtain a judgment result, wherein the judgment result is that the target device operates normally or the target device operates abnormally.

[0095] Optionally, the above system further includes:

[0096] The data calling module is used to determine the preset window and step size corresponding to part or all of the data in the multivariate time series data according to the time series data characteristics stored in the preset database.

[0097] Optionally, the first error generation submodule includes: a first training submodule,

[0098] The first training submodule uses the data of normal operation of the target device as input to train the generation model to obtain a preset generation model, wherein the output of the preset generation model is the reconstruction error between the simulated sample and the real sample of the normal operation of the target device.

[0099] Optionally, the second error generation submodule includes: a second training submodule,

[0100] The second training submodule trains the discrimination model using the data of normal operation of the target device and the data of abnormal operation of the target device to obtain a preset discrimination model, wherein the output of the preset discrimination model is that the target device is normal or the target device is abnormal.

[0101] Optionally, the subsequence division module 203 includes:

[0102] The first preprocessing submodule is used to use a preset missing value processing algorithm and a data standardization algorithm to remove missing data points and erroneous data points from the multivariate time series data to obtain first multivariate time series data.

[0103] The second preprocessing submodule is used to use a feature selection algorithm to perform dimensionality reduction and interference dimension removal processing on the first multivariate time series data to obtain second multivariate time series data, wherein the number of data in the second multivariate time series data is not greater than the number of data in the multivariate time series data.

[0104] The embodiment of the present invention provides a device for detecting device abnormality, such as Figure 3 As shown, including:

[0105] Processor 301;

[0106] The memory 302 is used to store instructions executable by the processor 301 .

[0107] The processor 301 is configured to execute instructions to implement any of the above-mentioned methods for detecting device abnormalities.

[0108] An embodiment of the present invention provides a computer-readable storage medium. When instructions in the computer-readable storage medium are executed by a processor of a device, a device for device abnormality detection can execute any of the above-mentioned methods for device abnormality detection.

[0109] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0110] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0111] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0112] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. It should also be noted that the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0113] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0114] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for detecting device anomaly, characterized in that: The method comprises: Within a preset time period, a three-light acquisition device is used to acquire data from a target device to obtain three-light fusion image sequence data, wherein the three-light acquisition device is a device that can simultaneously acquire image data of three light spectrum bands of the target device, and each three-light fusion image in the three-light fusion image sequence data carries image features of three light spectrum bands, and the three light spectrum bands include an infrared light spectrum band, an ultraviolet light spectrum band, and a visible light spectrum band; Inputting the three-light fusion image sequence data into an abnormal feature extraction model for feature extraction to obtain multivariate time series data; Based on a preset window and a preset step size, the preset missing value processing algorithm and the data normalization algorithm are used to remove missing data points and erroneous data points from the multivariate time series sequence data to obtain first multivariate time series sequence data; the first multivariate time series sequence data is subjected to dimensionality reduction processing and interference dimension removal processing using a feature selection algorithm to obtain second multivariate time series sequence data, and the second multivariate time series sequence data is determined as multivariate subsequence segment data, wherein the number of data in the second multivariate time series sequence data is not greater than the number of data in the multivariate time series sequence data; the multivariate subsequence segment data is input into a preset anomaly detection model so that the preset anomaly detection model performs error judgment on the multivariate subsequence segment data to obtain an abnormal state judgment result of the target device.

2. The method according to claim 1, characterized in that The preset anomaly detection model is composed of a preset generation model, a preset discrimination model and a result fusion model. The multivariate subsequence segment data is input into the preset anomaly detection model so that the preset anomaly detection model performs error judgment on the multivariate subsequence segment data to obtain the abnormal state judgment result of the target device, including: Inputting the multivariate subsequence segment data into the preset generation model, so that the preset generation model generates simulated samples according to the multivariate subsequence segment data, and calculates the reconstruction error between the simulated samples and the real samples of the normal operation of the target device; Inputting the multivariate subsequence segment data into the preset discriminant model so that the preset discriminant model generates a discriminant error according to the multivariate subsequence segment data, wherein the discriminant error refers to whether the target device is normal or the target device is abnormal; The reconstruction error and the discrimination error are input into the result fusion model for weighted result fusion to obtain a judgment result, wherein the judgment result is that the target device operates normally or the target device operates abnormally.

3. The method according to claim 1, characterized in that: The method further comprises: According to the characteristics of the time series data stored in the preset database, the preset window and the preset step size corresponding to part of the data or all of the data in the multivariate time series data are determined.

4. The method according to claim 2, characterized in that: The training process of the preset generation model includes: The generation model is trained using the data of the normal operation of the target device as input to obtain the preset generation model, wherein the output of the preset generation model is the reconstruction error between the simulated sample and the real sample of the normal operation of the target device.

5. The method according to claim 2, characterized in that: The training process of the preset discriminant model includes: The discriminant model is trained using the data of the normal operation of the target device and the data of the abnormal operation of the target device to obtain the preset discriminant model, wherein the output of the preset discriminant model is that the target device is normal or the target device is abnormal.

6. A system for detecting equipment anomalies, characterized in that: The system comprises: A data acquisition module, used to use a three-light acquisition device to collect data from a target device within a preset time period to obtain three-light fusion image sequence data, wherein the three-light acquisition device is a device that can simultaneously collect image data of three light spectrum segments of the target device, and each three-light fusion image in the three-light fusion image sequence data carries image features of three light spectrum segments, and the three light spectrum segments include an infrared light spectrum segment, an ultraviolet light spectrum segment, and a visible light spectrum segment; A feature extraction module is used to input the three-light fusion image sequence data into an abnormal feature extraction model for feature extraction to obtain multivariate time series data; A subsequence division module is used to remove missing data points and erroneous data points from the multivariate time series data based on a preset window and a preset step size, using a preset missing value processing algorithm and a data normalization algorithm, to obtain first multivariate time series data; perform dimensionality reduction processing and interference dimension removal processing on the first multivariate time series data using a feature selection algorithm to obtain second multivariate time series data, and determine the second multivariate time series data as multivariate subsequence segment data, wherein the number of data in the second multivariate time series data is not greater than the number of data in the multivariate time series data; The result output module is used to input the multivariable subsequence segment data into a preset anomaly detection model so that the preset anomaly detection model performs error judgment on the multivariable subsequence segment data to obtain an abnormal state judgment result of the target device.

7. The system according to claim 6, characterized in that The system further comprises: The data calling module is used to determine the preset window and the preset step size corresponding to part of or all of the data in the multivariate time series data according to the time series data characteristics stored in the preset database.

8. A device for detecting device anomalies, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the method for device abnormality detection according to any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of a device for device abnormality detection, the device for device abnormality detection is enabled to perform the method for device abnormality detection as described in any one of claims 1 to 5.

Images