Power plant safety event identification method and device based on large model

Abstract

The invention belongs to the technical field of power plant safety supervision, and discloses a power plant safety event identification method and device based on a large model. The method comprises the following steps: acquiring a monitoring video image of a power plant through monitoring equipment; performing dimension reduction processing on the monitoring video image to obtain a processed monitoring video image; extracting local features from the processed monitoring video image by using a large security event recognition model, and constructing an abnormal score map corresponding to the monitoring video image according to the local features, the large security event recognition model being an unsupervised deep learning model; and identifying the safety event of the power plant based on the abnormal score graph. Through the above mode, the security event is identified in real time by using the security event identification large model, the identification accuracy is improved, and the safe operation of the power plant can be effectively ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of power plant safety supervision, and in particular to a method and device for identifying power plant safety events based on a large model. Background Art

[0002] As a power supply unit, power plants play an important role in social production and life. The safe operation and safety detection of power plants are also of great significance.

[0003] The safe operation of power plants mainly involves detecting a series of safety-related events in power plants, and ensuring the safe operation of power plants by real-time monitoring of safety events. At present, the identification of safety events in power plants mainly uses manually defined features to make manual judgments on safety events appearing in monitoring videos. However, this method is limited by human experience, which affects the accuracy of identification and cannot effectively ensure the safe operation of power plants.

[0004] The above contents are only used to assist in understanding the technical solution of the present invention and do not constitute an admission that the above contents are prior art. Summary of the invention

[0005] The main purpose of the present invention is to provide a method and device for identifying power plant safety events based on a large model, aiming to solve the technical problem that the current method is limited by human experience, affects the accuracy of identification, and cannot effectively ensure the safe operation of the power plant.

[0006] To achieve the above object, the present invention provides a method for identifying power plant safety events based on a large model, and the method for identifying power plant safety events based on a large model comprises the following steps:

[0007] Acquire monitoring video images of the power plant through monitoring equipment;

[0008] Performing dimensionality reduction processing on the surveillance video image to obtain a processed surveillance video image;

[0009] Extracting local features from the processed surveillance video image using a security event recognition model, and constructing an abnormality score map corresponding to the surveillance video image based on the local features, wherein the security event recognition model is an unsupervised deep learning model;

[0010] Safety events of the power plant are identified based on the abnormal score graph.

[0011] In some embodiments, the performing dimensionality reduction processing on the surveillance video image to obtain a processed surveillance video image includes:

[0012] Determine that the surveillance video image has N images, the size of the image is W×h, and connect each image by column to form an m=W×h-dimensional column vector x i ;

[0013] The column vector x of N images i Connect together to form a matrix X, and calculate the covariance matrix based on the matrix X;

[0014] Constructing a mapping matrix U based on the covariance matrix, wherein the mapping matrix U is constructed by eigenvectors corresponding to r eigenvalues;

[0015] The column vector of the monitoring video image is reduced from m=W×h dimension to r dimension through the mapping matrix U to obtain a processed monitoring video image.

[0016] In some embodiments, the extracting local features from the processed surveillance video image using the security event recognition large model includes:

[0017] Dividing the processed surveillance video image into a plurality of sub-surveillance video images;

[0018] For any sub-monitoring video image, a corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed;

[0019] Subtract the mean of the image block set from any image block to obtain the corresponding first matrix;

[0020] Connecting the first matrices corresponding to the sub-monitoring video images to obtain a first target matrix;

[0021] Constructing a filter kernel according to the first target matrix;

[0022] The filter kernel is used to extract local features from the processed surveillance video image.

[0023] In some embodiments, constructing a filter kernel according to the first target matrix includes:

[0024] Constructing a primary filter kernel according to the target matrix;

[0025] Obtaining a mapping matrix corresponding to any sub-monitoring video image according to the first-level filter kernel;

[0026] Taking out the corresponding image block from the mapping matrix, and subtracting the mean value of the mapping matrix from any image block to obtain the corresponding second matrix;

[0027] Connecting the second matrices corresponding to the sub-monitoring video images to obtain a second target matrix;

[0028] Constructing a secondary filter kernel according to the second target matrix;

[0029] Accordingly, the extracting local features from the processed surveillance video image using the filter kernel includes:

[0030] The secondary filter kernel is used to extract local features from the processed surveillance video image.

[0031] In some embodiments, constructing an abnormality score map corresponding to the surveillance video image according to the local features includes:

[0032] Clustering the local features using a K-mediods clustering algorithm to obtain a number of cluster centers;

[0033] Calculating anomaly scores based on the plurality of cluster centers;

[0034] The abnormal score is mapped back to the corresponding position in the monitoring video image to obtain an abnormal score map corresponding to the monitoring video image.

[0035] In some embodiments, the identifying the safety event of the power plant based on the abnormal score graph includes:

[0036] Determine the abnormal values ​​of the safety events corresponding to each area of ​​the power plant based on the abnormal score map;

[0037] Compare the abnormal value with a preset threshold to determine the abnormal area and abnormal event corresponding to the abnormal value exceeding the preset threshold;

[0038] Determine the number of consecutive frames of surveillance video images corresponding to which the abnormal score exceeds a preset threshold;

[0039] A duration of the abnormality is determined based on the number of images.

[0040] In some embodiments, the method further comprises:

[0041] After identifying the abnormal area, abnormal event, and abnormal duration based on the abnormal score map, determining the object that triggered the abnormal event;

[0042] Performing facial recognition on the subject to identify the offender;

[0043] Send an alarm to the violator to stop the behavior that triggers the abnormal event.

[0044] In addition, to achieve the above-mentioned purpose, the present invention also proposes a power plant safety event identification device based on a large model, and the power plant safety event identification device based on a large model includes:

[0045] An acquisition module, used to acquire monitoring video images of the power plant through monitoring equipment;

[0046] A processing module, used for performing dimensionality reduction processing on the surveillance video image to obtain a processed surveillance video image;

[0047] An extraction module, used to extract local features from the processed surveillance video image using a security event recognition model, and construct an abnormal score map corresponding to the surveillance video image according to the local features, wherein the security event recognition model is an unsupervised deep learning model;

[0048] An identification module is used to identify safety events of the power plant based on the abnormal score graph.

[0049] In some embodiments, the processing module is used to divide the processed surveillance video image into a plurality of sub-surveillance video images;

[0050] For any sub-monitoring video image, a corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed;

[0051] Subtract the mean of the image block set from any image block to obtain the corresponding first matrix;

[0052] Connecting the first matrices corresponding to the sub-monitoring video images to obtain a first target matrix;

[0053] Constructing a filter kernel according to the first target matrix;

[0054] The filter kernel is used to extract local features from the processed surveillance video image.

[0055] In some embodiments, the extraction module is used to divide the processed surveillance video image into a number of sub-surveillance video images;

[0056] For any sub-monitoring video image, a corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed;

[0057] Subtract the mean of the image block set from any image block to obtain the corresponding first matrix;

[0058] Connecting the first matrices corresponding to the sub-monitoring video images to obtain a first target matrix;

[0059] Constructing a filter kernel according to the first target matrix;

[0060] The filter kernel is used to extract local features from the processed surveillance video image.

[0061] The present invention obtains the monitoring video image of the power plant through the monitoring equipment; performs dimensionality reduction processing on the monitoring video image to obtain the processed monitoring video image; uses the security event recognition big model to extract local features from the processed monitoring video image, and constructs the abnormal score map corresponding to the monitoring video image according to the local features, and the security event recognition big model is an unsupervised deep learning model; and identifies the security events of the power plant based on the abnormal score map. In the above manner, the security event recognition big model is used to perform real-time identification of security events, which improves the identification accuracy and can effectively ensure the safe operation of the power plant. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a flow chart of the first embodiment of the method for identifying power plant safety events based on a large model according to the present invention;

[0063] Figure 2 It is a structural block diagram of the first embodiment of the power plant safety event identification device based on a large model of the present invention.

[0064] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0065] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0066] The embodiment of the present invention provides a method for identifying power plant safety events based on a large model, referring to Figure 1 , Figure 1 The figure is a flow chart of a first embodiment of a method for identifying power plant safety events based on a large model according to the present invention.

[0067] In this embodiment, the power plant safety event identification method based on the large model includes the following steps:

[0068] Step S10: Acquire monitoring video images of the power plant through monitoring equipment.

[0069] In this embodiment, the executor of this embodiment is a power plant safety event identification device based on a large model, wherein the power plant safety event identification device based on a large model has functions such as data processing, data communication and program running. The power plant safety event identification device based on a large model can be a computer terminal device or other network device, and of course it can also be other devices with similar functions, and this embodiment does not limit this.

[0070] It should be noted that the safe operation of power plants mainly involves the detection of a series of safety-related events in power plants, and real-time monitoring of safety events to ensure the safe operation of power plants. At present, the identification of power plant safety events mainly uses manually defined features to make manual judgments on safety events appearing in monitoring videos. However, this method is limited by human experience, which affects the accuracy of identification and cannot effectively ensure the safe operation of power plants.

[0071] In order to solve the above technical problems, this embodiment obtains the monitoring video image of the power plant through the monitoring equipment; performs dimensionality reduction processing on the monitoring video image to obtain the processed monitoring video image; uses the security event recognition large model to extract local features from the processed monitoring video image, and constructs the abnormal score map corresponding to the monitoring video image according to the local features, and the security event recognition large model is an unsupervised deep learning model; based on the abnormal score map, the security events of the power plant are identified. Through the above method, the security event recognition large model is used to perform real-time identification of security events, which improves the identification accuracy and can effectively ensure the safe operation of the power plant. Specifically, it can be implemented in the following way.

[0072] In a specific implementation, in this embodiment, image recognition is used to identify safety events in a power plant, and monitoring video images can be collected in real time by various monitoring devices installed in the power plant.

[0073] Step S20: performing dimensionality reduction processing on the surveillance video image to obtain a processed surveillance video image.

[0074] After obtaining the surveillance video image, in order to improve the accuracy of subsequent recognition, in this embodiment, it is necessary to remove noise from the surveillance video image and retain the most important information, which can be achieved through dimensionality reduction processing. The specific process of dimensionality reduction processing is to determine that the surveillance video image has N images, the size of the image is W×h, and each image is connected by column to form an m=W×h-dimensional column vector x i ; The column vector x of N images i The matrix X is connected together to form a matrix X, and the covariance matrix is ​​calculated based on the matrix X; a mapping matrix U is constructed based on the covariance matrix, and the mapping matrix U is constructed by the eigenvectors corresponding to the r eigenvalues; the column vector of the monitoring video image is reduced from m=W×h dimension to r dimension through the mapping matrix U to obtain the processed monitoring video image.

[0075] It should be noted that, assuming that the N images are {I1,I2,...,I N}, the matrix X formed by connecting the column vectors is [x1,x2,...,x N ], the corresponding covariance matrix is ​​as follows:

[0076]

[0077] Where μ represents the mean vector.

[0078] By solving the eigenvalue and eigenvector of φ, we can get the eigenvalues ​​arranged from large to small {λ1,λ2,...,λ wh}, the corresponding eigenvector is {u1,u2,...,u wh}, and then select the eigenvectors corresponding to the first r eigenvalues ​​to form a mapping matrix U = [u1,u2,...,u r ], the number corresponding to r can be set according to actual needs, and this is not limited in this embodiment. After obtaining the mapping matrix U, the original W×h-dimensional image projection is linearly transformed to: i =U T x i , the original image is reduced to an r-dimensional column vector, removing the redundant information in the original image.

[0079] Step S30: extracting local features from the processed surveillance video image using the security event recognition large model, and constructing an abnormality score map corresponding to the surveillance video image based on the local features.

[0080] In the specific implementation, in this embodiment, a large security event recognition model is used to extract local features from the processed surveillance video image. The specific extraction process is to divide the processed surveillance video image into a number of sub-surveillance video images; for any sub-surveillance video image, the corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed; the mean of the image block set is subtracted from any image block to obtain a corresponding first matrix; the first matrices corresponding to the sub-surveillance video images are connected to obtain a first target matrix; a filter kernel is constructed according to the first target matrix; and the filter kernel is used to extract local features from the processed surveillance video image.

[0081] It should be noted that the size of the sub-monitoring video image is h×w, and the size of the image block taken out at each spatial point is k1×k2, so the constructed image block set is {x i,1 ,x i,2 ,...,x i,(h-k1+1)×(w-k2+1)}, where each item x i,j Represents x i The first matrix is ​​represented as The first target matrix obtained by concatenating all the first matrices is According to the first target matrix obtained above, a filter kernel can be further constructed, and local features can be automatically extracted using the filter kernel.

[0082] Furthermore, in this embodiment, the process of constructing a filter kernel according to the first target matrix is ​​specifically as follows: constructing a first-level filter kernel according to the first target matrix; obtaining a mapping matrix corresponding to any sub-monitoring video image according to the first-level filter kernel; taking out the corresponding image block from the mapping matrix, subtracting the mean of the mapping matrix from any image block to obtain the corresponding second matrix; connecting the second matrices corresponding to each sub-monitoring video image to obtain a second target matrix; and constructing a second-level filter kernel according to the second target matrix.

[0083] It should be noted that the formula of the first-level filter kernel is as follows:

[0084]

[0085] in, Representation Matrix The eigenvector corresponding to the lth eigenvalue is, It is expressed as converting a vector into a matrix. The mapping matrix is Then, in the same way as above, the size of the extracted image block is also k1×k2, and the second matrix constructed is The second target matrix obtained by concatenating all the second matrices is Y=[Y 1 ,Y 2 ,...,Y L1 ], where L1 represents the number of primary filter cores, and the formula for the secondary filter core is as follows:

[0086]

[0087] Among them, q l (YY T ) represents the matrix YY T The eigenvector corresponding to the lth eigenvalue.

[0088] After extracting the local features, in this embodiment, it is necessary to further construct an anomaly score map, which corresponds to the anomaly values ​​of each spatial position in the surveillance video image. The specific process of constructing the anomaly score map is to cluster the local features using the K-mediods clustering algorithm to obtain a number of cluster centers; calculate the anomaly scores based on the several cluster centers; map the anomaly scores back to the corresponding positions in the surveillance video image to obtain the anomaly score map corresponding to the surveillance video image.

[0089] It should be noted that the K-mediods clustering algorithm is used to cluster local features into k categories, and k cluster centers C = {c1, c2, ..., c k}, the calculation formula of the outlier value is as follows:

[0090] g(x)=min(xc k ) T (xc k ),

[0091] Among them, c k is any of the above cluster centers, and x is the extracted local feature. In order to eliminate local noise interference, the above-obtained outliers are smoothed to obtain the processed g'(x), and finally the classifier is used to obtain the final outlier score. The classifier is as follows:

[0092]

[0093] Step S40: identifying safety events of the power plant based on the abnormal score graph.

[0094] The identification process is specifically to determine the abnormal values ​​of the safety events corresponding to each area of ​​the power plant based on the abnormal score map. Different areas and different safety events have corresponding preset thresholds. If the preset threshold is exceeded, the corresponding area in the abnormal score map can be determined as an abnormal area, and the corresponding safety event is an abnormal event.

[0095] After determining the above-mentioned abnormal areas and abnormal events, in this embodiment, it is necessary to further determine the duration of the abnormality. It can be understood that the monitoring video image is a number of continuous images. Therefore, it is possible to determine a number of continuous frames of monitoring video images corresponding to abnormal values ​​exceeding the threshold, and then count the number of images, that is, the number of frames. The time of each frame can be determined by setting the frame rate of the monitoring device, and the duration of the abnormality can be calculated by multiplying the time of each frame by the above-mentioned number of images.

[0096] Furthermore, after obtaining the above-mentioned recognition results, if there is an abnormal security event, the object that triggers the abnormal event is further determined, and face recognition is performed on the object to determine the violator; an alarm is sent to the violator to make the violator stop the behavior that triggers the abnormal event. Specifically, an alarm can be sent to the mobile terminal of the violator, and if the violator still has an abnormal security event within a preset time after receiving the alarm information, the violator is reminded by a buzzer prompt.

[0097] This embodiment obtains the monitoring video image of the power plant through the monitoring equipment; performs dimensionality reduction processing on the monitoring video image to obtain the processed monitoring video image; uses the security event recognition big model to extract local features from the processed monitoring video image, and constructs the abnormal score map corresponding to the monitoring video image according to the local features, and the security event recognition big model is an unsupervised deep learning model; and identifies the security events of the power plant based on the abnormal score map. In the above manner, the security event recognition big model is used to perform real-time identification of security events, which improves the identification accuracy and can effectively ensure the safe operation of the power plant.

[0098] Reference Figure 2 , Figure 2 It is a structural block diagram of the first embodiment of the power plant safety event identification device based on a large model of the present invention.

[0099] like Figure 2 As shown, the power plant safety event identification device based on a large model proposed in an embodiment of the present invention includes:

[0100] An acquisition module 10 is used to acquire monitoring video images of the power plant through monitoring equipment;

[0101] The processing module 20 is used to perform dimensionality reduction processing on the surveillance video image to obtain a processed surveillance video image;

[0102] An extraction module 30 is used to extract local features from the processed surveillance video image using a security event recognition model, and to construct an abnormal score map corresponding to the surveillance video image according to the local features, wherein the security event recognition model is an unsupervised deep learning model;

[0103] The identification module 40 is used to identify the safety events of the power plant based on the abnormal score graph.

[0104] This embodiment obtains the monitoring video image of the power plant through the monitoring equipment; performs dimensionality reduction processing on the monitoring video image to obtain the processed monitoring video image; uses the security event recognition big model to extract local features from the processed monitoring video image, and constructs the abnormal score map corresponding to the monitoring video image according to the local features, and the security event recognition big model is an unsupervised deep learning model; and identifies the security events of the power plant based on the abnormal score map. In the above manner, the security event recognition big model is used to perform real-time identification of security events, which improves the identification accuracy and can effectively ensure the safe operation of the power plant.

[0105] In some embodiments, the processing module 20 is used to divide the processed surveillance video image into a plurality of sub-surveillance video images;

[0106] For any sub-monitoring video image, a corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed;

[0107] Subtract the mean of the image block set from any image block to obtain the corresponding first matrix;

[0108] Connecting the first matrices corresponding to the sub-monitoring video images to obtain a first target matrix;

[0109] Constructing a filter kernel according to the first target matrix;

[0110] The filter kernel is used to extract local features from the processed surveillance video image.

[0111] In some embodiments, the extraction module 30 is used to divide the processed surveillance video image into a plurality of sub-surveillance video images;

[0112] For any sub-monitoring video image, a corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed;

[0113] Subtract the mean of the image block set from any image block to obtain the corresponding first matrix;

[0114] Connecting the first matrices corresponding to the sub-monitoring video images to obtain a first target matrix;

[0115] Constructing a filter kernel according to the first target matrix;

[0116] The filter kernel is used to extract local features from the processed surveillance video image.

[0117] In some embodiments, the extraction module 30 is used to construct a primary filter kernel according to the target matrix;

[0118] Obtaining a mapping matrix corresponding to any sub-monitoring video image according to the first-level filter kernel;

[0119] Taking out the corresponding image block from the mapping matrix, and subtracting the mean value of the mapping matrix from any image block to obtain the corresponding second matrix;

[0120] Connecting the second matrices corresponding to the sub-monitoring video images to obtain a second target matrix;

[0121] Constructing a secondary filter kernel according to the second target matrix;

[0122] Accordingly, the extracting local features from the processed surveillance video image using the filter kernel includes:

[0123] The secondary filter kernel is used to extract local features from the processed surveillance video image.

[0124] In some embodiments, the extraction module 30 is used to cluster the local features using a K-mediods clustering algorithm to obtain a number of cluster centers;

[0125] Calculating anomaly scores based on the plurality of cluster centers;

[0126] The abnormal score is mapped back to the corresponding position in the monitoring video image to obtain an abnormal score map corresponding to the monitoring video image.

[0127] In some embodiments, the identification module 40 is used to determine the abnormal values ​​of the security events corresponding to each area of ​​the power plant based on the abnormal score map;

[0128] Compare the abnormal value with a preset threshold to determine the abnormal area and abnormal event corresponding to the abnormal value exceeding the preset threshold;

[0129] Determine the number of consecutive frames of surveillance video images corresponding to which the abnormal score exceeds a preset threshold;

[0130] A duration of the abnormality is determined based on the number of images.

[0131] In some embodiments, the identification module 40 is used to determine the object that triggers the abnormal event after identifying the abnormal area, abnormal event and abnormal duration based on the abnormal score map;

[0132] Performing facial recognition on the subject to identify the offender;

[0133] Send an alarm to the violator to stop the behavior that triggers the abnormal event.

[0134] An embodiment of the present application also provides a power plant safety event identification device based on a large model, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus, and the memory is used to store a power plant safety event identification program based on a large model; the processor is used to implement the above-mentioned power plant safety event identification method based on the large model when executing the program stored in the memory.

[0135] The communication bus mentioned in the above-mentioned power plant safety event identification device based on a large model can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0136] The communication interface is used for communication between the above-mentioned large model-based power plant safety event identification device and other devices.

[0137] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0138] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, and discrete hardware components.

[0139] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integrated. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0140] 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. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article 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, article 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, article or device including the elements.

[0141] 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.

[0142] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

[0143] It should be understood that the above is only an example and does not constitute any limitation on the technical solution of the present invention. In specific applications, technicians in this field can make settings as needed, and the present invention does not limit this.

[0144] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of them according to actual needs to achieve the purpose of the present embodiment, and no limitation is made here.

[0145] In addition, for technical details that are not described in detail in this embodiment, please refer to the large model-based power plant safety event identification method provided in any embodiment of the present invention, and will not be repeated here.

[0146] In addition, it should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system 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, article or system. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or system including the element.

[0147] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0148] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as a read-only memory (ROM) / RAM, a magnetic disk, or an optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the present invention.

[0149] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

[0150] It is understandable that the system provided by the embodiment of the present invention corresponds to the method provided by the embodiment of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above method.

Claims

1. A method for identifying power plant safety events based on a large model, characterized in that: The large model-based power plant safety event identification method includes: Acquire monitoring video images of the power plant through monitoring equipment; Performing dimensionality reduction processing on the surveillance video image to obtain a processed surveillance video image; Extracting local features from the processed surveillance video image using a security event recognition model, and constructing an abnormality score map corresponding to the surveillance video image based on the local features, wherein the security event recognition model is an unsupervised deep learning model; Safety events of the power plant are identified based on the abnormal score graph.

2. The method for identifying power plant safety events based on a large model according to claim 1, characterized in that: The performing dimensionality reduction processing on the surveillance video image to obtain a processed surveillance video image includes: Determine that the surveillance video image has N images, the size of the image is W×h, and connect each image by column to form an m=W×h-dimensional column vector x i ; The column vector x of N images i Connect together to form a matrix X, and calculate the covariance matrix based on the matrix X; Constructing a mapping matrix U based on the covariance matrix, wherein the mapping matrix U is constructed by eigenvectors corresponding to r eigenvalues; The column vector of the monitoring video image is reduced from m=W×h dimension to r dimension through the mapping matrix U to obtain a processed monitoring video image.

3. The method for identifying power plant safety events based on a large model according to claim 1, characterized in that: The method of extracting local features from the processed surveillance video image using the security event recognition large model includes: Dividing the processed surveillance video image into a plurality of sub-surveillance video images; For any sub-monitoring video image, a corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed; Subtract the mean of the image block set from any image block to obtain the corresponding first matrix; Connecting the first matrices corresponding to the sub-monitoring video images to obtain a first target matrix; Constructing a filter kernel according to the first target matrix; The filter kernel is used to extract local features from the processed surveillance video image.

4. The method for identifying power plant safety events based on a large model according to claim 3, characterized in that: The step of constructing a filter kernel according to the first target matrix includes: Constructing a primary filter kernel according to the target matrix; Obtaining a mapping matrix corresponding to any sub-monitoring video image according to the first-level filter kernel; Taking out the corresponding image block from the mapping matrix, and subtracting the mean value of the mapping matrix from any image block to obtain the corresponding second matrix; Connecting the second matrices corresponding to the sub-monitoring video images to obtain a second target matrix; Constructing a secondary filter kernel according to the second target matrix; Accordingly, the extracting local features from the processed surveillance video image using the filter kernel includes: The secondary filter kernel is used to extract local features from the processed surveillance video image.

5. The method for identifying power plant safety events based on a large model according to claim 1, characterized in that: The constructing an abnormality score map corresponding to the surveillance video image according to the local features includes: Clustering the local features using a K-mediods clustering algorithm to obtain a number of cluster centers; Calculating anomaly scores based on the plurality of cluster centers; The abnormal score is mapped back to the corresponding position in the monitoring video image to obtain an abnormal score map corresponding to the monitoring video image.

6. The method for identifying power plant safety events based on a large model according to claim 1, characterized in that: The identifying the safety event of the power plant based on the abnormal score graph includes: Determine the abnormal values ​​of the safety events corresponding to each area of ​​the power plant based on the abnormal score map; Compare the abnormal value with a preset threshold to determine the abnormal area and abnormal event corresponding to the abnormal value exceeding the preset threshold; Determine the number of consecutive frames of surveillance video images corresponding to which the abnormal score exceeds a preset threshold; A duration of the abnormality is determined based on the number of images.

7. The method for identifying power plant safety events based on a large model according to any one of claims 1 to 6, characterized in that: The method further comprises: After identifying the abnormal area, abnormal event, and abnormal duration based on the abnormal score map, determining the object that triggered the abnormal event; Performing facial recognition on the subject to identify the offender; Send an alarm to the violator to stop the behavior that triggers the abnormal event.

8. A power plant safety event identification device based on a large model, characterized in that: The power plant safety event identification device based on the large model includes: An acquisition module, used to acquire monitoring video images of the power plant through monitoring equipment; A processing module, used for performing dimensionality reduction processing on the surveillance video image to obtain a processed surveillance video image; An extraction module, used to extract local features from the processed surveillance video image using a security event recognition model, and construct an abnormal score map corresponding to the surveillance video image according to the local features, wherein the security event recognition model is an unsupervised deep learning model; An identification module is used to identify safety events of the power plant based on the abnormal score graph.

9. The power plant safety event identification device based on a large model according to claim 8, characterized in that: The processing module is used to divide the processed surveillance video image into a plurality of sub-surveillance video images; For any sub-monitoring video image, a corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed; Subtract the mean of the image block set from any image block to obtain the corresponding first matrix; Connecting the first matrices corresponding to the sub-monitoring video images to obtain a first target matrix; Constructing a filter kernel according to the first target matrix; The filter kernel is used to extract local features from the processed surveillance video image.

10. The power plant safety event identification device based on a large model according to claim 8, characterized in that: The extraction module is used to divide the processed surveillance video image into a plurality of sub-surveillance video images; For any sub-monitoring video image, a corresponding image block is taken out at each spatial point, and a corresponding image block set is constructed; Subtract the mean of the image block set from any image block to obtain the corresponding first matrix; Connecting the first matrices corresponding to the sub-monitoring video images to obtain a first target matrix; Constructing a filter kernel according to the first target matrix; The filter kernel is used to extract local features from the processed surveillance video image.

Images