A method, storage medium, and equipment for intelligent data sensing in computer rooms using a star-based large-scale model for knowledge distillation.

Abstract

This invention discloses a method, storage medium, and device for intelligent sensing of data center data using a star-shaped big data model for knowledge distillation. The method includes: collecting historical data from data center log files under various abnormal scenarios; constructing a random forest model for extracting key features from the data center data; using the star-shaped big data model as the teacher model and the random forest model as the student model; inputting the historical data into the teacher and student models respectively; guiding the student model training based on the knowledge distillation loss between the teacher and student models until the knowledge distillation loss function converges, thus completing the training of the random forest model; collecting data center data in real time and inputting it into the trained random forest model to extract key features; and inputting the extracted key features into the fanotify component for anomaly identification of the data center data. This invention uses a lightweight random forest model to achieve intelligent sensing of data center data, thereby realizing anomaly identification.

Description

Technical Field

[0001] This invention relates to the field of data anomaly identification technology in data centers, specifically to a data center data intelligent sensing method, storage medium, and device that utilizes a star-scale model for knowledge distillation. Background Technology

[0002] Data centers host a company's core business, such as financial transactions, online services, and cloud computing. Any anomalies in the data center, such as server downtime or network latency, can lead to business failures. By monitoring data center data in real time, performing intelligent analysis, and implementing automated responses, anomalies can be detected in advance, significantly reducing the risk of failure, improving service quality, and ensuring data security.

[0003] On the one hand, data in data centers is complex in dimensions, and real abnormal data points may be masked by noise effects from multiple unrelated dimensions, making it difficult for existing data center data sensing technologies to capture anomalies from high-dimensional data. On the other hand, in order to improve the accuracy of data center data sensing, most current technologies use complex structural models such as deep neural networks and ensemble machine learning models for data center data sensing. This requires the use of high-performance GPUs or TPUs for training and inference, which significantly increases hardware investment. At the same time, complex structural models require longer training and inference times, which cannot meet the requirements of real-time monitoring of data centers. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a method, storage medium, and device for intelligent sensing of data center data using a star-shaped large model for knowledge distillation. The random forest model for data center data sensing is optimized by using knowledge distillation, which greatly reduces inference latency. At the same time, the optimized random forest model can extract key features from high-dimensional data center data for data center anomaly identification, which greatly improves the accuracy of anomaly identification.

[0005] To achieve the above technical objectives, the present invention adopts the following technical solution: a data center intelligent sensing method using a star-shaped large model for knowledge distillation, characterized by comprising the following steps: Step S1: Collect historical data on various abnormal scenarios from the data center's log files; Step S2: Construct a random forest model for extracting key features from data center data; Step S3: Use the Starry Sky model as the teacher model and the Random Forest model as the student model. Input the historical computer lab data into the teacher model and the student model respectively. Guide the training of the student model based on the knowledge distillation loss between the teacher model and the student model until the knowledge distillation loss function converges, thus completing the training of the Random Forest model. Step S4: Collect data from the computer room in real time and input it into the trained random forest model to extract key features; Step S5: Input the extracted key features into the fanotify component for anomaly identification of data center data.

[0006] Furthermore, step S3 includes the following sub-steps: Step S3.1: Input historical data from the computer room into the Starry Sky Big Data Model to extract key features; Step S3.2: Use historical data from the computer room as input to the random forest model, use key features extracted from the Starry Sky Big Data Model as soft labels for the random forest model, use manually labeled key features from historical data as true labels for the random forest model, train the random forest model, and calculate the knowledge distillation loss function. Step S3.3: Adjust the hyperparameters of the random forest model, determine the pruning path of each subtree in the random forest model according to the pruning reward, and update the random forest model; Step S3.4: Repeat steps S3.2-S3.3 for the updated random forest model until the knowledge distillation loss function converges, thus completing the training of the random forest model.

[0007] Furthermore, the knowledge distillation loss function The calculation process is as follows:

[0008] in, This represents the error loss between the key features extracted by the random forest model and the true labels. , This indicates the number of samples used in the training process. i express N index, J This indicates the number of key features extracted from each sample. j express J index, Indicates the first i The first sample extracted j One key feature Indicates the first i In the nth sample j A real label, express Weighting coefficients; This represents the error loss between the key features extracted by the random forest model and the soft labels. , Indicates the first i In the nth sample j A soft label, express Weighting coefficients; This represents the error loss of intermediate features between the Random Forest model and the Starry Sky model. , Indicates the first i Intermediate features of a sample in a random forest model Indicates the first i Features of each sample in the star-shaped large model express Weighting coefficients; .

[0009] Furthermore, the process of determining the pruning path for each subtree in the random forest model based on pruning costs is as follows: i. For nodes in the random forest model, calculate the pruning benefit of cutting off the corresponding subtree; ii. Find the internal node with the minimum pruning benefit from cutting its corresponding subtree from all internal nodes in the random forest model, subtract the corresponding subtree, and obtain the new tree; iii. Repeat steps i-ii until only the root node remains in all trees.

[0010] Furthermore, the pruning benefit of cutting off the corresponding subtree The calculation process is as follows:

[0011] in, Represented by node t The subtree rooted in the tree Subtree The number of leaf nodes on Indicates t The node represents the error loss between the key features extracted from the leaf nodes and the true labels. Represented by node t subtree with root Error loss between extracted key features and true labels Indicates t The error loss between the key features extracted from leaf nodes and the soft labels is calculated. Represented by node t subtree with root Error loss between extracted key features and soft labels.

[0012] Furthermore, the specific process of step S4 is as follows: input the real-time collected data from the data center into the trained random forest model, calculate the score of the data center data in each dimension, and extract the data center data with the top m scores as key features.

[0013] Furthermore, the scoring of data center data in each dimension. The calculation process is as follows:

[0014] in, This represents the number of subtrees in the trained random forest model. h express index, This indicates that the data from a certain data center is in the [number]th [period]. h Scores in subtrees , Indicates the first h The number of nodes in each subtree Indicates the first s Nodes The amount of importance has decreased.

[0015] Further, the specific process of step S5 is as follows: configure the rules for identifying data anomalies in the data center in the fanotify component, match the extracted key features with the rules for identifying data anomalies in the data center, and if the matching degree exceeds the set threshold, determine that there is anomaly in the data center data; otherwise, the data center data is running normally.

[0016] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program that enables a computer to execute the aforementioned intelligent sensing method for data center data using a star-scale model for knowledge distillation.

[0017] Furthermore, the present invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the aforementioned intelligent sensing method for data center data using the star-scale model for knowledge distillation.

[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) The intelligent perception method for computer room data using knowledge distillation of the Starry Sky Model in this invention uses the Starry Sky Model as the teacher model. The Starry Sky Model has cross-domain knowledge and can quickly and accurately capture the key features in the computer room data. Through knowledge distillation, the random forest model can learn the generalized knowledge of the Starry Sky Model in different computer room scenarios. This makes the optimized random forest model better maintain performance stability when facing new computer room environments and reduce the risk of overfitting. In addition, the random forest model after knowledge distillation has fewer parameters and lower computational complexity, and can run quickly under limited computing resources to meet the needs of real-time perception of computer room data. (2) The present invention calculates the pruning benefit of removing subtrees in the random forest model based on the knowledge distillation loss function, which can more accurately evaluate the contribution of each subtree to the performance of the random forest model, and only remove those parts that have a small impact on the knowledge distillation loss, thereby achieving more precise pruning while ensuring the performance of the random forest model. This invention utilizes an optimized random forest model to extract key features from high-dimensional data center data for data center anomaly identification, which greatly improves the accuracy of anomaly identification and reduces inference latency. Attached Figure Description

[0019] Figure 1 This is a flowchart of the intelligent perception method for data center data using the star-scale model for knowledge distillation, as described in this invention. Figure 2 This is a flowchart of knowledge distillation in this invention. Detailed Implementation

[0020] The technical solution of the present invention will be further explained and described below with reference to the accompanying drawings.

[0021] like Figure 1 This is a flowchart of the intelligent perception method for data center data using the star-scale model for knowledge distillation, as described in this invention. The anomaly identification method includes the following steps: Step S1: Collect historical data from the data center log files under various anomaly scenarios. These anomaly scenarios include: hardware failures caused by server crashes, network equipment malfunctions, storage system problems, or power distribution failures; environmental anomalies caused by uncontrolled temperature and humidity, data center leaks, or abnormal air quality; operational errors caused by configuration errors, physical damage, or abuse of privileges; external threats caused by power outages, network attacks, or natural disasters; and software anomalies caused by operating system crashes, application server failures, or log system anomalies. Therefore, for various anomaly scenarios, the data center data to be collected includes: hardware equipment data, environmental monitoring data, network performance data, security log data, and business continuity data. Through intelligent perception of this data center data, anomalies in the data center can be identified.

[0022] Step S2: Construct a random forest model for extracting key features from data center data. Since data center data often contains a large number of features and complex patterns, the original random forest model cannot learn fully due to structural limitations, resulting in a large error in the extraction of key features from the data center.

[0023] Step S3: Therefore, in this invention, the Starry Sky Model is used as the teacher model, and the Random Forest Model is used as the student model. Historical data from the computer room is input into the teacher model and the student model respectively. The student model is trained based on the knowledge distillation loss between the teacher model and the student model until the knowledge distillation loss function converges, thus completing the training of the Random Forest Model. Utilizing the cross-domain knowledge of the Starry Sky Model, it can quickly and accurately capture key features in the computer room data. Through knowledge distillation, the Random Forest Model can learn the generalization knowledge of the Starry Sky Model in different computer room scenarios. This allows the optimized Random Forest Model to better maintain performance stability when facing new computer room environments, reducing the risk of overfitting. Furthermore, due to the reduced number of parameters and lower computational complexity, the knowledge-distilled Random Forest Model can run quickly with limited computing resources, meeting the needs of real-time perception of computer room data. Figure 2 This includes the following sub-steps: Step S3.1: Input historical data from the computer room into the Starry Sky Big Data Model to extract key features; Step S3.2: Use historical data from the computer room as input to the random forest model, use key features extracted from the Starry Sky Big Data Model as soft labels for the random forest model, use manually labeled key features from historical data as true labels for the random forest model, train the random forest model, and calculate the knowledge distillation loss function. The knowledge distillation function in this invention includes an error loss between the key features extracted by the random forest model and the true labels. Error loss between key features extracted by the random forest model and soft labels And the error loss of intermediate features between the random forest model and the star model. ,in, Used to constrain the consistency between the output of the random forest model and the true label, ensuring that the random forest model learns the basic pattern; This allows the random forest model to learn the knowledge imparted by the star model, helping it capture more nuanced decision-making logic and gradually approximate the output distribution of the star model, thereby improving the sensitivity of data center anomaly detection. This is used to align the intermediate layer features of the Random Forest model and the Starry Sky Model, forcing the Random Forest model to learn the higher-order abstract representation of the Starry Sky Model. The knowledge distillation loss function in this invention... The calculation process is as follows:

[0024] in, , This indicates the number of samples used in the training process. i express N index, J This indicates the number of key features extracted from each sample.j express J index, Indicates the first i The first sample extracted j One key feature Indicates the first i In the nth sample j A real label, express Weighting coefficients; , Indicates the first i In the nth sample j A soft label, express Weighting coefficients; , Indicates the first i Intermediate features of a sample in a random forest model Indicates the first i Features of each sample in the star-shaped large model express Weighting coefficients; .

[0025] Step S3.3: Adjust the hyperparameters of the random forest model and determine the pruning path for each subtree in the random forest model according to the pruning reward, then update the random forest model. During the pruning process of the random forest model, using knowledge distillation loss as a guide ensures that the pruned random forest model still retains the key knowledge learned from the large-scale model. This is crucial for data center data perception tasks, as the key knowledge may involve key features for identifying data center anomalies, thereby improving pruning efficiency.

[0026] In this invention, the process of determining the pruning path of each subtree in the random forest model based on pruning cost is as follows: i. For each node in the random forest model, calculate the pruning benefit of removing the corresponding subtree. :

[0027] in, Represented by node t The subtree rooted in the tree Subtree The number of leaf nodes on Indicates t The node represents the error loss between the key features extracted from the leaf nodes and the true labels. Represented by node t subtree with root Error loss between extracted key features and true labels Indicates t The error loss between the key features extracted from leaf nodes and the soft labels is calculated. Represented by node t subtree with root Error loss between extracted key features and soft labels.

[0028] ii. Find the internal node with the minimum pruning benefit from cutting off the corresponding subtree from all internal nodes in the random forest model, subtract the corresponding subtree, and obtain a new tree. By using the pruning benefit, we can more accurately evaluate the contribution of each subtree to the performance of the random forest model, and only prune those parts that have a small impact on the knowledge distillation loss, so as to achieve more precise pruning while ensuring the performance of the random forest model. iii. Repeat steps i-ii until only the root node remains in all trees.

[0029] Step S3.4: Repeat steps S3.2-S3.3 for the updated random forest model until the knowledge distillation loss function converges, thus completing the training of the random forest model.

[0030] Step S4: Real-time data collection of the data center is input into the trained random forest model to extract key features. This allows for the extraction of key features from high-dimensional data center data for anomaly identification, significantly improving anomaly identification accuracy and reducing inference latency. Specifically, the real-time data center data is input into the trained random forest model to calculate the score of the data center data in each dimension, and the top m data center data in terms of score are extracted as key features.

[0031] Scoring of each dimension of data center data in this invention The calculation process is as follows:

[0032] in, This represents the number of subtrees in the trained random forest model. h express index, This indicates that the data from a certain data center is in the [number]th [period]. h Scores in subtrees , Indicates the first h The number of nodes in each subtree Indicates the first s Nodes The amount of importance has decreased.

[0033] Step S5: Input the extracted key features into the fanotify component for anomaly detection in the data center. The fanotify component runs directly in kernel space, eliminating the need for frequent context switching and avoiding impact on the performance of various systems in the data center. Furthermore, utilizing the file system monitoring mechanism of the fanotify component, it can capture file access, modification, deletion, and other operations in real time. Specifically, configure the rules for anomaly detection in the fanotify component, and match the extracted key features with these rules. If the matching degree exceeds a set threshold, the data center data is determined to be abnormal; otherwise, the data center data operates normally.

[0034] This invention utilizes a knowledge distillation method for intelligent perception of data center data using a star-shaped large model. By constructing a lightweight random forest model through knowledge distillation, it extracts key features from high-dimensional data center data, significantly improving the accuracy of key feature extraction and reducing latency. Furthermore, it uses the fanotify component to identify abnormal data center data from key features, reducing overhead and false alarm rate, and providing reliable support for early warning of data center anomalies. This allows for proactive measures to mitigate the risk of data center anomalies.

[0035] In one technical solution of the present invention, a computer-readable storage medium is also provided, storing a computer program that enables a computer to execute the aforementioned method for intelligent perception and anomaly identification of computer room data using a star-scale model for knowledge distillation.

[0036] In one technical solution of the present invention, an electronic device is also provided, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described method for intelligent perception and anomaly identification of computer room data using the star-scale model for knowledge distillation.

[0037] In the embodiments disclosed in this application, a computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0038] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this application can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software 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 beyond the scope of this application.

[0039] The above are merely preferred embodiments of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent sensing of data in a computer room using a star-based large-scale model for knowledge distillation, characterized in that, Includes the following steps: Step S1: Collect historical data on various abnormal scenarios from the data center's log files; Step S2: Construct a random forest model for extracting key features from data center data; Step S3: Use the Starry Sky model as the teacher model and the Random Forest model as the student model. Input historical computer lab data into the teacher and student models respectively. Guide the student model training based on the knowledge distillation loss between the teacher and student models until the knowledge distillation loss function converges, completing the training of the Random Forest model; including the following sub-steps: Step S3.1: Input historical data from the computer room into the Starry Sky Big Data Model to extract key features; Step S3.2: Use historical data from the computer room as input to the random forest model, use key features extracted from the Starry Sky Big Data Model as soft labels for the random forest model, use manually labeled key features from historical data as true labels for the random forest model, train the random forest model, and calculate the knowledge distillation loss function. Step S3.3: Adjust the hyperparameters of the random forest model, determine the pruning path of each subtree in the random forest model according to the pruning reward, and update the random forest model; Step S3.4: Repeat steps S3.2-S3.3 for the updated random forest model until the knowledge distillation loss function converges, thus completing the training of the random forest model; Knowledge distillation loss function The calculation process is as follows: in, This represents the error loss between the key features extracted by the random forest model and the true labels. , This indicates the number of samples used in the training process. i express N index, J This indicates the number of key features extracted from each sample. j express J index, Indicates the first i The first sample extracted j One key feature Indicates the first i In the nth sample j A real label, express Weighting coefficients; This represents the error loss between the key features extracted by the random forest model and the soft labels. , Indicates the first i In the nth sample j A soft label, express Weighting coefficients; This represents the error loss of intermediate features between the Random Forest model and the Starry Sky model. , Indicates the first i Intermediate features of a sample in a random forest model Indicates the first i Features of each sample in the star-shaped large model express Weighting coefficients; ; Step S4: Collect data from the computer room in real time and input it into the trained random forest model to extract key features; Step S5: Input the extracted key features into the fanotify component for anomaly identification of data center data.

2. The intelligent sensing method for data center using a star-based large-scale model for knowledge distillation as described in claim 1, characterized in that, The process of determining the pruning path for each subtree in the random forest model based on pruning cost is as follows: i. For nodes in the random forest model, calculate the pruning benefit of cutting off the corresponding subtree; ii. Find the internal node with the minimum pruning benefit from cutting its corresponding subtree from all internal nodes in the random forest model, subtract the corresponding subtree, and obtain the new tree; iii. Repeat steps i-ii until only the root node remains in all trees.

3. The intelligent sensing method for data center using a star-based large-scale model for knowledge distillation as described in claim 2, characterized in that... The pruning benefits of cutting off the corresponding subtree The calculation process is as follows: in, Represented by node t The subtree rooted in the tree Subtree The number of leaf nodes on Indicated by t The node represents the error loss between the key features extracted from the leaf nodes and the true labels. Represented by node t subtree with root Error loss between extracted key features and true labels Indicated by t The error loss between the key features extracted from leaf nodes and the soft labels is calculated. Represented by node t subtree with root Error loss between extracted key features and soft labels.

4. The intelligent sensing method for data center using a star-based large-scale model for knowledge distillation as described in claim 3, characterized in that, The specific process of step S4 is as follows: input the real-time collected data from the data center into the trained random forest model, calculate the score of the data center data in each dimension, and extract the data center data with the top m scores as key features.

5. The intelligent sensing method for data center using a star-based large-scale model for knowledge distillation as described in claim 4, characterized in that... Scoring of data center data in each dimension The calculation process is as follows: in, This represents the number of subtrees in the trained random forest model. h express index, This indicates that the data from a certain data center is in the [number]th [period]. h Scores in subtrees , Indicates the first h The number of nodes in each subtree Indicates the first s Nodes The amount of importance has decreased.

6. The intelligent sensing method for data center using a star-based large-scale model for knowledge distillation as described in claim 1, characterized in that, The specific process of step S5 is as follows: Configure the rules for identifying abnormal data in the data center in the fanotify component, match the extracted key features with the rules for identifying abnormal data in the data center, and if the matching degree exceeds the set threshold, it is determined that there is an abnormality in the data center data. Otherwise, the data in the computer room will operate normally.

7. A computer-readable storage medium storing a computer program, characterized in that, The computer program causes the computer to execute the intelligent perception method for data center using the star-scale model for knowledge distillation as described in any one of claims 1-6.

8. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the intelligent sensing method for data center data using the star-shaped large model for knowledge distillation as described in any one of claims 1-6.

Images