Methods, apparatus, equipment, and media for monitoring the status of data centers.

By processing data center feature data through neural network models and utilizing pre-training techniques and self-attention mechanisms to learn equipment status, the problem of low efficiency in fault location in large data centers has been solved, achieving efficient and accurate status monitoring and fault location.

CN114564367BActive Publication Date: 2026-11-17INDUSTRIAL AND COMMERCIAL BANK OF CHINA
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202210228424.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-09
Publication Date
2026-11-17
Estimated Expiration
2042-03-09

AI Technical Summary

Technical Problem

Existing technologies for locating faults in large data centers are inefficient, make it difficult to analyze the status of each node within the data center cluster and the mutual influence between nodes, and consume a lot of manpower and resources.

Method used

The model uses a neural network model to process the characteristic data of the data center. The neural network model is trained through pre-training technology. The self-attention mechanism is used to learn the status of data center equipment and their mutual influence. Pre-training tasks are constructed to improve the model's ability to resist interference. The data center status is monitored through cluster status classifier and time-series status classifier.

Benefits of technology

It improves the efficiency and accuracy of data center status monitoring, can quickly locate faulty equipment, reduce the false alarm rate, and is suitable for various types of data centers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114564367B_ABST
    Figure CN114564367B_ABST
Patent Text Reader

Abstract

This disclosure provides a method and apparatus for monitoring the status of a data center, applicable to the field of artificial intelligence technology. The method includes: collecting feature data from N devices in the data center at R time points to obtain R feature data sequences, wherein a start flag is used to identify the starting position in each feature data sequence; obtaining a feature matrix based on each feature data sequence, wherein the start flag and the feature data from each of the N devices are respectively converted into feature vectors in the feature matrix; processing the feature matrix using a neural network model to obtain an output matrix; and obtaining the data center status information based on the vector corresponding to the start flag in the output matrix. This disclosure also provides a training method and apparatus for a neural network model for monitoring the status of a data center, as well as an electronic device, storage medium, and program product.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and more specifically, relates to a method and apparatus for monitoring the state of a data center, a method and apparatus for training a neural network model for monitoring the state of a data center, as well as an electronic device, medium and program product. Background Art

[0002] Large-scale data centers have relatively complex configurations. In daily production activities, if a fault occurs in any link, the troubleshooting of the root cause of the fault is relatively complicated. In the prior art, fault location methods for large-scale data centers are mainly divided into two categories: traditional alarm definition and alarm correlation analysis. However, both of these methods perform manual correlation analysis to troubleshoot the root cause of the fault after receiving an alarm signal, which makes it difficult to clearly analyze the state of each node in the data center cluster and the mutual influence between nodes. They are not only low in efficiency and consume a large amount of manpower and material resources, but also difficult to be universally applied to various types of data centers. Summary of the Invention

[0003] In view of the above problems, the present disclosure provides a method and apparatus for monitoring the state of a data center, a method and apparatus for training a neural network model for monitoring the state of a data center, as well as an electronic device, medium and program product, which can improve the efficiency and accuracy of data center state monitoring.

[0004] In a first aspect of the embodiments of the present disclosure, a method for monitoring the state of a data center is provided. The method includes: collecting respective feature data of N devices in the data center at R moments to obtain R feature data sequences, wherein a start flag bit is used to identify the start position in each of the feature data sequences, and R and N are integers greater than or equal to 1 respectively; obtaining a feature matrix based on each of the feature data sequences, wherein the start flag bit in the feature data sequence and the respective feature data of the N devices are correspondingly converted into feature vectors in the feature matrix; processing the feature matrix using a neural network model to obtain an output matrix output by the neural network model, wherein vectors in the output matrix are in one-to-one correspondence with the feature vectors in the feature matrix; wherein the neural network model is obtained by training through a pre-training technique; and obtaining state information of the data center based on the vector corresponding to the start flag bit in the output matrix, wherein the state information includes at least one of operation state information or time state information of the data center.

[0005] According to an embodiment of this disclosure, the status information includes the operational status information of the data center. Obtaining the status information of the data center based on the vector corresponding to the start flag in the output matrix includes: classifying the vector corresponding to the start flag in the output matrix using a cluster status classifier to obtain a first classification result; wherein the first classification result is used to indicate whether there is an anomaly in the data center.

[0006] According to an embodiment of this disclosure, the method further includes: when the first classification result indicates that the data center is abnormal, using a device status classifier corresponding to each of the N devices to classify the vectors corresponding to each device in the output matrix to obtain a second classification result; wherein the second classification result is used to indicate whether the device is faulty.

[0007] According to embodiments of this disclosure, the neural network model is trained as follows: M feature data sequences collected at M historical moments are obtained to obtain M historical feature data sequences for training, where M is an integer greater than or equal to 1; the feature matrix obtained based on each historical feature data sequence is used as a historical feature matrix; the historical feature matrix is ​​input into the neural network model to obtain a first historical output matrix output by the neural network model; the feature vectors corresponding to some devices in the historical feature matrix are data masked to obtain a half-masked feature matrix; the half-masked feature matrix is ​​input into the neural network model to obtain a second historical output matrix output by the neural network model; and the neural network model is back-optimized based on the difference between the first historical output matrix and the second historical output matrix.

[0008] According to embodiments of this disclosure, the step of data masking the feature vectors corresponding to a portion of the devices in the historical feature matrix to obtain a half-masked feature matrix includes at least one of the following: replacing the feature vectors corresponding to the portion of the devices with zero vectors; or replacing the feature vectors corresponding to the portion of the devices with the feature vectors corresponding to other devices.

[0009] According to an embodiment of this disclosure, when R is greater than 1 and the state information includes the time state information of the data center, obtaining the state information of the data center based on the vectors corresponding to the start flag in the output matrix includes: inputting each pair of vectors corresponding to the start flag in the output matrix into a time-series state classifier; and obtaining a third classification result output by the time-series state classifier, wherein the third classification result is used to indicate whether the times corresponding to the two paired input vectors have a sequential order.

[0010] According to an embodiment of this disclosure, the method further includes: sorting the R output matrices in chronological order according to the third classification result; and sorting the R feature data sequences and the R feature matrices accordingly according to the order of the R output matrices.

[0011] According to embodiments of this disclosure, the method further includes training the neural network model by: acquiring Q feature data sequences collected at Q historical moments to obtain Q historical feature data sequences for training, where Q is an integer greater than or equal to 1; obtaining a feature matrix based on each historical feature data sequence as a historical feature matrix; inputting the historical feature matrix into the neural network model to obtain a historical output matrix output by the neural network model; extracting a vector corresponding to a start flag bit from each historical output matrix as a historical flag bit vector; pairing the extracted Q historical flag bit vectors into two categories, wherein in one category, the moments corresponding to each pair of historical flag bit vectors have a chronological order, and in the other category, the moments corresponding to each pair of historical flag bit vectors have a random relationship; inputting each pair of historical flag bit vectors into the temporal state classifier to obtain the third classification result output by the temporal state classifier; and jointly training the neural network model and the temporal state classifier based on the difference between the third classification result output by the temporal state classifier and the category to which each pair of historical flag bit vectors belongs.

[0012] According to embodiments of this disclosure, the neural network model is equipped with a self-attention mechanism.

[0013] According to embodiments of this disclosure, the neural network model is a transformer model.

[0014] A second aspect of this disclosure also provides a method for training a neural network model for monitoring the status of a data center. The training method includes: acquiring feature data of N devices in the data center collected at M time points to obtain M historical feature data sequences, wherein a start flag is used to identify the starting position in each historical feature data sequence; wherein R and N are integers greater than or equal to 1; obtaining a historical feature matrix based on each historical feature data sequence, wherein the start flag in the historical feature data sequence and the feature data of the N devices are respectively converted into feature vectors in the historical feature matrix; inputting the historical feature matrix into the neural network model to obtain a first historical output matrix output by the neural network model; performing data masking on the feature vectors corresponding to some devices in the historical feature matrix to obtain a half-masked feature matrix; inputting the half-masked feature matrix into the neural network model to obtain a second historical output matrix output by the neural network model; and back-optimizing the neural network model based on the difference between the first historical output matrix and the second historical output matrix.

[0015] According to embodiments of this disclosure, the step of data masking the feature vectors corresponding to a portion of the devices in the historical feature matrix to obtain a half-masked feature matrix includes at least one of the following: replacing the feature vectors corresponding to the portion of the devices with zero vectors; or replacing the feature vectors corresponding to the portion of the devices with the feature vectors corresponding to other devices.

[0016] A third aspect of this disclosure also provides a method for training a neural network model for monitoring the status of a data center. The training method includes: acquiring feature data from N devices in the data center collected at Q time points to obtain Q historical feature data sequences, wherein a start position is identified in each historical feature data sequence by a start flag; wherein Q and N are integers greater than or equal to 1; obtaining a historical feature matrix based on each historical feature data sequence, wherein the start flag in the historical feature data sequence and the feature data of each of the N devices are respectively converted into feature vectors in the historical feature matrix; inputting the historical feature matrix into the neural network model to obtain a historical output matrix output by the neural network model; and extracting the vector corresponding to the start flag from each historical output matrix. The historical marker vectors are extracted and paired into two categories. In one category, the times corresponding to each pair of historical marker vectors are ordered, while in the other category, the times corresponding to each pair of historical marker vectors are randomly ordered. Each pair of historical marker vectors is input into a temporal state classifier to obtain a third classification result output by the temporal state classifier. The third classification result indicates whether the times corresponding to the two pairs of historical marker vectors are ordered. Based on the difference between the third classification result output by the temporal state classifier and the category to which each pair of historical marker vectors belongs, the neural network model and the temporal state classifier are optimized in reverse.

[0017] A fourth aspect of this disclosure provides an apparatus for monitoring the status of a data center. The apparatus includes a data acquisition module, a data preprocessing module, a neural network model, and a monitoring module. The data acquisition module acquires feature data from N devices in the data center at R time points, obtaining R feature data sequences, wherein a start flag is used to identify the starting position in each feature data sequence; where R and N are integers greater than or equal to 1. The data preprocessing module obtains a feature matrix based on each feature data sequence, wherein the start flag in the feature data sequence and the feature data from each of the N devices are respectively converted into feature vectors in the feature matrix. The neural network model is trained using pre-training techniques, wherein the neural network model processes the feature matrix and obtains an output matrix, wherein the vectors in the output matrix correspond one-to-one with the feature vectors in the feature matrix. The monitoring module obtains the status information of the data center based on the vectors in the output matrix corresponding to the start flag, the status information including at least one of the data center's operational status information or time status information.

[0018] According to an embodiment of this disclosure, the monitoring module includes a cluster state separator. The cluster state classifier is used to classify the vectors corresponding to the start flag in the output matrix to obtain a first classification result; wherein, the first classification result is used to indicate whether there is an anomaly in the data center.

[0019] According to embodiments of this disclosure, the monitoring module further includes N device status classifiers. Each of the N device status classifiers corresponds one-to-one with one of the N devices. Each device status classifier is used to classify the vector in the output matrix corresponding to that device to obtain a second classification result; wherein the second classification result is used to indicate whether a device has a fault.

[0020] According to embodiments of this disclosure, the monitoring module further includes a time-series state classifier. The time-series state classifier is further configured to take pairs of vectors corresponding to the start flag in every two output matrices as input and output a third classification result, the third classification result indicating whether the times corresponding to the two input vector pairs have a chronological order, so that the R output matrices are sorted according to their temporal chronological order.

[0021] A fifth aspect of this disclosure provides a training apparatus for a neural network model used to monitor the status of a data center. The training apparatus includes a data acquisition module, a data preprocessing module, a first type of pre-training task construction module, and a first pre-training module. The data acquisition module acquires feature data from N devices in the data center collected at M time points, resulting in M ​​historical feature data sequences. Each historical feature data sequence is identified by a start flag; where R and N are integers greater than or equal to 1. The data preprocessing module obtains a historical feature matrix based on each historical feature data sequence, where the start flag in the historical feature data sequence and the feature data from each of the N devices are respectively converted into feature vectors in the historical feature matrix. The first type of pre-training task construction module performs data masking on the feature vectors corresponding to some devices in the historical feature matrix to obtain a half-masked feature matrix. The first pre-training module is used to: input the historical feature matrix into the neural network model to obtain a first historical output matrix output by the neural network model; input the semi-masked feature matrix into the neural network model to obtain a second historical output matrix output by the neural network model; and optimize the neural network model in reverse based on the difference between the first historical output matrix and the second historical output matrix.

[0022] A sixth aspect of this disclosure provides a training apparatus for a neural network model used to monitor the status of a data center. The training apparatus includes a data acquisition module, a data preprocessing module, a second type of pre-training task construction module, and a second pre-training module. The data acquisition module acquires feature data from N devices in the data center collected at Q time points, resulting in Q historical feature data sequences. Each historical feature data sequence is identified by a start flag, where Q and N are integers greater than or equal to 1. The data preprocessing module obtains a historical feature matrix based on each historical feature data sequence, wherein the start flag in the historical feature data sequence and the feature data from each of the N devices are respectively converted into feature vectors in the historical feature matrix. The second type of pre-training task construction module is used for: inputting the historical feature matrix into the neural network model to obtain the historical output matrix output by the neural network model; extracting the vector corresponding to the start flag bit from each of the historical output matrices as historical flag bit vectors; and pairing the extracted Q historical flag bit vectors into two categories, wherein the times corresponding to each pair of historical flag bit vectors in one category have a chronological order, and the times corresponding to each pair of historical flag bit vectors in the other category have a random relationship; the second pre-training module is used for: inputting each pair of historical flag bit vectors into a temporal state classifier to obtain the third classification result output by the temporal state classifier, wherein the third classification result is used to indicate whether the times corresponding to the two pairs of input historical flag bit vectors have a chronological order; and optimizing the neural network model and the temporal state classifier in reverse based on the difference between the third classification result output by the temporal state classifier and the category to which each pair of historical flag bit vectors belongs.

[0023] A seventh aspect of this disclosure provides an electronic device. The electronic device includes one or more processors and one or more memories. The one or more memories are used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors perform the methods provided in the first, second, or third aspects described above.

[0024] An eighth aspect of this disclosure also provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the methods provided in the first, second, or third aspects described above.

[0025] A ninth aspect of this disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the methods provided in the first, second, or third aspects described above. Attached Figure Description

[0026] The foregoing contents, as well as other objects, features, and advantages of this disclosure, will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0027] Figure 1 The illustration schematically depicts an application scenario of a method, apparatus, device, medium, and program product for monitoring the status of a data center according to embodiments of the present disclosure;

[0028] Figure 2 A flowchart illustrating a method for monitoring data center status according to an embodiment of the present disclosure is shown schematically.

[0029] Figure 3 The schematic illustration shows the input-output process of a neural network model according to an embodiment of the present disclosure;

[0030] Figure 4 The schematic illustration shows a process architecture for monitoring the status of a data center according to an embodiment of the present disclosure;

[0031] Figure 5 The schematic illustration shows a process architecture for monitoring the status of a data center according to another embodiment of the present disclosure;

[0032] Figure 6 A flowchart illustrating a training method for a neural network model according to an embodiment of the present disclosure is shown schematically.

[0033] Figure 7 A flowchart illustrating a method for monitoring the status of a data center according to another embodiment of this disclosure is shown schematically.

[0034] Figure 8 A flowchart illustrating a training method for a neural network model according to another embodiment of the present disclosure is shown.

[0035] Figure 9 A block diagram of an apparatus for monitoring the status of a data center is shown schematically according to an embodiment of the present disclosure;

[0036] Figure 10 A block diagram schematically illustrates a training apparatus for a neural network model according to an embodiment of the present disclosure;

[0037] Figure 11 A block diagram schematically illustrates a training apparatus for a neural network model according to another embodiment of the present disclosure; and

[0038] Figure 12 A block diagram schematically illustrates an electronic device suitable for implementing a method for monitoring data center status or a training method for a neural network model according to embodiments of the present disclosure. Detailed Implementation

[0039] The embodiments of the present disclosure will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the disclosure. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of the present disclosure for ease of explanation. However, it will be apparent that one or more embodiments may be practiced without these specific details. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concepts of the present disclosure.

[0040] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0041] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0042] When using expressions such as "at least one of A, B, and C", they should generally be interpreted in accordance with the meaning that is commonly understood by a person skilled in the art (e.g., "a system having at least one of A, B, and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B, and C, etc.).

[0043] It should be understood in this document that any number of elements in the specification and figures are for illustrative purposes and not for limitation, and that any naming is for distinction only and has no limiting meaning.

[0044] This disclosure provides a method for monitoring the status of a data center. Specifically, firstly, feature data from N devices in the data center are collected at R time points, resulting in R feature data sequences. Each feature data sequence is marked with a start flag, where R and N are integers greater than or equal to 1. Then, a feature matrix is ​​obtained based on each feature data sequence, where the start flag and the feature data from each of the N devices are respectively converted into feature vectors in the feature matrix. Next, a neural network model is used to process the feature matrix, resulting in an output matrix. The vectors in the output matrix correspond one-to-one with the feature vectors in the feature matrix. The neural network model is trained using pre-training techniques. Subsequently, based on the vectors in the output matrix corresponding to the start flags, the status information of the data center is obtained. This status information may include at least one of the data center's operational status information or time status information.

[0045] According to embodiments of this disclosure, a neural network model can be obtained using pre-training technology to process the features of the data center collected at different times. The output data obtained after processing by the neural network model can be used to determine the operational status information of the data center (e.g., whether there are any anomalies), or the time information status (e.g., whether the data obtained at different times have a sequential order), and / or the status of various devices within the data center (e.g., whether each node has a fault), thereby effectively monitoring the status of the data center. It can be widely applied to various data centers and has strong versatility.

[0046] The core idea of ​​pre-training technology is to abandon random initialization of the model, pre-train the model with a large amount of unlabeled training data to obtain a general language representation containing contextual information, and then fine-tune the model according to the specific task to be completed.

[0047] Specifically, in this embodiment of the disclosure, when training the neural network model using pre-training technology, features of various devices in the data center collected at historical moments can be used to form a large amount of unlabeled data to train the neural network model. This allows the neural network model to learn, through self-attention learning, the individual states of each device in the data center at different times, the mutual influences between devices, and the overall state of the cluster. Then, a corresponding pre-training task can be constructed according to the purpose of monitoring the cluster to optimize the neural network model.

[0048] For example, a first-class pre-training task can be constructed by masking the features of some devices in the data center at a certain moment, and then inputting them into a neural network model. This allows the neural network model to use pre-learned knowledge to combat interference in the data, achieving the goal of obtaining the correct output from all devices in the cluster even with only partially real input. In this way, the ability of the neural network model to resist various interference or random factors in the input data can be improved, thereby enhancing the stability and accuracy of using the neural network model to monitor the status of the data center.

[0049] For example, a second type of pre-training task can be constructed, enabling the neural network model to learn the characteristics of individual devices and the overall changes in the data center over time. This allows it to distinguish whether data at two different moments have a temporal order. Thus, when the neural network model is used to monitor the data center's status, if an anomaly is detected at one or more moments, the time range affected by the anomaly can be determined by sorting the collected data in chronological order. This includes information such as the start time, end time, and duration of the anomaly. This facilitates accurate troubleshooting of data center faults over time.

[0050] Based on the first and second type of pre-training tasks described above, embodiments of this disclosure also provide a method for training a neural network model for monitoring the status of a data center. Furthermore, embodiments of this disclosure also provide apparatus, electronic devices, media, and program products corresponding to the methods of the various embodiments of this disclosure.

[0051] It should be noted that the methods, apparatus, electronic devices, media and program products provided in the various embodiments of this disclosure can be used in the financial field, or in any field other than the financial field. This disclosure does not limit the application field.

[0052] Figure 1 This illustration schematically depicts an application scenario of a method, apparatus, device, medium, and program product for monitoring data center status according to embodiments of the present disclosure. It should be noted that... Figure 1 The examples shown are merely examples of system architectures that can be applied to the embodiments of this disclosure, in order to help those skilled in the art understand the technical content of this disclosure, but do not mean that the embodiments of this disclosure cannot be used in other devices, systems, environments or scenarios.

[0053] like Figure 1As shown, application scenario 100 according to this embodiment may include at least one terminal device (three are shown in the figure, 101, 102, and 103), a server 104, and a data center 105. Terminal devices 101, 102, and 103 communicate with server 104, and data center 105 communicates with server 104. Data center 105 includes a cluster of N devices 10, where N is an integer greater than or equal to 1.

[0054] Users can interact with server 105 using terminal devices 101, 102, and 103 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as email clients, data center status monitoring software, etc. (for example only). Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0055] Server 104 may be a server that provides various services, such as a backend management server that supports the websites browsed by users using terminal devices 101, 102, and 103 (for example only).

[0056] Server 104 can execute the method for monitoring the status of a data center according to embodiments of this disclosure, and monitor the status of data center 105. A neural network model may be deployed in server 104. This neural network model may include a self-attention mechanism. In one embodiment, the neural network model may be a transformer model. This neural network model can be trained using pre-training techniques.

[0057] Server 104 can collect characteristic data of each device 10 in data center 105 by communicating with data center 105, and then process the data according to the method of this disclosure embodiment to monitor the status of data center 105, and can send the monitoring results to terminal devices 101, 102, 103 for users to view.

[0058] It should be noted that the method for monitoring the status of a data center provided in this disclosure can generally be executed by server 104. Correspondingly, the apparatus for monitoring the status of a data center provided in this disclosure can generally be located in server 104. The method for monitoring the status of a data center provided in this disclosure can also be executed by a server or server cluster that is different from server 104 and capable of communicating with terminal devices 101, 102, 103 and data center 105. Correspondingly, the apparatus for monitoring the status of a data center provided in this disclosure can also be located in a server or server cluster that is different from server 104 and capable of communicating with terminal devices 101, 102, 103 and data center 105.

[0059] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0060] The following will be based on Figure 1 The described scene, through Figures 2-8 The present disclosure provides a detailed description of a method for monitoring the status of a data center, and a method for training a neural network model involved therein.

[0061] Figure 2 A flowchart illustrating a method for monitoring data center status according to an embodiment of the present disclosure is shown schematically.

[0062] like Figure 2 As shown, the method for monitoring the status of a data center according to an embodiment of this disclosure may include operations S210 to S240.

[0063] In operation S210, feature data of each of the N devices 10 in data center 105 are collected at R time points to obtain R feature data sequences. In each feature data sequence, the starting position is identified by a start flag bit, where R and N are integers greater than or equal to 1.

[0064] According to embodiments of this disclosure, characteristic data is collected from each device 10 within the cluster of data center 105, including but not limited to CPU, memory, and I / O characteristic data. CPU-related characteristic data may include, for example, CPU utilization, CPU operating speed, and CPU power. Memory-related characteristic data may include, for example, memory utilization and memory read / write speed. I / O-related characteristic data may include, for example, I / O throughput, average I / O response time, and number of I / O operations per unit time. In one embodiment, hundreds of metrics can be collected for each device 10, meaning these hundreds of metrics can be used to characterize a specific device 10 in the cluster at a given moment.

[0065] Each feature data sequence consists of feature data from N devices 10 at that time, along with a start flag. That is, hundreds of indicators from each device 10 correspond to a single unit in this feature data sequence. Thus, the feature data sequence contains N such units arranged in a predetermined order. When data is continuously collected from the data center 105, the start flags can be used to distinguish feature data sequences from different times.

[0066] In operation S220, a feature matrix is ​​obtained based on each feature data sequence, wherein the start flag bit in the feature data sequence and the feature data of each of the N devices 10 are respectively converted into feature vectors in the feature matrix.

[0067] The start flag and the feature data of each of the N devices 10 are processed and converted into feature vectors. For example, the start flag can be processed into a zero vector or a vector associated with the acquisition time, while the feature data of each of the N devices 10 are processed into data of uniform length, uniform specification or format, thereby obtaining the corresponding feature vectors.

[0068] For example, the feature data collected from each device 10 can be cleaned, and outliers (e.g., garbled characters or special characters in a certain indicator) can be padded with 0s. The feature data of each device 10 can then be padded to a uniform length X (e.g., X = 156). If some indicators are not collected sufficiently or data is lost during transmission, the insufficient length can be padded with 0s. If excessively long data is collected, it can be truncated to length X. In this way, the feature data of each device is processed into uniform length values.

[0069] Furthermore, according to embodiments of this disclosure, the feature data of each device 10 can be encoded according to its device type. For example, a server running Oracle uses encoding parameter 1 for encoding; a server running MySQL uses encoding parameter 2 for encoding. During encoding, in one embodiment, the encoding parameter can be added to each bit of the feature data of length X. In another embodiment, an additional bit can be added to the end of the feature data of length X to fill in the encoding parameter.

[0070] Furthermore, normalization can be performed on the cleaned and / or encoded data.

[0071] In operation S230, a neural network model is used to process the feature matrix, resulting in an output matrix. The vectors in the output matrix correspond one-to-one with the eigenvectors in the feature matrix. The neural network model is trained using pre-training techniques. The input and output process of the neural network model can be found in [reference needed]. Figure 3 The illustration.

[0072] Figure 3 The input and output process of a neural network model according to an embodiment of the present disclosure is illustrated schematically.

[0073] like Figure 3 As shown, the input feature matrix to the neural network model 30 is, for example, [s, Machine1, Machine2, Machine3, Machine4, ..., MachineN]. The neural network model 30 is trained using pre-training techniques. As described above, pre-training techniques enable the neural network model 30 to learn, through self-attention learning, the states of N devices 10 in the data center 105, the mutual influences between the devices 10, and the relationship between the overall state of the cluster and the relationships between the devices 10 in the cluster. Based on the learned knowledge, the neural network model 30 can obtain the output matrix [Fs, F1, F2, F3, F4, ..., FN] through self-attention mechanism learning and processing. The vectors in the output matrix correspond one-to-one with the eigenvectors in the feature matrix. Of course, the length of each vector in the output matrix can be different from the length of the eigenvectors in the feature matrix, which is determined according to the output dimension settings of the neural network model 30.

[0074] In operation S240, based on the vector corresponding to the start flag bit in the output matrix, the status information of data center 105 is obtained. This status information includes at least one of the operating status information or time status information of data center 105.

[0075] like Figure 3 As shown, since the start flag in the feature matrix is ​​converted into the feature vector S, the vector in the output matrix corresponding to the start flag is Fs.

[0076] According to embodiments of this disclosure, when training the neural network model 30, corresponding pre-training tasks can be constructed, such as the first type of training task and the second type of training task mentioned above. The vector Fs in the output matrix of the neural network model 30 contains the operating status information of the data center 105 (e.g., whether there is an anomaly) and / or the temporal status information of the data center 105 (e.g., whether two successively obtained feature data sequences have a temporal order). Furthermore, the operating status and / or temporal transition characteristics of the data center 105 can be monitored through further processing of the vector Fs.

[0077] Figure 4 The schematic illustration shows a process architecture for monitoring the status of a data center according to an embodiment of the present disclosure.

[0078] like Figure 4As shown, in one embodiment, the vector Fs corresponding to the start flag in the output matrix can be used as the input of the cluster state classifier 41. The cluster state classifier 41 uses the vector Fs to perform classification, obtaining a first classification result A1. The first classification result A1 is used to indicate whether there is an anomaly in the data center 105. The cluster state classifier 41 is a machine learning model. For example, the cluster state classifier 41 can be trained using the output matrix of the neural network model 30 as the training input data and the operating state of the data center 105 at each historical time as the label, given that the anomaly of the data center 105's operating state at each historical time is known.

[0079] Alternatively, in another embodiment, the vector corresponding to the start flag in every two output matrices can be...<Fs-i,Fs-j> Pairs are input into the temporal state classifier 42, where i ≠ j, and Fs-i and Fs-j represent vectors obtained from data collected at different times i and j. Then, the temporal state classifier 42 is used to...<Fs-i,Fs-j> The relationships between the input vectors are classified, and a third classification result A3 is output. The third classification result A3 is used to indicate whether the times corresponding to the two input vectors have a chronological order.

[0080] During actual data acquisition, factors such as network latency and signal interference may occur, causing the data collected from data center 105 to arrive at server 104 in a different order. For example, data collected earlier may arrive at server 104 later due to network latency. Therefore, according to the embodiments of this disclosure, when monitoring the status of the data center, it is necessary to verify and confirm the order of the characteristic data sequences collected in operation S210 in order to accurately troubleshoot the root cause in case of a fault.

[0081] The temporal state classifier 42 can be a classifier obtained by learning and training the time transition characteristics of Fs in the output matrix. For example, the neural network model 30 and the temporal state classifier 42 can be jointly trained so that the vector Fs corresponding to the start flag in the output matrix of the neural network model 30 can reflect the temporal characteristics of the data center 105.

[0082] Figure 5 The schematic illustration shows a process architecture for monitoring the status of a data center according to another embodiment of the present disclosure.

[0083] Combination Figure 5When the first classification result A1 output by the cluster status classifier 41 indicates that there is an anomaly in the data center 105, the device status classifier 50 corresponding to each of the N devices 10 can be used to classify the vector (i.e., one of F1 to FN) corresponding to each device 10 in the output matrix to obtain the second classification result A2. The second classification result A2 is used to indicate whether the corresponding device 10 is faulty. This allows the faulty device 10 within the data center 105 to be located quickly, facilitating rapid troubleshooting of the root cause of the fault.

[0084] Figure 6 A flowchart illustrating a method for training a neural network model according to an embodiment of the present disclosure is shown. Figure 6 The flowchart illustrates the process of constructing the first type of pre-training task to train the neural network model 30.

[0085] like Figure 6 As shown, the training method may include operations S610 to S660.

[0086] In operation S610, M feature data sequences collected at M historical moments are obtained to obtain M historical feature data sequences for training, where M is an integer greater than or equal to 1.

[0087] In operation S620, the feature matrix obtained based on each historical feature data sequence is used as the historical feature matrix.

[0088] In operation S630, the historical feature matrix is ​​input into the neural network model 30 to obtain the first historical output matrix output by the neural network model 30.

[0089] When operating the S640, the feature vectors corresponding to some devices in the historical feature matrix are masked to obtain a half-masked feature matrix.

[0090] For example, the feature vectors corresponding to some devices can be replaced with zero vectors, or the feature vectors corresponding to some devices can be replaced with feature vectors corresponding to other devices 10. The percentage of devices in the data center 105 cluster that are subject to data masking can be set. In one embodiment, n% (the value of n% is designed as needed, for example, no more than 20%) of the devices in the cluster can be randomly selected for data modification. The data of some of these n% of devices can be replaced with zero vectors, the data of another portion of the devices can be randomly replaced with data from other devices in the cluster, while the data of the remaining devices remains unchanged.

[0091] In operation S650, the semi-masked feature matrix is ​​input into the neural network model 30 to obtain the second historical output matrix output by the neural network model 30. The input / output process of the neural network model 30 in operations S630 and S650 in this embodiment can be referred to... Figure 3 The illustration.

[0092] In operation S660, the neural network model 30 is optimized in reverse based on the difference between the first historical output matrix and the second historical output matrix.

[0093] The desired outcome of this training is that even if the data input to the neural network model 30 is partially missing or partially out of order, the neural network model 30 can still output the correct vector for all devices 10 based on the actual state of most devices 10 in the data center 105 (i.e., Figure 3 (F1 to FN in the model). For example, after this training, even if machine2 is replaced by the zero vector and machine4 is replaced by machine1 in the feature matrix, the neural network model 30 will still output... Figure 3 The output matrix in the process.

[0094] In this way, the neural network model 30 can fully learn the interrelationships and mutual influences among the N devices 10, and thereby improve the ability of the neural network model 30 to resist various interference factors or random factors in the input data. After all, data loss during data acquisition, or network latency or interference during data transmission, can affect the authenticity and integrity of the data obtained by the server 104.

[0095] As can be seen, according to the embodiments of this disclosure, the stability and accuracy of using the neural network model 30 to monitor the status of the data center 105 can be improved, thereby reducing the misjudgment rate when using the neural network model 30 to monitor the status of the data center 105.

[0096] Figure 7 A flowchart illustrating a method for monitoring the status of a data center according to another embodiment of this disclosure is shown schematically.

[0097] like Figure 7 As shown, the method for monitoring the status of a data center according to this embodiment may include operations S210 to S230 and operations S740 to S770.

[0098] First, through operations S210 to S230, the output matrix of neural network model 30 is obtained. For details, please refer to the relevant description above; it will not be repeated here.

[0099] Next, in the S740 operation, the vectors corresponding to the start flag in every two output matrices are paired and input into the time-series state classifier 42. For more information, please refer to... Figure 4 Mid-vector<Fs-i,Fs-j> Pairs are input into the time-series state classifier 42.

[0100] Then, by operating S750, the third classification result A3 of the temporal state classifier 42 is obtained. The third classification result A3 is used to indicate whether the times corresponding to the two pairs of input vectors have a sequential order. For example, the vectors in all output matrices can be traversed in pairs to find the output matrices that are sequentially adjacent in time for each output matrix.

[0101] After this, in operation S760, the R output matrices can be sorted according to their chronological order based on the third classification result A3.

[0102] And in operation S770, the R feature data sequences and R feature matrices are sorted in the order of the R output matrices.

[0103] According to embodiments of this disclosure, when an anomaly occurs in the operation of data center 105, the start time of the anomaly and its spread over time can be determined based on the sorting results. Furthermore, when combined with... Figure 5 Once the method shown is used to understand the status of faulty equipment inside data center 105, it is possible to clearly understand how the anomaly spreads within data center 105 over time, facilitating rapid troubleshooting.

[0104] Figure 8 A flowchart illustrating a method for training a neural network model according to another embodiment of this disclosure is shown. Figure 8 The flowchart illustrates the process of constructing a second type of pre-training task to train the neural network model 30.

[0105] like Figure 8 As shown, the training method may include operations S810 to S870.

[0106] In operation S810, Q feature data sequences collected at Q historical time points are obtained to obtain Q historical feature data sequences for training, where Q is an integer greater than or equal to 1.

[0107] In operation S820, the feature matrix obtained based on each historical feature data sequence is used as the historical feature matrix.

[0108] In operation S830, the historical feature matrix is ​​input into neural network model 30, resulting in the historical output matrix of neural network model 30. The input and output process of neural network model 30 can be found in [reference needed]. Figure 3The illustration.

[0109] In operation S840, a vector corresponding to the start flag is extracted from each historical output matrix to serve as the historical flag vector.

[0110] In operation S850, the extracted Q historical flag bit vectors are paired and divided into two categories, with the pairs of historical flag bit vectors arranged in a predetermined order. In one category, the times corresponding to each pair of historical flag bit vectors are sequential, while in the other category, the times corresponding to each pair of historical flag bit vectors are random.

[0111] In operation S860, each pair of historical flag bit vectors is input into the temporal state classifier 42 to obtain the third classification result A3 output by the temporal state classifier 42.

[0112] In operation S870, based on the difference between the third classification result A3 output by the temporal state classifier 42 and the category to which each pair of historical flag vectors belong, the neural network model 30 and the temporal state classifier 42 are jointly trained. In this way, the vector Fs corresponding to the start flag in the output matrix of the neural network model 30 can reflect the temporal characteristics of the data center 105.

[0113] Based on the methods for monitoring data center status and the training methods for neural network models described in the above embodiments, this disclosure also provides a device for monitoring data center status and a training device for a neural network model for monitoring data center status. The following will be combined with... Figures 9-11 These devices are described in detail.

[0114] Figure 9 A block diagram of an apparatus 900 for monitoring the status of a data center is shown schematically according to an embodiment of the present disclosure.

[0115] like Figure 9 As shown, the device 900 for monitoring the status of a data center may include a data acquisition module 910, a data preprocessing module 920, a neural network model 930, and a monitoring module 940. The device 900 can be used to perform the aforementioned... Figures 2-5 ,and Figure 7 The method described.

[0116] The data acquisition module 910 is used to acquire feature data from N devices in the data center at R time points, resulting in R feature data sequences. Each feature data sequence is marked with a start flag to indicate its starting position; where R and N are integers greater than or equal to 1. In one embodiment, the data acquisition module 910 can perform the operation S210 described above.

[0117] The data preprocessing module 920 is used to obtain a feature matrix based on each feature data sequence, wherein the start flag bit in the feature data sequence and the feature data of each of the N devices are respectively converted into feature vectors in the feature matrix. In one embodiment, the data preprocessing module 920 can refer to the operation S220 described above.

[0118] The neural network model 930 is trained using pre-training techniques. The neural network model 930 processes the feature matrix and obtains an output matrix, where each vector in the output matrix corresponds one-to-one with an eigenvector in the feature matrix. In one embodiment, the neural network model 930 is used to perform the operation S230 described above.

[0119] The monitoring module 940 is used to obtain the status information of the data center based on the vector corresponding to the start flag bit in the output matrix. The status information includes at least one of the data center's operational status information or time status information. In one embodiment, the monitoring module 940 can perform the operation S240 described above.

[0120] According to embodiments of this disclosure, the monitoring module 940 includes a cluster state separator. The cluster state classifier is used to classify the vectors corresponding to the start flag bits in the output matrix to obtain a first classification result; wherein the first classification result is used to indicate whether there is an anomaly in the data center.

[0121] According to embodiments of this disclosure, the monitoring module 940 further includes N device status classifiers. Each of the N device status classifiers corresponds one-to-one with one of the N devices. Each device status classifier is used to classify the vector corresponding to that device in the output matrix to obtain a second classification result; wherein the second classification result is used to indicate whether a device is faulty.

[0122] According to embodiments of this disclosure, the monitoring module 940 further includes a temporal state classifier. The temporal state classifier is further configured to take pairs of vectors corresponding to the start flag in each pair of output matrices as input and output a third classification result. The third classification result indicates whether the times corresponding to the two input vector pairs have a chronological order, so that the R output matrices are sorted according to their temporal order. Accordingly, the monitoring module 940 can execute operations S740 to S770 described above.

[0123] Figure 10 A block diagram of a training apparatus 1000 for a neural network model according to an embodiment of the present disclosure is shown schematically.

[0124] like Figure 10As shown, the training device 1000 for monitoring the status of a neural network model may include a data acquisition module 1010, a data preprocessing module 1020, a first type of pre-training task construction module 1030, and a first pre-training module 1040. This training device 1000 can be used to perform reference... Figure 6 The training method described.

[0125] The data acquisition module 1010 is used to acquire feature data from N devices in the data center collected at M time points, resulting in M ​​historical feature data sequences. Each historical feature data sequence is marked with a start flag, where R and N are integers greater than or equal to 1. In one embodiment, the data acquisition module 1010 can perform the operation S610 described above.

[0126] The data preprocessing module 1020 is used to obtain a historical feature matrix based on each historical feature data sequence, wherein the start flag bit in the historical feature data sequence and the feature data of each of the N devices are respectively converted into feature vectors in the historical feature matrix. In one embodiment, the data preprocessing module 1020 can be used to perform the operation S620 described above. In one embodiment, the data processing procedures of the data processing module 1020 and the data processing module 920 are consistent.

[0127] The first type of pre-training task construction module 1030 is used to perform data masking on the feature vectors corresponding to some devices in the historical feature matrix to obtain a half-masked feature matrix. In one embodiment, the first type of pre-training task construction module 1030 can perform the operation S640 described above.

[0128] The first pre-training module 1040 is used to: input the historical feature matrix into the neural network model to obtain a first historical output matrix output by the neural network model; input the semi-masked feature matrix into the neural network model to obtain a second historical output matrix output by the neural network model; and optimize the neural network model in reverse based on the difference between the first historical output matrix and the second historical output matrix. In one embodiment, the first pre-training module 1040 can perform the operations S630, S650 and S660 described above.

[0129] Figure 11 A block diagram of a training apparatus 1100 for a neural network model according to another embodiment of the present disclosure is shown schematically.

[0130] like Figure 11As shown, the training device 1100 for a neural network model used to monitor data center status may include a data acquisition module 1110, a data preprocessing module 1120, a second type of pre-training task construction module 1130, and a second pre-training module 1140. This device 1100 can be used to implement a reference... Figure 8 The training method described.

[0131] The data acquisition module 1110 is used to acquire feature data from N devices in the data center collected at Q time points, resulting in Q historical feature data sequences. Each historical feature data sequence is marked with a start flag, where Q and N are integers greater than or equal to 1. In one embodiment, the data acquisition module 1110 can be used to perform operation S810. In one embodiment, the data processing procedure of the data acquisition module 1110 is consistent with that of the data acquisition module 1010.

[0132] The data preprocessing module 1120 is used to obtain a historical feature matrix based on each historical feature data sequence, wherein the start flag bit in the historical feature data sequence and the feature data of each of the N devices are respectively converted into feature vectors in the historical feature matrix. In one embodiment, the data preprocessing module 1120 can perform the operation S820 described above. In one embodiment, the data preprocessing module 1120 has the same data processing procedure as the data preprocessing module 1020.

[0133] The second type of pre-training task construction module 1130 is used to: input the historical feature matrix into the neural network model to obtain the historical output matrix of the neural network model; extract the vector corresponding to the start flag bit from each historical output matrix as the historical flag bit vector; and pair the extracted Q historical flag bit vectors into two categories, wherein the time corresponding to each pair of historical flag bit vectors in one category has a chronological order, and the time corresponding to each pair of historical flag bit vectors in the other category is random. In one embodiment, the second type of pre-training task construction module 1130 can perform the operations S840 and S850 described above.

[0134] The second pre-training module 1140 is used to: input each pair of historical flag bit vectors into a temporal state classifier to obtain a third classification result output by the temporal state classifier, the third classification result indicating whether the times corresponding to the two input pairs of historical flag bit vectors have a chronological order; and based on the difference between the third classification result output by the temporal state classifier and the category to which each pair of historical flag bit vectors belongs, to back-optimize the neural network model and the temporal state classifier. In one embodiment, the second pre-training module 1140 can perform operations S860 and S870 described above.

[0135] According to some embodiments of this disclosure, at least one of training device 1000 or training device 1100 can be integrated into training device 900 for monitoring the status of a neural network model in a data center, for training the neural network model 930 accordingly.

[0136] According to embodiments of this disclosure, any multiple modules selected from the data acquisition module 910, data preprocessing module 920, neural network model 930, monitoring module 940, data acquisition module 1010, data preprocessing module 1020, first type pre-training task construction module 1030, first pre-training module 1040, data acquisition module 1110, data preprocessing module 1120, second type pre-training task construction module 1130, or second pre-training module 1140 can be combined into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this disclosure, at least one of the following components—data acquisition module 910, data preprocessing module 920, neural network model 930, monitoring module 940, data acquisition module 1010, data preprocessing module 1020, first type pre-training task construction module 1030, first pre-training module 1040, data acquisition module 1110, data preprocessing module 1120, second type pre-training task construction module 1130, or second pre-training module 1140—can be at least partially implemented as hardware circuitry, such as a field-programmable gate array (FPGA), a programmable logic array (PLA), a system-on-a-chip, a system-on-a-substrate, a system-on-package, an application-specific integrated circuit (ASIC), or any other reasonable means of integrating or packaging the circuitry, or implemented in software, hardware, or firmware, or in any suitable combination of any of these three implementation methods. Alternatively, at least one of the following modules may be implemented, at least partially, as a computer program module: data acquisition module 910, data preprocessing module 920, neural network model 930, monitoring module 940, data acquisition module 1010, data preprocessing module 1020, first type pre-training task construction module 1030, first pre-training module 1040, data acquisition module 1110, data preprocessing module 1120, second type pre-training task construction module 1130, or second pre-training module 1140. When the computer program module is run, it can perform the corresponding function.

[0137] Figure 12 A block diagram schematically illustrates an electronic device suitable for implementing a method for monitoring data center status or a training method for a neural network model according to embodiments of the present disclosure.

[0138] like Figure 12As shown, an electronic device 1200 according to an embodiment of the present disclosure includes a processor 1201, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1202 or a program loaded from a storage portion 1208 into a random access memory (RAM) 1203. The processor 1201 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1201 may also include onboard memory for caching purposes. The processor 1201 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0139] RAM 1203 stores various programs and data required for the operation of electronic device 1200. Processor 1201, ROM 1202, and RAM 1203 are interconnected via bus 1204. Processor 1201 performs various operations of the method flow according to embodiments of the present disclosure by executing programs in ROM 1202 and / or RAM 1203. It should be noted that the programs may also be stored in one or more memories other than ROM 1202 and RAM 1203. Processor 1201 may also perform various operations of the method flow according to embodiments of the present disclosure by executing programs stored in said one or more memories.

[0140] According to embodiments of this disclosure, the electronic device 1200 may further include an input / output (I / O) interface 1205, which is also connected to the bus 1204. The electronic device 1200 may also include one or more of the following components connected to the I / O interface 1205: an input section 1206 including a keyboard, mouse, etc.; an output section 1207 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1208 including a hard disk, etc.; and a communication section 1209 including a network interface card such as a LAN card, modem, etc. The communication section 1209 performs communication processing via a network such as the Internet. A drive 1210 is also connected to the I / O interface 1205 as needed. A removable medium 1211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 1210 as needed so that computer programs read from it can be installed into the storage section 1208 as needed.

[0141] This disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs that, when executed, implement the method according to the embodiments of this disclosure.

[0142] According to embodiments of this disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, such as including, but not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this disclosure, the computer-readable storage medium may include ROM 1202 and / or RAM 1203 and / or one or more memories other than ROM 1202 and RAM 1203 described above.

[0143] Embodiments of this disclosure also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this disclosure.

[0144] When the computer program is executed by the processor 1201, it performs the functions defined in the system / apparatus of this disclosure embodiments. According to embodiments of this disclosure, the systems, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0145] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 1209, and / or installed from the removable medium 1211. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0146] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1209, and / or installed from the removable medium 1211. When the computer program is executed by the processor 1201, it performs the functions defined in the system of this disclosure embodiment. According to embodiments of this disclosure, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0147] According to embodiments of this disclosure, program code for executing the computer programs provided in embodiments of this disclosure can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C", or similar programming languages. The program code can execute entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0148] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0149] Those skilled in the art will understand that the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways, even if such combinations or combinations are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure can be combined or combined in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or combinations fall within the scope of this disclosure.

[0150] The embodiments of this disclosure have been described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of this disclosure. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this disclosure is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this disclosure, and all such substitutions and modifications should fall within the scope of this disclosure.

Claims

1. A method for monitoring the status of a data center, comprising: The feature data of N devices in the data center are collected at R time points to obtain R feature data sequences. The starting position of each feature data sequence is identified by a start flag. The feature data of each device includes at least one of CPU features, memory features and IO features. The start flag is used to distinguish feature data sequences at different time points. R and N are integers greater than or equal to 1. A feature matrix is ​​obtained based on each of the feature data sequences, wherein the start flag in the feature data sequence and the feature data of each of the N devices are respectively converted into feature vectors in the feature matrix, including: processing the start flag into a vector associated with the acquisition time, and processing the feature data of each of the N devices into data of uniform length and uniform specification or format; The feature matrix is ​​processed using a neural network model to obtain an output matrix, wherein the vectors in the output matrix correspond one-to-one with the feature vectors in the feature matrix; wherein the neural network model is trained using pre-training techniques; and Based on the vector corresponding to the start flag in the output matrix, the status information of the data center is obtained. This status information includes the data center's operational status information and time status information, including: The vectors corresponding to the start flag in the output matrix are classified using a cluster status classifier to obtain a first classification result, wherein the first classification result is information indicating whether the data center has abnormal operating status; and When R is greater than 1, the vectors corresponding to the start flag in each pair of output matrices are input into the temporal state classifier, and the third classification result output by the temporal state classifier is obtained. The third classification result is used to indicate whether the times corresponding to the two pairs of input vectors have a sequential time state information.

2. The method of claim 1, wherein, The method further includes: When the first classification result indicates that the data center is abnormal, the device status classifier corresponding to each of the N devices is used to classify the vector corresponding to each device in the output matrix to obtain a second classification result; wherein, the second classification result is used to indicate whether the device is faulty.

3. The method of claim 1, wherein, The neural network model is trained in the following manner: Obtain M feature data sequences collected at M historical moments to obtain M historical feature data sequences for training, where M is an integer greater than or equal to 1; The feature matrix obtained based on each of the historical feature data sequences is used as the historical feature matrix; The historical feature matrix is ​​input into the neural network model to obtain the first historical output matrix output by the neural network model; The feature vectors corresponding to some devices in the historical feature matrix are data masked to obtain a half-masked feature matrix; The semi-masked feature matrix is ​​input into the neural network model to obtain the second historical output matrix output by the neural network model; and Based on the difference between the first historical output matrix and the second historical output matrix, the neural network model is optimized in reverse.

4. The method of claim 3, wherein, The step of data masking the feature vectors corresponding to a portion of the devices in the historical feature matrix to obtain a half-masked feature matrix includes at least one of the following: Replace the feature vectors corresponding to the aforementioned devices with zero vectors; or Replace the feature vectors corresponding to some of the devices with the feature vectors corresponding to other devices.

5. The method of claim 1, wherein, The method further includes: Based on the third classification result, the R output matrices are sorted according to their chronological order; and The R feature data sequences and the R feature matrices are sorted accordingly based on the order of the R output matrices.

6. The method of claim 1, wherein, The method further includes training the neural network model in the following manner: Q feature data sequences collected at Q historical moments are obtained to obtain Q historical feature data sequences for training, where Q is an integer greater than 1; The feature matrix obtained based on each of the historical feature data sequences is used as the historical feature matrix; The historical feature matrix is ​​input into the neural network model to obtain the historical output matrix output by the neural network model; Extract the vector corresponding to the start flag bit from each of the historical output matrices to serve as the historical flag bit vector; The extracted Q historical flag bit vectors are paired up and divided into two categories. In one category, the times corresponding to each pair of historical flag bit vectors have a chronological order, while in the other category, the times corresponding to each pair of historical flag bit vectors have a random relationship. Each pair of historical flag bit vectors is input into the temporal state classifier to obtain the third classification result output by the temporal state classifier; as well as Based on the difference between the third classification result output by the temporal state classifier and the category to which each pair of historical flag bit vectors belong, the neural network model and the temporal state classifier are jointly trained.

7. The method of any one of claims 1-6, wherein, The neural network model incorporates a self-attention mechanism.

8. The method of claim 7, wherein, The neural network model is a transformer model.

9. An apparatus for monitoring the status of a data center, comprising: The data acquisition module is used to acquire feature data of N devices in the data center at R time points to obtain R feature data sequences. In each feature data sequence, the starting position is identified by a start flag. The feature data of each device includes at least one of CPU features, memory features, and IO features. The start flag is used to distinguish feature data sequences at different time points. R and N are integers greater than or equal to 1. The data preprocessing module is used to obtain a feature matrix based on each of the feature data sequences, wherein the start flag in the feature data sequence and the feature data of each of the N devices are respectively converted into feature vectors in the feature matrix, including: processing the start flag into a vector associated with the acquisition time, and processing the feature data of each of the N devices into data of uniform length and uniform specification or format; The neural network model is trained using pre-training techniques; the neural network model is used to process the feature matrix and obtain an output matrix, wherein the vectors in the output matrix correspond one-to-one with the feature vectors in the feature matrix; and The monitoring module is used to obtain the status information of the data center based on the vector corresponding to the start flag in the output matrix. The status information includes the operation status information and time status information of the data center. The monitoring module is specifically used for: The vectors corresponding to the start flag in the output matrix are classified using a cluster status classifier to obtain a first classification result, wherein the first classification result is information indicating whether the data center has abnormal operating status; and When R is greater than 1, the vectors corresponding to the start flag in each pair of output matrices are input into the temporal state classifier, and the third classification result output by the temporal state classifier is obtained. The third classification result is used to indicate whether the times corresponding to the two pairs of input vectors have a sequential time state information.

10. An electronic device, comprising: One or more processors; Storage device for storing one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors perform the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having stored thereon executable instructions that, when executed by a processor, cause the processor to perform the method according to any one of claims 1 to 8.

12. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 8.

Citation Information

Patent Citations

  • Micro-grid state online monitoring method and device, computer equipment and storage medium

    CN113595246A