Abnormality detection method, device and equipment and storage medium

Abstract

The invention discloses an anomaly detection method, device and equipment and a storage medium, and relates to the technical field of intelligent operation and maintaining.The anomaly detection method comprises the steps that target heterogeneous data of a target machine room precise air conditioner are collected in real time through target heterogeneous equipment, and the target heterogeneous data comprise target sound data, target image data and target operation data; detecting the target sound data based on a preset unsupervised auto-encoder, detecting the target image data based on a preset deep learning visual model, and detecting the target operation data based on a preset static threshold algorithm and a preset state condition to obtain each detection result; and if any detection result is abnormal, whether the target machine room precision air conditioner is target fault equipment or not is detected on the basis of a preset multi-mode feature cross validation mechanism, the target sound data, the target image data and the target operation data. Therefore, the false alarm rate can be reduced, and the reliability of the detection result of the precise air conditioning equipment in the machine room is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent operation and maintenance, and particularly relates to an anomaly detection method, device, equipment and storage medium. Background Art

[0002] Currently, the operation and maintenance management of the precision air conditioning system in the computer room mainly relies on traditional manual inspections and single-dimensional threshold alarm mechanisms. Manual inspections adopt a periodic operation mode, with a time monitoring blind spot of up to several hours, making it difficult to capture sudden abnormal operating conditions of the equipment. The existing single-sensor monitoring system has weak anti-interference ability. Taking voiceprint detection as an example, the acoustic feature overlap between abnormal sounds of the air conditioner body and environmental noise reaches , resulting in a high false alarm rate being maintained for a long time. Operation and maintenance personnel need to spend a lot of time on alarm review, and currently, the temperature monitoring system based on threshold triggering can only alarm after obvious faults such as equipment overheating occur.

[0003] In summary, how to improve the reliability of the detection results of precision air conditioning equipment in the computer room is a technical problem that needs to be solved urgently at present. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an anomaly detection method, device, equipment and storage medium, which can improve the reliability of the detection results of precision air conditioning equipment in the computer room. The specific solutions are as follows:

[0005] In a first aspect, the present application provides an anomaly detection method, including:

[0006] Real-time collecting target heterogeneous data of a target computer room precision air conditioner through a target heterogeneous device, where the target heterogeneous data includes target sound data, target image data, and target operation data;

[0007] Detecting the target sound data based on a preset unsupervised autoencoder, detecting the target image data based on a preset deep learning vision model, and detecting the target operation data based on a preset static threshold algorithm and preset state conditions to obtain respective detection results;

[0008] If any of the detection results indicates an anomaly, then detect whether the target computer room precision air conditioner is a target faulty device based on a preset multimodal feature cross-validation mechanism, the target sound data, the target image data, and the target operation data.

[0009] Optionally, the real-time collecting of the target heterogeneous data of the target computer room precision air conditioner through the target heterogeneous device includes:

[0010] Determining a list of target sound collectors corresponding to the target computer room precision air conditioner, and real-time collecting the target sound data from each target sound collector in the list of target sound collectors according to a first preset time condition;

[0011] Use the target camera corresponding to the precision air conditioner in the target computer room to collect the target image data in real time, and collect the target operation data of the precision air conditioner in the target computer room in real time based on the target northbound interface of the preset data center infrastructure management system.

[0012] Optionally, after collecting the target heterogeneous data of the precision air conditioner in the target computer room through the target heterogeneous device, it further includes:

[0013] Upload the target sound data, the target image data, and the target operation data to a preset information storage list through a preset file upload interface;

[0014] Identify the target sound data based on a preset sound data identification method, identify the target image data based on a preset image data identification method, and identify the target operation data based on a preset operation data identification method;

[0015] Bind the target sound data, the target image data, and the target operation data to the device number corresponding to the precision air conditioner in the target computer room.

[0016] Optionally, the detecting the target sound data based on a preset unsupervised autoencoder includes:

[0017] Train an initial unsupervised autoencoder based on the historical sound data under preset normal operating conditions and preset control parameters to obtain a target unsupervised autoencoder; wherein, the preset control parameters include any one or a combination of several of batch size, number of iterations, and learning rate;

[0018] Determine the target Mel spectrogram corresponding to the target sound data, and input the target Mel spectrogram into the target unsupervised autoencoder to obtain the target eigenvalue corresponding to the target sound data, and determine the target root mean square error corresponding to the target sound data based on the target eigenvalue;

[0019] If the target root mean square error is greater than a preset root mean square error threshold, it is determined that the detection result corresponding to the target sound data indicates an abnormality.

[0020] Optionally, the detecting the target image data based on a preset deep learning vision model includes:

[0021] Train an initial deep learning vision model based on a preset transfer learning method and a preset image data set to obtain a target deep learning vision model, and extract the target smoke texture feature, the target flame color feature, and the target dynamic optical flow feature of the target image data based on the target deep learning vision model;

[0022] Using the target deep learning vision model, determine the target anomaly probability of the target image data based on the target smoke texture feature, the target flame color feature, and the target dynamic optical flow feature;

[0023] If the target anomaly probability is greater than a preset anomaly probability threshold, then detect the target image data based on a preset three-level verification mechanism.

[0024] Optionally, the detecting the target image data based on a preset three-level verification mechanism includes:

[0025] Determine the target local pixels of the target image data based on a preset local area condition, so as to judge whether there is a target pixel cluster in the target image data that satisfies a preset flame spectral feature condition based on the target local pixels;

[0026] Judge whether there is a target smoke area in the target image data that satisfies a preset edge blur feature condition based on a preset contour detection algorithm, and determine the historical image data corresponding to the target image data and the target illumination similarity of the target image data;

[0027] Determine the detection result corresponding to the target image data according to the target pixel cluster, the target smoke area, and the target illumination similarity.

[0028] Optionally, the detecting whether the target precision air conditioner in the target computer room is a target faulty device based on a preset multi-modal feature cross-verification mechanism, the target sound data, the target image data, and the target operation data includes:

[0029] Determine a target time period based on the current moment and a second preset time condition, and obtain the to-be-verified sound data, to-be-verified image data, and to-be-verified operation data within the target time period from the target heterogeneous data;

[0030] If it is determined that the target precision air conditioner in the target computer room is the target faulty device based on the to-be-verified sound data, the to-be-verified image data, the to-be-verified operation data, and the preset multi-modal feature cross-verification mechanism, then determine the abnormal data corresponding to the target precision air conditioner in the target computer room, and generate a corresponding target alarm message based on the abnormal data and the device number corresponding to the target precision air conditioner in the target computer room.

[0031] In a second aspect, the present application provides an anomaly detection device, including:

[0032] A target heterogeneous data acquisition module, configured to collect target heterogeneous data of a target precision air conditioner in a target computer room in real time through a target heterogeneous device, where the target heterogeneous data includes target sound data, target image data, and target operation data;

[0033] A detection result acquisition module, configured to detect the target sound data based on a preset unsupervised autoencoder, detect the target image data based on a preset deep learning vision model, and detect the target operation data based on a preset static threshold algorithm and a preset state condition, so as to obtain each detection result;

[0034] A target computer room precision air conditioner detection module, configured to, if any of the detection results indicates an abnormality, detect whether the target computer room precision air conditioner is a target faulty device based on a preset multi-modal feature cross-validation mechanism, the target sound data, the target image data, and the target operation data.

[0035] In a third aspect, the present application provides an electronic device, including:

[0036] A memory, configured to store a computer program;

[0037] A processor, configured to execute the computer program to implement the foregoing anomaly detection method.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium, configured to store a computer program; wherein, when the computer program is executed by a processor, the foregoing anomaly detection method is implemented.

[0039] In the present application, first, target heterogeneous data of a target computer room precision air conditioner is collected in real time by a target heterogeneous device, and the target heterogeneous data includes target sound data, target image data, and target operation data; then, the target sound data is detected based on a preset unsupervised autoencoder, the target image data is detected based on a preset deep learning vision model, and the target operation data is detected based on a preset static threshold algorithm and a preset state condition, so as to obtain each detection result; if any of the detection results indicates an abnormality, it is detected whether the target computer room precision air conditioner is a target faulty device based on a preset multi-modal feature cross-validation mechanism, the target sound data, the target image data, and the target operation data. As can be seen from the above, in the present application, the target sound data, the target image data, and the target operation data of the target computer room precision air conditioner are collected in real time, and then a preset unsupervised autoencoder is used to implement the detection of the target sound data, a preset deep learning vision model is used to detect the target image data, a preset static threshold algorithm and a preset state condition are used to detect the target operation data in real time, and finally, a multi-modal feature cross-validation mechanism is used to effectively distinguish equipment failures from environmental interferences. In this way, the present application can achieve all-weather automatic monitoring of the operating state of the computer room precision air conditioner, have the ability to respond to abnormal events within seconds, and can also greatly reduce the frequency of manual inspections. In this way, the present application can greatly reduce the false alarm rate and significantly improve the reliability of the detection results of the computer room precision air conditioner equipment. Description of the Drawings

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on the provided drawings.

[0041] Figure 1 It is a system architecture diagram of an anomaly detection solution provided by this application;

[0042] Figure 2 It is a flowchart of an anomaly detection method provided by this application;

[0043] Figure 3 It is a flowchart of a specific voice data detection provided by this application;

[0044] Figure 4 It is a flowchart of a specific image data detection provided by this application;

[0045] Figure 5 It is a schematic structural diagram of an anomaly detection device provided by this application;

[0046] Figure 6 It is a structural diagram of an electronic device provided by this application. Specific embodiments

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0048] The current operation and maintenance management of the precision air conditioning system in the computer room mainly relies on traditional manual inspections and one-dimensional threshold alarm mechanisms. Manual inspections adopt a periodic operation mode, with time monitoring blind spots of up to several hours, making it difficult to capture sudden abnormal conditions of equipment. The existing single-sensor monitoring system has weak anti-interference ability. Taking voiceprint detection as an example, the acoustic feature overlap between abnormal noises of the air conditioner body and environmental noises reaches , resulting in a long-term high false alarm rate. Maintenance personnel need to spend a lot of time on alarm review, and currently, the temperature monitoring system based on threshold triggering can only alarm after obvious faults such as equipment overheating occur. Therefore, this application provides an anomaly detection solution that can improve the reliability of the detection results of precision air conditioning equipment in the computer room.

[0049] In the anomaly detection solution of this application, the system framework adopted can be specifically referred to Figure 1As shown in the figure, it may specifically include: a data acquisition module, a multi-dimensional perception device management module, an information storage module, an intelligent analysis module, a message module, etc. Among them, the information storage module is mainly used to save various information of the system, including structured data such as device monitoring item logs, collected temperature information, and system configuration tables, as well as unstructured data such as sounds and pictures collected during the operation process. The main function of the data acquisition module is to manage each collector, realize the unified access of heterogeneous devices such as device operation sound acquisition, camera video / image acquisition, and DCIM (DataCenter Infrastructure Management) monitoring data acquisition, and the data acquisition module uses scheduled tasks to obtain data regularly. The intelligent analysis module includes an acoustic anomaly detection unit, a visual anomaly detection unit, a DCIM threshold alarm unit, a multi-modal cross-verification unit, etc., which are used to perform intelligent analysis on the collected data. The message module realizes the notification of abnormal event messages and pushes alarm information including device numbers, abnormal types, and abnormal data.

[0050] See Figure 2 As shown in the figure, an embodiment of the present invention discloses an anomaly detection method, which may include:

[0051] Step S11, real-time collect target heterogeneous data of a precision air conditioner in a target computer room through a target heterogeneous device, where the target heterogeneous data includes target sound data, target image data, and target operation data.

[0052] In this embodiment, multi-source heterogeneous data of the precision air conditioner in the computer room can be real-time collected through the matrix acoustic collector, visual monitoring device, and DCIM system of the data acquisition module. The above real-time collection of target heterogeneous data of the precision air conditioner in the target computer room through the target heterogeneous device may include: determining a list of target sound collectors corresponding to the precision air conditioner in the target computer room, and real-time collecting the target sound data from each target sound collector in the list of target sound collectors according to the first preset time condition; using the target camera corresponding to the precision air conditioner in the target computer room to real-time collect the target image data, and real-time collecting the target operation data of the precision air conditioner in the target computer room based on the target northbound interface of the preset data center infrastructure management system. It can be understood that after the above real-time collection of target heterogeneous data of the precision air conditioner in the target computer room through the target heterogeneous device, it may further include: uploading the target sound data, the target image data, and the target operation data to a preset information storage list through a preset file upload interface; identifying the target sound data based on a preset sound data identification method, identifying the target image data based on a preset image data identification method, and identifying the target operation data based on a preset operation data identification method; binding the target sound data, the target image data, and the target operation data to the device number corresponding to the precision air conditioner in the target computer room.

[0053] In the first specific implementation manner, in order to collect target sound data, an omnidirectional sound collection can be achieved by installing a matrix-type sound collector around the precision air conditioner in the target computer room. The sound collection process is as follows:

[0054] (1) Obtain the list of currently available target sound collectors from the multi-dimensional perception device management module;

[0055] (2) Traverse the list of target sound collectors, and for each sound collector, intercept 10s of sound data;

[0056] (3) If the acquisition of the target sound collector list fails, call the exception interface of the message module to send a data acquisition failure notification to the patrol personnel;

[0057] (4) If the acquisition is successful, call the file upload interface of the information storage module to upload the data, mark the data as sound data (voice), and bind the device number corresponding to the precision air conditioner in the target computer room;

[0058] (5) Complete the sound collection for the precision air conditioner equipment in the current target computer room, and start the sound collection for the precision air conditioner equipment in the next computer room.

[0059] In the second specific implementation manner, in order to collect target image data, a camera can be arranged above the precision air conditioner equipment in the target computer room to obtain a screenshot of the current camera, so as to obtain the target image data. Then call the file upload interface of the information storage module to upload the data, mark the data as image data (image) and bind the device number corresponding to the precision air conditioner in the target computer room.

[0060] In the third specific implementation manner, in order to collect target operation data, the northbound interface of the third-party DCIM system can be docked, and the DCIM monitoring data of the precision air conditioner equipment in the current target computer room, that is, the target operation data, can be obtained through the target northbound interface, including but not limited to return air temperature, filter clogging status, power supply status, internal fan status, etc. After the data collection is completed, the file upload interface of the information storage module can be called to save the collected target operation data, mark the data as DCIM data and bind the device number corresponding to the precision air conditioner in the target computer room.

[0061] Step S12: Detect the target sound data based on a preset unsupervised autoencoder, detect the target image data based on a preset deep learning vision model, and detect the target operation data based on a preset static threshold algorithm and preset state conditions, so as to obtain each detection result.

[0062] In this embodiment, a sound detection model can be used to detect target sound data. Generally, positive and negative samples are required for model training. To train the sound detection model, artificial air conditioner abnormal noises need to be created as negative samples. However, there are various faults in the precision air conditioners in the computer room, and the abnormal sounds emitted are also different. For sounds not included in the dataset, the classification accuracy of the classification algorithm is usually not high. Refer to Figure 3 As shown, to improve the accuracy of sound detection, the above method for detecting the target sound data based on a preset unsupervised autoencoder may include: training an initial unsupervised autoencoder based on historical sound data under preset normal operating conditions and preset control parameters to obtain a target unsupervised autoencoder; wherein, the preset control parameters include any one or a combination of several of batch size, number of iterations, and learning rate; determining a target Mel spectrogram corresponding to the target sound data, and inputting the target Mel spectrogram into the target unsupervised autoencoder to obtain a target eigenvalue corresponding to the target sound data, and determining a target root mean square error corresponding to the target sound data based on the target eigenvalue; if the target root mean square error is greater than a preset root mean square error threshold, it is determined that the detection result corresponding to the target sound data indicates an abnormality. It can be understood that there is no abnormal sound data in the training stage of the sound detection model. Therefore, when an abnormal sound is input during sound detection, corresponding losses such as RMSE (Root Mean Square Error) will generally be relatively high. Then, a reasonable root mean square error threshold can be set to determine whether the target sound data is an abnormal sound. Specifically, determine the Mel spectrogram corresponding to the historical sound data under normal operating conditions, and train the initial unsupervised autoencoder using the Mel spectrogram corresponding to the historical sound data and the preset control parameters to obtain a target unsupervised autoencoder. Then, use the target unsupervised autoencoder to determine the detection result corresponding to the target sound data. The sound data detection logic is as follows:

[0063] (1) Call the information storage module to obtain the target sound data;

[0064] (2) Determine the target Mel spectrogram corresponding to the target sound data, and use the target unsupervised autoencoder to extract the target eigenvalue corresponding to the target sound data based on the target Mel spectrogram, and calculate the target RMSE of the target sound data according to the target eigenvalue;

[0065] (3) If the target RMSE is greater than the preset root mean square error threshold, call the sound abnormality interface of the message module, and after message cross-verification passes, notify the inspection personnel;

[0066] (4) If the target RMSE is less than the preset root mean square error threshold, proceed to detect the next sound data.

[0067] Refer to Figure 4As shown, it should be noted that the above-mentioned detection of the target image data based on a preset deep learning visual model may include: training an initial deep learning visual model based on a preset transfer learning method and a preset image data set to obtain a target deep learning visual model, and extracting the target smoke texture feature, target flame color feature, and target dynamic optical flow feature of the target image data based on the target deep learning visual model; using the target deep learning visual model to determine the target anomaly probability of the target image data based on the target smoke texture feature, the target flame color feature, and the target dynamic optical flow feature; if the target anomaly probability is greater than a preset anomaly probability threshold, detecting the target image data based on a preset three-level verification mechanism. Specifically, the above-mentioned detection of the target image data based on a preset three-level verification mechanism may include: determining the target local pixels of the target image data based on a preset local area condition, so as to judge whether there is a target pixel cluster in the target image data that satisfies the preset flame spectrum feature condition based on the target local pixels; judging whether there is a target smoke area in the target image data that satisfies the preset edge blur feature condition based on a preset contour detection algorithm, and determining the historical image data corresponding to the target image data and the target illumination similarity of the target image data; determining the detection result corresponding to the target image data according to the target pixel cluster, the target smoke area, and the target illumination similarity. In a specific implementation manner, first, the initial deep learning visual model can be fine-tuned and trained through transfer learning and thousands of photos of computer room precision air conditioners with normal scenes and manually labeled flame and smoke abnormal scenes to obtain a target deep learning visual model. Then, the target deep learning visual model is used to detect the target image data. Specifically, first, the information storage module is called to obtain the target image data, each collected image is traversed, and after preprocessing, it is input into the target deep learning visual model, where the preprocessing includes but is not limited to resizing the image to a unified size, normalization processing, filtering processing, etc. The target smoke texture feature, target flame color feature, and target dynamic optical flow feature of the image are extracted by using the target deep learning visual model, and the target anomaly probability corresponding to the target image data is output according to the target smoke texture feature, target flame color feature, and target dynamic optical flow feature. If the target anomaly probability is greater than a preset anomaly probability threshold, for example, 0.9, the three-level verification mechanism can be triggered. The logic of the three-level verification mechanism is as follows:

[0068] (1) Analyze the local area pixels of the target image data to detect whether there is a pixel cluster in the target image data that satisfies the flame spectrum feature condition, that is, extract the red and orange highlighted areas in the target image data and detect the matching degree between the red and orange highlighted areas and the flame spectrum;

[0069] (2) Verify whether there is a smoke area that meets the edge blur feature condition in the target image data based on the contour detection algorithm, that is, verify whether there is a smoke diffusion form in the target image data through edge blur degree analysis;

[0070] (3) Determine the illumination similarity between the historical image data corresponding to the target image data and the target image data to exclude environmental interference.

[0071] When the three-level verification passes, the system calls the fire smoke alarm interface of the message module to temporarily store the message containing the device number, abnormal position coordinates, and screenshot. After the message cross-verification passes, the message is notified to the patrol personnel.

[0072] In this embodiment, the target operation data can be detected based on a preset static threshold algorithm and preset state conditions. Specifically, the system calls the information storage module to obtain the target operation data, and can perform threshold comparison for the return air temperature. When the value exceeds or is lower than then it can immediately call the message module to send a "temperature limit alarm" notification to the patrol personnel, along with the device number of the precision air conditioner in the computer room and the current temperature value. The system can judge whether there is a blockage through the filter clogging state, detect whether the power supply is interrupted in real time through the power supply state, monitor the change of the running and stopping states through the internal fan state. If any state is abnormal, it can immediately trigger the fault alarm of the corresponding device, such as "power interruption" or "fan stop". All alarm information includes the device number of the precision air conditioner in the computer room, the name of the abnormal index, the current value, and the occurrence time. After the message cross-verification passes, it is pushed to the patrol personnel through the message module.

[0073] Step S13: If any of the detection results indicates an abnormality, based on the preset multi-modal feature cross-verification mechanism, the target sound data, the target image data, and the target operation data, detect whether the target precision air conditioner in the computer room is the target faulty device.

[0074] In this embodiment, in order to reduce environmental interference, detecting whether the target precision air conditioner in the target computer room is a target faulty device based on the preset multi-modal feature cross-validation mechanism, the target sound data, the target image data, and the target operation data may include: determining a target time period based on the current moment and a second preset time condition, and obtaining the to-be-verified sound data, to-be-verified image data, and to-be-verified operation data within the target time period from the target heterogeneous data; if it is determined that the target precision air conditioner in the target computer room is the target faulty device based on the to-be-verified sound data, the to-be-verified image data, the to-be-verified operation data, and the preset multi-modal feature cross-validation mechanism, determining the abnormal data corresponding to the target precision air conditioner, and generating a corresponding target alarm message based on the abnormal data and the device number corresponding to the target precision air conditioner. In a specific implementation manner, if there is an abnormal detection result, the DCIM temperature curve, image data, and sound data within 3 minutes before and after the current moment can be synchronously retrieved to verify whether there are coupling features such as temperature, smoke diffusion, and abnormal sound. If so, it is determined that the target precision air conditioner in the target computer room is the target faulty device, a target alarm message including the device number, abnormal type, and abnormal data corresponding to the target precision air conditioner is generated, and the target alarm message is sent to the patrol personnel.

[0075] As can be seen from the above, in this embodiment, the target heterogeneous data of the target precision air conditioner in the target computer room is first collected in real time by the target heterogeneous device, and the target heterogeneous data includes target sound data, target image data, and target operation data; then the target sound data is detected based on the preset unsupervised autoencoder, the target image data is detected based on the preset deep learning vision model, and the target operation data is detected based on the preset static threshold algorithm and preset state conditions to obtain each detection result; if any of the detection results indicates an abnormality, it is detected whether the target precision air conditioner in the target computer room is a target faulty device based on the preset multi-modal feature cross-validation mechanism, the target sound data, the target image data, and the target operation data. As can be seen from the above, in this embodiment, the target sound data, target image data, and target operation data of the target precision air conditioner in the target computer room are collected in real time, and then the target sound data detection is realized by using the preset unsupervised autoencoder, the target image data is detected based on the preset deep learning vision model, the target operation data is detected in real time based on the preset static threshold algorithm and preset state conditions, and finally the device fault and environmental interference are effectively distinguished through the multi-modal feature cross-validation mechanism. In this way, this embodiment can realize the all-weather automatic monitoring of the operation status of the precision air conditioner in the computer room, have the ability to respond to abnormal events in seconds, and can also greatly reduce the frequency of manual inspections. In this way, this embodiment can greatly reduce the false alarm rate and significantly improve the reliability of the detection results of the precision air conditioner equipment in the computer room.

[0076] Correspondingly, referring to Figure 5As shown in the figure, an embodiment of the present application further provides an anomaly detection device, which may include:

[0077] A target heterogeneous data acquisition module 11, configured to collect target heterogeneous data of a target computer room precision air conditioner in real time through a target heterogeneous device, where the target heterogeneous data includes target sound data, target image data, and target operation data;

[0078] A detection result acquisition module 12, configured to detect the target sound data based on a preset unsupervised autoencoder, detect the target image data based on a preset deep learning vision model, and detect the target operation data based on a preset static threshold algorithm and a preset state condition to obtain respective detection results;

[0079] A target computer room precision air conditioner detection module 13, configured to, if any of the detection results indicates an anomaly, detect whether the target computer room precision air conditioner is a target faulty device based on a preset multimodal feature cross-verification mechanism, the target sound data, the target image data, and the target operation data.

[0080] As can be seen from the above, in the present application, first, target heterogeneous data of a target computer room precision air conditioner is collected in real time through a target heterogeneous device, where the target heterogeneous data includes target sound data, target image data, and target operation data; then, the target sound data is detected based on a preset unsupervised autoencoder, the target image data is detected based on a preset deep learning vision model, and the target operation data is detected based on a preset static threshold algorithm and a preset state condition to obtain respective detection results; if any of the detection results indicates an anomaly, it is detected whether the target computer room precision air conditioner is a target faulty device based on a preset multimodal feature cross-verification mechanism, the target sound data, the target image data, and the target operation data. As can be seen from the above, in the present application, the target sound data, target image data, and target operation data of the target computer room precision air conditioner are collected in real time, then the target sound data detection is implemented using a preset unsupervised autoencoder, the target image data is detected based on a preset deep learning vision model, the target operation data is detected in real time based on a preset static threshold algorithm and a preset state condition, and finally, the device faults and environmental interferences are effectively distinguished through a multimodal feature cross-verification mechanism. In this way, the present application can realize all-weather automatic monitoring of the operating state of the computer room precision air conditioner, have the ability to respond to anomaly events within seconds, and can also greatly reduce the frequency of manual inspections. In this way, the present application can greatly reduce the false alarm rate and significantly improve the reliability of the detection results of the computer room precision air conditioner equipment.

[0081] In some specific embodiments, the target heterogeneous data acquisition module 11 may include:

[0082] A target sound data acquisition unit, configured to determine a list of target sound collectors corresponding to the target precision air conditioner in the target machine room, and collect the target sound data in real time from each target sound collector in the list of target sound collectors according to a first preset time condition;

[0083] A target operation data acquisition unit, configured to use a target camera corresponding to the target precision air conditioner in the target machine room to collect the target image data in real time, and collect the target operation data of the target precision air conditioner in the target machine room in real time based on a target northbound interface of a preset data center infrastructure management system.

[0084] In some specific embodiments, the anomaly detection device may further include:

[0085] A data upload module, configured to upload the target sound data, the target image data, and the target operation data to a preset information storage list through a preset file upload interface;

[0086] A data identification module, configured to identify the target sound data based on a preset sound data identification method, identify the target image data based on a preset image data identification method, and identify the target operation data based on a preset operation data identification method;

[0087] A data binding module, configured to bind the target sound data, the target image data, and the target operation data to a device number corresponding to the target precision air conditioner in the target machine room.

[0088] In some specific embodiments, the detection result acquisition module 12 may include:

[0089] A target unsupervised autoencoder determination unit, configured to train an initial unsupervised autoencoder based on historical sound data under preset normal operating conditions and preset control parameters to obtain a target unsupervised autoencoder; wherein the preset control parameters include any one or a combination of several of batch size, number of iterations, and learning rate;

[0090] A target root mean square error determination unit, configured to determine a target Mel spectrogram corresponding to the target sound data, input the target Mel spectrogram into the target unsupervised autoencoder to obtain a target feature value corresponding to the target sound data, and determine a target root mean square error corresponding to the target sound data based on the target feature value;

[0091] A first detection result determination unit, configured to determine that the detection result corresponding to the target sound data indicates an anomaly when the target root mean square error is greater than a preset root mean square error threshold.

[0092] In some specific embodiments, the detection result acquisition module 12 may include:

[0093] A target image data feature extraction sub-module, which is used to train an initial deep learning vision model based on a preset transfer learning method and a preset image data set to obtain a target deep learning vision model, and extract the target smoke texture feature, target flame color feature and target dynamic optical flow feature of the target image data based on the target deep learning vision model;

[0094] A target abnormal probability determination sub-module, which is used to determine the target abnormal probability of the target image data based on the target smoke texture feature, the target flame color feature and the target dynamic optical flow feature by using the target deep learning vision model;

[0095] A target image data detection sub-module, which is used to detect the target image data based on a preset three-level verification mechanism if the target abnormal probability is greater than a preset abnormal probability threshold.

[0096] In some specific embodiments, the target image data detection sub-module may include:

[0097] A target local pixel determination unit, which is used to determine the target local pixel of the target image data based on a preset local area condition, so as to judge whether there is a target pixel cluster in the target image data that satisfies a preset flame spectral feature condition based on the target local pixel;

[0098] A target illumination similarity determination unit, which is used to judge whether there is a target smoke area in the target image data that satisfies a preset edge blur feature condition based on a preset contour detection algorithm, and determine the historical image data corresponding to the target image data and the target illumination similarity of the target image data;

[0099] A second detection result determination unit, which is used to determine the detection result corresponding to the target image data according to the target pixel cluster, the target smoke area and the target illumination similarity.

[0100] In some specific embodiments, the target computer room precision air conditioner detection module 13 may include:

[0101] A target time period determination unit, which is used to determine a target time period based on the current moment and a second preset time condition, and obtain the to-be-verified sound data, to-be-verified image data and to-be-verified operation data within the target time period from the target heterogeneous data;

[0102] A target alarm information generation unit, configured to determine abnormal data corresponding to the target precision air conditioner for the computer room if it is determined that the target precision air conditioner for the computer room is the target faulty device based on the to-be-verified sound data, the to-be-verified image data, the to-be-verified operation data, and the preset multi-modal feature cross-verification mechanism, and generate corresponding target alarm information based on the abnormal data and the device number corresponding to the target precision air conditioner for the computer room.

[0103] Further, an embodiment of the present application also discloses an electronic device. Figure 6 It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment, and the content in the figure cannot be considered as any limitation to the scope of use of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the abnormal detection method disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0104] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows is any communication protocol applicable to the technical solution of the present application, and specific limitations are not imposed thereon here; the input / output interface 25 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application needs, and specific limitations are not made here.

[0105] In addition, as a carrier for resource storage, the memory 22 may be a read-only memory, a random access memory, a disk, or an optical disc, etc., and the resources stored thereon may include an operating system 221, a computer program 222, etc., and the storage method may be short-term storage or permanent storage.

[0106] Among them, the operating system 221 is used to manage and control each hardware device and the computer program 222 on the electronic device 20, and it may be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the abnormal detection method executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program that can be used to complete other specific tasks.

[0107] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the foregoing disclosed anomaly detection method is implemented. For the specific steps of this method, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details will not be elaborated herein.

[0108] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and reference can be made to the description in the method part for relevant parts.

[0109] Those skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed in this article can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0110] The steps of the method or algorithm described in combination with the embodiments disclosed in this article can be directly implemented by hardware, a software module executed by a processor, or a combination of the two. The software module can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0111] Finally, it should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0112] The above has introduced the technical solution provided by this application in detail. Specific examples are used in this text to elaborate on the principle and implementation manner of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manner and application scope. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. An anomaly detection method, characterized in that, Including: Real-time collecting target heterogeneous data of a precision air conditioner in a target computer room through a target heterogeneous device, where the target heterogeneous data includes target sound data, target image data, and target operation data; Detecting the target sound data based on a preset unsupervised autoencoder, detecting the target image data based on a preset deep learning vision model, and detecting the target operation data based on a preset static threshold algorithm and preset state conditions to obtain respective detection results; If any of the detection results indicates an abnormality, detecting whether the precision air conditioner in the target computer room is a target faulty device based on a preset multi-modal feature cross-validation mechanism, the target sound data, the target image data, and the target operation data.

2. The anomaly detection method according to claim 1, wherein The real-time collecting of the target heterogeneous data of the precision air conditioner in the target computer room through the target heterogeneous device includes: Determining a list of target sound collectors corresponding to the precision air conditioner in the target computer room, and real-time collecting the target sound data from each target sound collector in the list of target sound collectors according to a first preset time condition; Using a target camera corresponding to the precision air conditioner in the target computer room to real-time collect the target image data, and real-time collecting the target operation data of the precision air conditioner in the target computer room based on a target northbound interface of a preset data center infrastructure management system.

3. The anomaly detection method according to claim 1, wherein After the real-time collecting of the target heterogeneous data of the precision air conditioner in the target computer room through the target heterogeneous device, it further includes: Uploading the target sound data, the target image data, and the target operation data to a preset information storage list through a preset file upload interface; Identifying the target sound data based on a preset sound data identification method, identifying the target image data based on a preset image data identification method, and identifying the target operation data based on a preset operation data identification method; Binding the target sound data, the target image data, and the target operation data to the device number corresponding to the precision air conditioner in the target computer room.

4. The anomaly detection method according to claim 1, wherein The detecting of the target sound data based on the preset unsupervised autoencoder includes: Training an initial unsupervised autoencoder based on historical sound data under preset normal working conditions and preset control parameters to obtain a target unsupervised autoencoder; where the preset control parameters include any one or a combination of several of batch size, number of iterations, and learning rate; Determining a target Mel spectrogram corresponding to the target sound data, and inputting the target Mel spectrogram into the target unsupervised autoencoder to obtain a target feature value corresponding to the target sound data, and determining a target root mean square error corresponding to the target sound data based on the target feature value; If the target root mean square error is greater than a preset root mean square error threshold, determining that the detection result corresponding to the target sound data indicates an abnormality.

5. The anomaly detection method according to claim 1, wherein The detecting of the target image data based on the preset deep learning vision model includes: Training an initial deep learning vision model based on a preset transfer learning method and a preset image data set to obtain a target deep learning vision model, and extracting a target smoke texture feature, a target flame color feature, and a target dynamic optical flow feature of the target image data based on the target deep learning vision model; Using the target deep learning vision model, determine the target anomaly probability of the target image data based on the target smoke texture feature, the target flame color feature, and the target dynamic optical flow feature; If the target anomaly probability is greater than a preset anomaly probability threshold, detect the target image data based on a preset three-level verification mechanism.

6. The anomaly detection method according to claim 5, characterized in that The detecting the target image data based on the preset three-level verification mechanism includes: Determine the target local pixels of the target image data based on a preset local area condition, so as to judge whether there is a target pixel cluster in the target image data that satisfies a preset flame spectral feature condition based on the target local pixels; Judge whether there is a target smoke area in the target image data that satisfies a preset edge blur feature condition based on a preset contour detection algorithm, and determine the historical image data corresponding to the target image data and the target light illumination similarity of the target image data; Determine the detection result corresponding to the target image data according to the target pixel cluster, the target smoke area, and the target light illumination similarity.

7. The anomaly detection method according to any one of claims 1 to 6, characterized in that The detecting whether the target precision air conditioner in the target computer room is a target faulty device based on the preset multi-modal feature cross-verification mechanism, the target sound data, the target image data, and the target operation data includes: Determine a target time period based on the current moment and a second preset time condition, and obtain the to-be-verified sound data, the to-be-verified image data, and the to-be-verified operation data within the target time period from the target heterogeneous data; If it is determined that the target precision air conditioner in the target computer room is the target faulty device based on the to-be-verified sound data, the to-be-verified image data, the to-be-verified operation data, and the preset multi-modal feature cross-verification mechanism, determine the abnormal data corresponding to the target precision air conditioner in the target computer room, and generate a corresponding target alarm message based on the abnormal data and the device number corresponding to the target precision air conditioner in the target computer room.

8. An anomaly detection device, characterized in that, Including: A target heterogeneous data acquisition module, configured to collect target heterogeneous data of a target precision air conditioner in a target computer room in real time through a target heterogeneous device, where the target heterogeneous data includes target sound data, target image data, and target operation data; A detection result acquisition module, configured to detect the target sound data based on a preset unsupervised autoencoder, detect the target image data based on a preset deep learning vision model, and detect the target operation data based on a preset static threshold algorithm and a preset state condition to obtain respective detection results; A target precision air conditioner detection module, configured to, if any of the detection results indicates an anomaly, detect whether the target precision air conditioner in the target computer room is a target faulty device based on a preset multi-modal feature cross-verification mechanism, the target sound data, the target image data, and the target operation data.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein, the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the anomaly detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, For storing a computer program, the computer program, when executed by a processor, implements the anomaly detection method according to any one of claims 1 to 7.

Images