Equipment information monitoring method and device

Abstract

The invention discloses an equipment information monitoring method and device. The method comprises the following steps: acquiring equipment information; the collected equipment information is preprocessed, and the preprocessing comprises the steps that the collected equipment information is compared with a set reference value to obtain the running state of the equipment, and historical flow data in the equipment information is learned through an adaptive weight time sequence analysis model to obtain the running state of the equipment; dynamically adjusting the weight of each time sequence according to the prediction error and learning result of the historical flow data so as to predict the future operation state of the equipment; and automatically uploading the preprocessed equipment information to a cloud server, and monitoring the equipment information through the cloud server. According to the invention, efficient, accurate and real-time management of the data center switch can be realized, and the stability and efficiency of the data center are improved.

Description

Technical Field

[0001] The present invention relates to the field of network communication technology, and in particular to a device information monitoring method and apparatus. Background Art

[0002] In today's network communications landscape, especially in data center switch management, device self-diagnosis is a crucial step. Traditional self-diagnosis methods rely heavily on manual operations, requiring technicians to manually collect and compare switch configuration information, performance data, and alarm information in log files. This approach is not only inefficient but also prone to human error, failing to meet the high real-time and accuracy requirements of modern data centers.

[0003] Furthermore, existing self-checking systems often lack the ability to effectively manage and analyze historical data. Most systems only provide simple feedback on current status, without the ability to deeply mine and analyze historical data. This makes comprehensive information traceability and predictive maintenance difficult. This limitation slows data centers' response to sudden failures or performance degradation, increasing the complexity and cost of operations and maintenance.

[0004] Therefore, how to achieve efficient, accurate and real-time management of equipment is a technical problem that needs to be solved urgently. Summary of the Invention

[0005] The main purpose of the present invention is to provide a device information monitoring method and apparatus, which can efficiently, accurately and in real time manage data center switches, thereby improving the stability and efficiency of the data center.

[0006] In a first aspect, the present application provides a device information monitoring method, wherein the method comprises the steps of: collecting device information;

[0007] Preprocessing the collected device information, wherein the preprocessing includes: comparing the collected device information with a set reference value to obtain the operating status of the device, and using an adaptive weighted time series analysis model to learn historical traffic data in the device information, and dynamically adjusting the weight of each time series based on the prediction error of the historical traffic data and the learning results to predict the future operating status of the device;

[0008] The pre-processed device information is automatically uploaded to the cloud server, and the device information is monitored by the cloud server.

[0009] In combination with the first aspect above, as an optional implementation manner, based on the acquired performance data of the device, historical network traffic data of the device is acquired from the performance data;

[0010] The adaptive weighted time series analysis model arranged on the device is used to process the time series characteristics of the historical traffic data through a sliding time window to capture the multiple time scale characteristics of the traffic data.

[0011] In combination with the first aspect above, as an optional implementation, the historical data is preprocessed, and the preprocessing includes: data cleaning, removing outliers, and filling missing values;

[0012] Selecting a time window length and dividing the pre-processed historical traffic data into a plurality of consecutive time windows in chronological order;

[0013] The data within each time window is processed and analyzed, and the data processing results within each time window are output or visualized.

[0014] In combination with the first aspect above, as an optional implementation method, the adaptive weighted time series analysis model is used to output the prediction error and learning results of the time series characteristics of the historical traffic data;

[0015] Based on the dynamic weight adjustment mechanism introduced in the adaptive weight time series analysis model, the weights of each time series are dynamically adjusted through the prediction error and learning results output by the adaptive weight time series analysis model, so that the adaptive weight time series analysis model can adapt to different traffic change patterns.

[0016] In combination with the first aspect above, as an optional implementation method, a corresponding comparison reference frame is set according to the historical data and actual operation status of the switch, and the collected device information is compared with the set reference value;

[0017] If the comparison result shows that the device information is not within the set reference value range, it is determined that the device operation status is abnormal; otherwise, it is determined that the device operation status is normal.

[0018] In combination with the first aspect above, as an optional implementation, if it is determined that the operating state of the device is abnormal, the fault data and the device status information are pre-processed to extract abnormal features;

[0019] Based on abnormal characteristics and change trends, classification algorithms are used to classify faults, and corresponding processing strategies are formulated based on the classification results.

[0020] In conjunction with the first aspect above, as an optional implementation, the configuration information of the switch is collected through multiple methods, wherein the multiple methods include a web crawler and a device API interface;

[0021] Collect switch performance data through network traffic analysis tools;

[0022] Collect alarm information by analyzing the switch's log files or receiving the device's alarm signals.

[0023] In combination with the first aspect above, as an optional implementation method, the processed device information is organized into a unified data format and automatically uploaded to the cloud server by periodic batch uploading or real-time synchronization of data streams;

[0024] After receiving the device information, the cloud server analyzes and manages the historical data in the device information through cloud storage and computing resources to monitor the device information.

[0025] In combination with the first aspect above, as an optional implementation method, the uploaded data is encrypted, where the encryption process includes: a symmetric encryption algorithm and an asymmetric encryption algorithm.

[0026] In a second aspect, the present application provides a device information monitoring apparatus, the apparatus comprising:

[0027] A collection module, which is used to collect device information;

[0028] a data processing module for preprocessing the collected device information, wherein the preprocessing includes: comparing the collected device information with a set reference value to obtain the operating status of the device; using an adaptive weighted time series analysis model to learn the historical flow data in the device information; and dynamically adjusting the weight of each time series based on the prediction error of the historical flow data and the learning results to predict the future operating status of the device;

[0029] The cloud upload module is used to automatically upload the pre-processed device information to the cloud server and monitor the device information through the cloud server.

[0030] The present application provides a device information monitoring method and apparatus, wherein the method includes the steps of: collecting device information; preprocessing the collected device information, wherein the preprocessing includes: comparing the collected device information with a set reference value to obtain the operating status of the device; using an adaptive weight time series analysis model to learn historical traffic data in the device information, and dynamically adjusting the weights of each time series based on the prediction error of the historical traffic data and the learning results to predict the future operating status of the device; automatically uploading the preprocessed device information to a cloud server, and monitoring the device information through the cloud server. The present application can efficiently, accurately, and in real time manage data center switches, thereby improving the stability and efficiency of the data center.

[0031] It should be understood that the foregoing general description and the following detailed description are exemplary only and are not restrictive of the invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0033] Figure 1 A flow chart of a device information monitoring method provided in an embodiment of the present application;

[0034] Figure 2 A schematic diagram of a device information monitoring apparatus provided in an embodiment of the present application;

[0035] Figure 3 A schematic diagram showing the future operating state of a device predicted in an embodiment of the present application;

[0036] Figure 4 A flowchart for predicting the future operating status of a device provided in an embodiment of the present application;

[0037] Figure 5 This is a schematic diagram of the abnormal data processing principle provided in the embodiments of this application. DETAILED DESCRIPTION

[0038] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0039] Furthermore, the drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Some of the blocks shown in the drawings are functional entities that do not necessarily correspond to physically or logically separate entities.

[0040] The embodiments of the present application are further described in detail below with reference to the accompanying drawings.

[0041] Reference Figure 1 , Figure 1 The figure shows a flow chart of a device information monitoring method provided by the present invention. Figure 1 As shown, the method includes the steps of:

[0042] Step S101: Collect device information.

[0043] Specifically, device information includes configuration information, performance data, and alarm information. Switch configuration information is collected through various methods, including web crawlers and device APIs.

[0044] Collect switch performance data through network traffic analysis tools;

[0045] Collect alarm information by analyzing the switch's log files or receiving the device's alarm signals.

[0046] For easier understanding, let's use an example to obtain switch configuration information using a web crawler or device API. 1) A web crawler automatically obtains switch configuration information, such as the device model and port configuration, by simulating a user accessing the switch's webpage or API. 2) The device API provides a programming interface for interacting with the switch. You can obtain switch configuration information by calling API functions.

[0047] Use network traffic analysis tools to obtain switch performance data (understand the traffic forwarding status on the switch at different time periods, whether there are bursts of traffic, etc., and combine the analyzed alarm information to understand the switch's operating status). Network traffic analysis tools can monitor network traffic in real time and obtain switch network traffic data, such as network traffic volume and packet loss rate.

[0048] Parse the switch's log files to obtain alarm information. Switches typically record alarm events that occur during operation and write them to log files. The data collection module can read the log files, parse the alarm information, and obtain detailed information about the switch alarm events.

[0049] In one embodiment, collection rules are set: Based on the type of switch information to be collected (e.g., configuration information, performance data, alarm information, etc.), corresponding collection rules are set. For example, configuration information can be obtained by reading the switch configuration file or using the device API; performance data can be obtained through network traffic analysis tools; and alarm information can be obtained by parsing the switch log file or receiving device alarm signals.

[0050] Step S102: Preprocessing the collected device information, wherein the preprocessing includes: comparing the collected device information with the set reference value to obtain the operating status of the device, and using the adaptive weight time series analysis model to learn the historical flow data in the device information, and dynamically adjusting the weight of each time series based on the prediction error and learning results of the historical flow data to predict the future operating status of the device.

[0051] Specifically, the collected device information is compared with the set reference value to obtain the operating status of the device. Specifically, according to the historical data and actual operating conditions of the switch, a corresponding comparison reference frame is set, and the collected device information is compared with the set reference value;

[0052] If the comparison result shows that the device information is not within the set reference value range, it is determined that the device operation status is abnormal; otherwise, it is determined that the device operation status is normal.

[0053] To illustrate, first, the collected switch information is compared with a predetermined reference value. This reference value can be an average value, a threshold, or a standard value calculated based on historical data. (For example, for traffic data, the actual utilization of each port can be calculated based on the actual network traffic model, with a 5% margin of error as the upper and lower thresholds.)

[0054] The switch's operating status is then determined based on the comparison results. If the comparison results are within the reference range, the switch is operating normally. If they are outside the reference range, an abnormality exists and requires action (e.g., automatic repair, early warning notification, etc., based on the corresponding fault information processing strategy).

[0055] In one embodiment, if it is determined that the equipment operating state is abnormal, the fault data and equipment state information are pre-processed to extract abnormal features;

[0056] Based on abnormal characteristics and change trends, classification algorithms are used to classify faults, and corresponding processing strategies are formulated based on the classification results.

[0057] It should be noted that obtaining the operating status of the device can also predict the future operating status of the switch, such as predicting the network traffic trend and fault probability in the future.

[0058] Specifically: The adaptive weighted time series analysis model is used to learn the historical traffic data in the device information. Based on the prediction error of the historical traffic data and the learning results, the weight of each time series is dynamically adjusted to predict the future operating status of the device, including:

[0059] Based on the performance data of the device obtained, historical network traffic data of the device is obtained from the performance data; using an adaptive weight time series analysis model arranged on the device, the time series characteristics of the historical traffic data are processed through a sliding time window to capture the multiple time scale characteristics of the traffic data.

[0060] The time series characteristics of the historical traffic data are processed by a sliding time window, including: preprocessing the historical data, wherein the preprocessing includes: data cleaning, removing outliers and filling missing values;

[0061] Selecting a time window length and dividing the pre-processed historical traffic data into a plurality of consecutive time windows in chronological order;

[0062] The data within each time window is processed and analyzed, and the data processing results within each time window are output or visualized.

[0063] Dynamically adjusting the weights of each time series based on the prediction error and learning results of the historical traffic data to predict the future operating status of the device, including: using the adaptive weight time series analysis model to output the prediction error and learning results of the time series characteristics of the historical traffic data;

[0064] Based on the dynamic weight adjustment mechanism introduced in the adaptive weight time series analysis model, the weights of each time series are dynamically adjusted through the prediction error and learning results output by the adaptive weight time series analysis model, so that the adaptive weight time series analysis model can adapt to different traffic change patterns.

[0065] To facilitate understanding, we propose a novel adaptive weighted time series analysis model, deployed on devices (switches). This model automatically learns the time series characteristics of historical traffic data and adaptively adjusts the weights of each time series based on these characteristics, enabling it to more accurately capture the complex patterns of traffic changes. Specifically, a sliding time window technique is used to process historical traffic data to capture the multi-timescale characteristics of traffic data. This sliding time window allows for the learning of traffic data at different time scales, leading to a more comprehensive understanding of traffic change patterns.

[0066] The data processing flow of the sliding time window technology can be summarized as the following steps:

[0067] Data preparation: First, you need to prepare the data for time series analysis. This data can be historical traffic data, time series data, or other data that requires time window processing.

[0068] Select the time window length: Based on the characteristics of the data and the purpose of the analysis, select an appropriate time window length. This length can be fixed or dynamically adjusted based on data changes.

[0069] Data preprocessing: Preprocess the prepared data, including data cleaning, removing outliers, filling missing values, etc., to ensure the quality and reliability of the data.

[0070] Sliding time window: Divide the data into time windows according to the time sequence. The data in each window can be analyzed and processed separately. The length of the window should be equal to or slightly larger than the length of the periodicity or trend characteristics of the time series data.

[0071] Data processing and analysis: Process and analyze the data within each time window. This can include calculating statistical indicators, model fitting, and forecasting.

[0072] Result output: Output or visualize the data processing results within each time window to facilitate further analysis and decision-making.

[0073] A dynamic weight adjustment mechanism is introduced, which dynamically adjusts the weight of each time series based on the prediction error and learning progress. This enables the model to better adapt to different traffic change patterns and improve the robustness and accuracy of the prediction.

[0074] Specifically, a weight adjustment function f(e,p) is defined, where e represents the prediction error and p represents the learning progress. This function calculates and updates the weight of each time series based on the real-time situation of the prediction error and learning progress. The weight adjustment function f(e,p) accepts two parameters: the prediction error e and the learning progress p. This function calculates and updates the weight of each time series based on the real-time situation of the prediction error and learning progress (see Figure 3 ).

[0075] Step S103: Automatically upload the pre-processed device information to the cloud server, and monitor the device information through the cloud server.

[0076] Specifically, the processed device information is organized into a unified data format and automatically uploaded to the cloud server through regular batch uploading or real-time synchronization of data streams;

[0077] After receiving the device information, the cloud server analyzes and manages the historical data in the device information through cloud storage and computing resources to monitor the device information.

[0078] In one embodiment, the uploaded data is encrypted, wherein the encryption includes: a symmetric encryption algorithm and an asymmetric encryption algorithm.

[0079] For easier understanding, let's take an example: the cloud upload module automatically uploads collected switch information to a cloud server. Leveraging cloud storage and computing resources, historical data can be analyzed and managed, providing comprehensive information traceability and predictive maintenance. To ensure data integrity and security, uploaded data is encrypted.

[0080] The implementation principle of the cloud upload module is as follows:

[0081] First, the collected switch information is organized into a unified data format.

[0082] The data is then uploaded to the cloud server via the Internet. The upload can be achieved through periodic batch uploads or real-time synchronous data streaming (wherein the upload can be achieved through various protocols, such as HTTP, FTP, etc.).

[0083] After receiving the data, the cloud server stores and manages it. Leveraging its powerful storage and computing resources, historical data can be analyzed and managed. For example, big data analysis techniques can be used to deeply mine switch operating data to predict future operating trends and potential failures. Furthermore, the process and results of exception handling can be tracked and analyzed to continuously optimize the system's exception handling.

[0084] At the same time, to ensure data integrity and security, the uploaded data needs to be encrypted. Encryption can be implemented using symmetric or asymmetric encryption algorithms to ensure the security and confidentiality of data during transmission and storage.

[0085] When implementing this invention, it's necessary to first set various switch information collection rules and comparison standards to facilitate subsequent automated operations. Furthermore, a set of exception handling strategies must be established to ensure effective handling of various switch issues. Finally, through connection to a cloud server, automatic information upload and storage is achieved.

[0086] Reference Figure 2 , Figure 2 FIG. 1 is a schematic diagram of a device information monitoring device provided by the present invention, such as Figure 2 As shown, the device includes:

[0087] Collection module 201: used to collect device information.

[0088] Data processing module 202: It is used to pre-process the collected device information, where the pre-processing includes: comparing the collected device information with the set reference value to obtain the operating status of the device, and using the adaptive weight time series analysis model to learn the historical flow data in the device information, and dynamically adjusting the weight of each time series based on the prediction error and learning results of the historical flow data to predict the future operating status of the device.

[0089] The cloud upload module 203 is used to automatically upload the pre-processed device information to the cloud server and monitor the device information through the cloud server.

[0090] Furthermore, in a possible implementation manner, the data processing module is further configured to obtain historical network traffic data of the device from the performance data based on the acquired performance data of the device;

[0091] The adaptive weighted time series analysis model arranged on the device is used to process the time series characteristics of the historical traffic data through a sliding time window to capture the multiple time scale characteristics of the traffic data.

[0092] Furthermore, in a possible implementation, the data processing module is further configured to pre-process the historical data, wherein the pre-processing includes: data cleaning, removing outliers, and filling missing values;

[0093] Selecting a time window length and dividing the pre-processed historical traffic data into a plurality of consecutive time windows in chronological order;

[0094] The data within each time window is processed and analyzed, and the data processing results within each time window are output or visualized.

[0095] Furthermore, in a possible implementation manner, the data processing module is further configured to output a prediction error and a learning result of the time series characteristics of the historical traffic data using the adaptive weighted time series analysis model;

[0096] Based on the dynamic weight adjustment mechanism introduced in the adaptive weight time series analysis model, the weights of each time series are dynamically adjusted through the prediction error and learning results output by the adaptive weight time series analysis model, so that the adaptive weight time series analysis model can adapt to different traffic change patterns.

[0097] Furthermore, in a possible implementation, the data processing module is further configured to set a corresponding comparison reference frame based on historical data and actual operation of the switch, and compare the collected device information with the set reference value;

[0098] If the comparison result shows that the device information is not within the set reference value range, it is determined that the device operation status is abnormal; otherwise, it is determined that the device operation status is normal.

[0099] Furthermore, in a possible implementation, the data processing module is further configured to pre-process the fault data and the device status information to extract abnormal features if it is determined that the device operating state is abnormal;

[0100] Based on abnormal characteristics and change trends, classification algorithms are used to classify faults, and corresponding processing strategies are formulated based on the classification results.

[0101] Furthermore, in a possible implementation, the collection module is further configured to collect switch configuration information through a variety of methods, including web crawlers and device API interfaces;

[0102] Collect switch performance data through network traffic analysis tools;

[0103] Collect alarm information by analyzing the switch's log files or receiving the device's alarm signals.

[0104] Furthermore, in a possible implementation, the cloud upload module is further configured to organize the processed device information into a unified data format and automatically upload it to the cloud server by periodically uploading in batches or by real-time synchronization of data streams;

[0105] After receiving the device information, the cloud server analyzes and manages the historical data in the device information through cloud storage and computing resources to monitor the device information.

[0106] Furthermore, in a possible implementation, the cloud upload module is further configured to perform encryption processing on the uploaded data, wherein the encryption processing includes: a symmetric encryption algorithm and an asymmetric encryption algorithm.

[0107] Reference Figure 3 , Figure 3 The figure shows the principle diagram of the future operation state of the prediction device provided by the present invention, as shown in FIG. Figure 3 As shown:

[0108] In one embodiment, the adaptive weighted time series analysis model is used to predict future traffic trends and failure rates based on historical data. The following steps may be used:

[0109] (1) Data preprocessing: Clean and smooth historical data to extract features for subsequent analysis.

[0110] (2) Feature selection: Select appropriate features based on actual conditions, which may include device status information, network traffic data, fault data, etc.

[0111] (3) Model selection: An adaptive weighted time series analysis model is proposed to learn the preprocessed data (the model can better capture the patterns and trends of network traffic changes and improve the efficiency and accuracy of network management. Secondly, the model can predict future traffic changes and provide a basis for network planning and optimization. For example, based on the prediction results, network congestion or abnormal situations can be discovered in advance, and timely measures can be taken to intervene and adjust. Finally, the model also has certain robustness and generalization capabilities and can adapt to different network environments and traffic characteristics).

[0112] (4) Model training: Use 70%-80% of historical data to train the model.

[0113] (5) Model evaluation: Based on the learning results and prediction errors of historical traffic data, the weight of each time series is dynamically adjusted, and adjustments and optimizations are performed based on the evaluation results. The best parameter combination is selected as the model. Specifically, a weight adjustment function f(e,p) is defined, where e represents the prediction error and p represents the learning progress. This function calculates and updates the weight of each time series based on the real-time status of the prediction error and learning progress. The weight adjustment function f(e,p) accepts two parameters: the prediction error e and the learning progress p. This function calculates and updates the weight of each time series based on the real-time status of the prediction error and learning progress.

[0114] The specific implementation can be as follows:

[0115] ① When the prediction error e is small, increase the weight of the corresponding time series to reward the model for correct predictions. Specifically, a proportional function can be used to increase the weight, for example:

[0116] f(e,p)=a*e+b*p+c, where a is a proportional coefficient greater than 0, and b and c are appropriate constants.

[0117] ② When the prediction error e is large, reduce the weight of the corresponding time series to punish the model's incorrect predictions. Specifically, an inverse proportional function can be used to reduce the weight, for example:

[0118] f(e,p)=d / (e+ε)+b*p+c, where d is a proportional coefficient greater than 0, ε is a small positive number, and b and c are appropriate constants.

[0119] ③ When the learning progress p is high, for example, between 0 and 0.5, increase the weight of the corresponding time series to encourage the model to learn more historical information. Specifically, you can use an increasing function to increase the weight, for example:

[0120] f(e,p)=a*e+b*p^2+c, where a is a proportional coefficient greater than 0, and b and c are appropriate constants.

[0121] ④ When the learning progress p is low, reduce the weight of the corresponding time series to prevent the model from over-relying on historical information. Specifically, a decreasing function can be used to reduce the weight, for example:

[0122] f(e,p)=a*e+b / (p+ε)+c, where a is a proportional coefficient greater than 0, ε is a small positive number, and b and c are appropriate constants.

[0123] Through the above strategies, we can fully learn and accurately capture the multi-timescale characteristics of historical traffic data. At the same time, the specific form of the function f(e,p) can be adjusted and optimized according to actual needs.

[0124] To determine the constants a, b, and c, you need to experiment and tune them based on the actual situation and data characteristics. The following are examples of actual values used in this method.

[0125] When the prediction error e is small, the weight adjustment function is: f(e,p)=a*e+b*p+c.

[0126] The actual value used in this method is f(e,p)=0.2*e+0.3*p+0.01.

[0127] When the prediction error e is large, the weight adjustment function is: f(e,p)=d / (e+ε)+b*p+c.

[0128] The actual value used in this method is f(e,p)=2 / (e+0.01)+0.3*p+0.01.

[0129] When the learning progress p is high, the weight adjustment function is: f(e,p)=a*e+b*p^2+c.

[0130] The actual value used in this method is f(e,p)=0.3*e+0.5*p^2+0.01.

[0131] When the learning progress p is low, the weight adjustment function is: f(e,p)=a*e+b / (p+ε)+c.

[0132] The actual value used in this method is f(e,p)=0.3*e+2 / (p+0.01)+0.01.

[0133] (6) Predicting future traffic trends and failure probabilities: The trained model is used to predict future traffic trends and failure probabilities, mainly inferring trends and probabilities based on traffic changes and information changes.

[0134] Reference Figure 4 , Figure 4 The figure shows a flow chart of the future operation state of the prediction device provided by the present invention, as shown in FIG. Figure 4 As shown:

[0135] In one embodiment, ① a new adaptive weighted time series analysis model is proposed, which automatically learns the time series characteristics of historical traffic data and adaptively adjusts the weights of each time series according to these characteristics, so as to more accurately capture the complex patterns of traffic changes.

[0136] ② Use sliding time window technology to process historical traffic data to capture the multi-time scale characteristics of traffic data. Sliding time windows can be used to learn traffic data at different time scales, thereby more comprehensively understanding the patterns of traffic changes.

[0137] ③Introducing a dynamic weight adjustment mechanism, which dynamically adjusts the weights of each time series based on the prediction error and learning progress. This enables the model to better adapt to different traffic change patterns and improve the robustness and accuracy of the prediction.

[0138] ① The adaptive weighted time series analysis model is a time series analysis model that automatically learns the time series characteristics of historical traffic data and adaptively adjusts the weights of each time series based on these characteristics, thereby being able to more accurately capture the complex patterns of traffic changes.

[0139] First, the model employs a sliding time window technique, dividing historical traffic data into multiple time windows, each containing data from a specific time range. Then, using time series analysis methods such as the ARIMA model, feature extraction is performed on the data from each time window. This captures traffic data characteristics within different time windows, such as periodicity, trend, and seasonality. This sliding time window technique allows observation of traffic data at different time scales, leading to a more comprehensive understanding of traffic patterns.

[0140] Next, the model introduces a dynamic weight adjustment mechanism. This mechanism dynamically adjusts the weights of each time series based on prediction error and learning progress. Specifically, if the weight of a time series is too large, causing the model's prediction error to increase, the weight of that series can be appropriately reduced. Conversely, if the weight of a time series is too small, causing the model's prediction error to decrease, the weight of that series can be appropriately increased. By dynamically adjusting weights, the model can better adapt to different traffic patterns and improve prediction accuracy.

[0141] Finally, the model uses time series prediction models, such as linear regression or deep learning models like LSTM, to make predictions. These models can analyze the extracted time series features and predict future traffic changes based on these features.

[0142] Applying the new adaptive weighted time series analysis model to switch products offers numerous advantages. First, the model can better capture patterns and trends in network traffic changes, improving the efficiency and accuracy of network management. Second, the model can predict future traffic changes, providing a basis for network planning and optimization. For example, based on the prediction results, network congestion or anomalies can be detected in advance, allowing timely intervention and adjustment measures. Finally, the model is robust and generalizable, adapting to diverse network environments and traffic characteristics.

[0143] ② Sliding Time Window technology is a data preprocessing method commonly used in time series analysis. This technology selects a fixed-length time window and slides the time series data onto this window to capture the multi-time scale characteristics of traffic data.

[0144] First, the principle of sliding time window technology is to divide historical traffic data into multiple consecutive time windows, each containing data within a certain time range. For example, if there is an hour of traffic data, it can be divided into 10 time windows of 10 minutes each. The data within each window can be analyzed and processed separately.

[0145] The advantage of sliding time window technology is that it can capture traffic data characteristics at different time scales. By selecting time windows of different lengths, traffic data characteristics can be captured within different time ranges, such as hourly, daily, or weekly. Furthermore, sliding time window technology can reduce data dimensionality, lower model complexity, and improve computational efficiency.

[0146] Secondly, sliding time window technology can be applied to traffic data analysis for switch products. Traffic data in switch products often exhibits characteristics such as periodicity, trending, and seasonality. Using sliding time window technology, traffic data can be sliced into time windows of varying lengths to capture these characteristics. For example, when analyzing switch traffic data, sliding time window technology can be used to divide historical traffic data into hourly time windows to capture hourly characteristics of traffic data. Time windows can also be set to a day or a week in length to capture other timescale characteristics of traffic data.

[0147] Finally, sliding time window technology is an efficient, flexible, and widely used method for time series analysis. Its application in traffic data analysis of switch products can improve understanding and knowledge of traffic data characteristics. By selecting time windows of varying lengths, traffic data characteristics at different time scales can be captured, providing a basis for network management and optimization.

[0148] The data processing flow of the sliding time window technology can be summarized as the following steps:

[0149] Data preparation: First, you need to prepare the data for time series analysis. This data can be historical traffic data, time series data, or other data that requires time window processing.

[0150] Select the time window length: Based on the characteristics of the data and the purpose of the analysis, select an appropriate time window length. This length can be fixed or dynamically adjusted based on data changes.

[0151] Data preprocessing: Preprocess the prepared data, including data cleaning, removing outliers, filling missing values, etc., to ensure the quality and reliability of the data.

[0152] Sliding time window: Divide the data into time windows according to the time sequence. The data in each window can be analyzed and processed separately. The length of the window should be equal to or slightly larger than the length of the periodicity or trend characteristics of the time series data.

[0153] Data processing and analysis: Process and analyze the data within each time window. This can include calculating statistical indicators, model fitting, and forecasting.

[0154] Result output: Output or visualize the data processing results within each time window to facilitate further analysis and decision-making.

[0155] The advantage of sliding time windows is that they can divide data into time windows of varying lengths to capture traffic data characteristics at different time scales. Using sliding time windows, time series data can be divided into multiple subsequences, each of which can be analyzed and processed. This approach can reduce data dimensionality, lower model complexity, improve computational efficiency, and enable better understanding and analysis of data characteristics and trends.

[0156] ③ Introducing a dynamic weight adjustment mechanism (it should be noted that ② and ③ are specific implementations of ①). This mechanism dynamically adjusts the weights of each time series based on the prediction error and learning progress. This enables the model to better adapt to different traffic patterns and improves the robustness and accuracy of the prediction. Specifically, a weight adjustment function f(e, p) is defined, where e represents the prediction error and p represents the learning progress. This function calculates and updates the weight of each time series based on the real-time prediction error and learning progress.

[0157] The weight adjustment function f(e,p) accepts two parameters: the prediction error e and the learning progress p. This function calculates and updates the weight of each time series based on the real-time situation of the prediction error and learning progress.

[0158] The specific implementation can be as follows:

[0159] ① When the prediction error e is small, increase the weight of the corresponding time series to reward the correct prediction of the model. Specifically, a direct proportional function can be used to increase the weight, for example:

[0160] f(e,p)=a*e+b*p+c, where a is a proportional coefficient greater than 0, and b and c are appropriate constants.

[0161] ② When the prediction error e is large, reduce the weight of the corresponding time series to punish the model's incorrect predictions. Specifically, an inverse proportional function can be used to reduce the weight, for example:

[0162] f(e,p)=d / (e+ε)+b*p+c, where d is a proportional coefficient greater than 0, ε is a small positive number, and b and c are appropriate constants.

[0163] ③ When the learning progress p is high, for example, between 0 and 0.5, increase the weight of the corresponding time series to encourage the model to learn more historical information. Specifically, you can use an increasing function to increase the weight, for example:

[0164] f(e,p)=a*e+b*p^2+c, where a is a proportional coefficient greater than 0, and b and c are appropriate constants.

[0165] ④ When the learning progress p is low, reduce the weight of the corresponding time series to prevent the model from over-relying on historical information. Specifically, a decreasing function can be used to reduce the weight, for example:

[0166] f(e,p)=a*e+b / (p+ε)+c, where a is a proportional coefficient greater than 0, ε is a small positive number, and b and c are appropriate constants.

[0167] Through the above strategies, we can fully learn and accurately capture the multi-timescale characteristics of historical traffic data. At the same time, the specific form of the function f(e,p) can be adjusted and optimized according to actual needs.

[0168] The weight adjustment function f(e,p) is based on observations and theoretical analysis of historical traffic data. It was found that traffic data exhibits certain time series characteristics, such as volatility, periodicity, and trends, which have a significant impact on traffic forecasting. Furthermore, it was noted that prediction error and learning progress are also key factors affecting model performance. Therefore, a weight adjustment function f(e,p) is proposed to dynamically adjust the weights of each time series based on prediction error and learning progress.

[0169] In the specific implementation of the weight adjustment function f(e,p), the reasons for choosing a direct proportional function to increase the weight, an inverse proportional function to decrease the weight, an increasing function to increase the weight, and a decreasing function to decrease the weight are as follows:

[0170] Direct Proportional Function: A direct proportional function's characteristic is that its value increases with the independent variable. Therefore, when the prediction error e is large, it is desirable to increase the weight of the time series to better fit the complex variations in traffic data. The direct proportional function effectively meets this requirement, increasing the weight as the prediction error e increases.

[0171] Inverse proportional function: The characteristic of an inverse proportional function is that its value gradually decreases as the independent variable increases. When the forecast error e is small, it is desirable to reduce the weight of the time series to avoid overfitting and the influence of noise. The inverse proportional function can well meet this requirement. As the forecast error e decreases, the weight also decreases.

[0172] Increasing function: An increasing function's value gradually increases as the independent variable increases. When the learning progress p is low, it is desirable to increase the weight of each time series to better capture long-term dependencies and cyclical patterns in traffic data. The increasing function can effectively meet this requirement, gradually increasing the weight as the learning progress p increases.

[0173] Decreasing function: A decreasing function's value decreases as the independent variable increases. When the learning progress p is high, it's desirable to reduce the weight of each time series to avoid overfitting and noise. The decreasing function effectively meets this requirement, gradually decreasing the weight as the learning progress p increases.

[0174] In summary, by selecting the direct proportional function, inverse proportional function, increasing function and decreasing function as the specific implementation of the weight adjustment function f(e,p), the weight of each time series can be dynamically adjusted according to the prediction error and learning progress, thereby improving the model's fitting ability for traffic data and prediction accuracy.

[0175] The range of the prediction error e can be determined according to the actual situation and data characteristics. Generally speaking, the prediction error can be divided into the following situations:

[0176] Small error range: When the prediction error is within a small range, the prediction result can be considered relatively accurate.

[0177] Medium error range: When the forecast error is within the medium range, it can be considered that the forecast result has a certain degree of uncertainty.

[0178] Out of range: When the forecast error exceeds the acceptable range, the forecast result can be considered unreliable.

[0179] In this method, the value of error e is determined according to the actual situation. When the error is less than 50% of the total sample size, an error below 20% is defined as a small error, 20%-50% is defined as a large error, and an error greater than 50% is defined as an unusable error, which requires re-learning.

[0180] When the sample size is above 50% of the total, the actual results are more reliable because of the large sample size. Therefore, errors below 10% are defined as small errors, 10%-40% are defined as large errors, and errors greater than 40% are defined as unusable errors, which require relearning.

[0181] The value range of learning progress p can be determined according to actual conditions. The following are some possible value ranges:

[0182] Rapid learning stage: When the model starts to learn quickly, the learning progress p can gradually increase, and the value range can be between 0 and 0.5.

[0183] Stable learning stage: When the model enters the stable learning stage, the value of p can increase accordingly, and the value of the learning progress p can be between 0.5 and 1.

[0184] To determine the constants a, b, and c, you need to experiment and tune them based on the actual situation and data characteristics. The following are examples of actual values used in this method.

[0185] When the prediction error e is small, the weight adjustment function is: f(e,p)=a*e+b*p+c.

[0186] The actual value used in this method is f(e,p)=0.2*e+0.3*p+0.01.

[0187] When the prediction error e is large, the weight adjustment function is: f(e,p)=d / (e+ε)+b*p+c.

[0188] The actual value used in this method is f(e,p)=2 / (e+0.01)+0.3*p+0.01.

[0189] When the learning progress p is high, the weight adjustment function is: f(e,p)=a*e+b*p^2+c.

[0190] The actual value used in this method is f(e,p)=0.3*e+0.5*p^2+0.01.

[0191] When the learning progress p is low, the weight adjustment function is: f(e,p)=a*e+b / (p+ε)+c.

[0192] The actual value used in this method is f(e,p)=0.3*e+2 / (p+0.01)+0.01.

[0193] The determination of constants a, b, and c requires experimentation and tuning based on actual conditions and data characteristics. The following are some simple reasoning formulas and processes:

[0194] ① When the prediction error e is small, the weight adjustment function is f(e,p)=a*e+b*p+c.

[0195] The value of a can range from 0 to 1, indicating the degree of reward for prediction error, and can be fine-tuned according to actual conditions. For example, when a = 0.2, the weight is increased by 20% for each unit of prediction error.

[0196] The value of b can range from 0 to 1, indicating the degree of reward for learning progress, and can also be fine-tuned according to actual conditions. In this method, the value b = 0.3 means that the weight increases by 30% for each unit of progress.

[0197] c is a constant that represents the minimum value of the weight to avoid the situation where the weight is zero. The value range of c can be determined according to the actual situation. In this method, the value c = 0.01.

[0198] When the learning progress is similar, learning will be rewarded based on the error situation. When the progress is consistent, the smaller the weight, the smaller the error, and the better the machine learning effect.

[0199] ② When the prediction error e is large, the weight adjustment function is f(e,p)=d / (e+ε)+b*p+c.

[0200] The value of d can range from 1 to positive infinity, indicating the degree of penalty for prediction error, and can also be fine-tuned according to actual conditions. For example, when d = 2, each unit of prediction error reduces the weight by 2 times, and this value is taken as 2 in this method.

[0201] ε is a small positive number that represents the penalty threshold for prediction error to avoid the denominator being zero. The value range of ε can be determined according to the actual situation. In this method, ε = 0.01.

[0202] The value of b can range from 0 to 1, indicating the degree of reward for learning progress, and can be fine-tuned according to actual conditions. In this weight function, the value b = 0.3 means that the weight increases by 30% for each unit of progress.

[0203] c is a constant and a reserved parameter in this formula. The value range of c can be determined according to the actual situation. In this method, the value c = 0.01.

[0204] When the learning progress is consistent, the larger the error, the smaller the weight, which will penalize the machine learning, thereby achieving a state where the machine learning reduces the error.

[0205] ③When the learning progress p is high, the weight adjustment function is f(e,p)=a*e+b*p^2+c.

[0206] The value of a can range from 0 to 1, indicating the degree of reward for prediction error, and can also be fine-tuned according to actual conditions. For example, when a = 0.1, the weight is increased by 10% for each unit of prediction error.

[0207] The value of b can range from 0 to positive infinity, indicating the degree of reward for learning progress, and can be fine-tuned based on actual conditions. For example, when b = 0.5, each additional unit of progress increases the weight by 50%. It should be noted that to avoid excessive weights, an upper limit can be added, for example, limiting the maximum value of b to 1. In this method, the value of b is 0.5.

[0208] c is a constant, which is a reserved parameter in this formula for weight fine-tuning. The value range of c can be determined according to the actual situation. In this method, the value c = 0.01.

[0209] ④ When the learning progress p is low, the weight adjustment function is f(e,p)=a*e+b / (p+ε)+c.

[0210] The value of a can range from 0 to 1, indicating the degree of reward for prediction error, and can also be fine-tuned according to actual conditions. For example, when a = 0.3, the weight is increased by 30% for each unit of prediction error.

[0211] The value of b can range from 0 to positive infinity, indicating the degree of penalty for learning progress, and can be fine-tuned based on actual conditions. For example, when b = 1, each unit of progress reduction reduces the weight by 1. It should be noted that to avoid negative numbers, a lower limit can be added, for example, limiting the minimum value of b to 1. The value selected in this method is 2.

[0212] ε is a small positive number that represents the penalty threshold for learning progress to avoid the denominator being zero. The value range of ε can be determined according to the actual situation, for example, ε = 0.01.

[0213] c is a constant, reserved parameter. The value range of c can be determined according to actual conditions. In this method, the value c = 0.01.

[0214] As the learning progress p increases, the function value gradually decreases and the weight gradually decreases.

[0215] When the errors are similar, the weights gradually increase as the learning progress p increases.

[0216] Reference Figure 5 , Figure 5 The diagram shows the principle of processing abnormal data provided by the present invention, as shown in FIG. Figure 5 As shown:

[0217] The comparison results automatically trigger the corresponding processing flow. If the switch has experienced an abnormality, fault diagnosis will be performed and corresponding treatment strategies will be generated, such as automatic repair and early warning notification. In addition, the module can also optimize and adjust non-abnormal information to improve the operating efficiency of the switch.

[0218] The implementation principle of the exception handling module is as follows:

[0219] First, the switch's operating status is determined based on the comparison results. If an abnormality is found, fault diagnosis is performed to analyze the cause and location of the abnormality.

[0220] Then, a corresponding handling strategy is generated based on the fault diagnosis results. This handling strategy can include automatic repair or early warning notification. Automatic repair can be achieved by calling repair commands or automatic repair tools provided by the switch; early warning notification can be sent by email or text message to notify the administrator for manual handling.

[0221] Furthermore, the exception handling module can optimize and adjust non-abnormal information to improve the switch's operating efficiency. For example, it can adjust the switch's port configuration and optimize routing strategies based on the network traffic analysis results.

[0222] The exception handling process includes:

[0223] Data preprocessing: Preprocess historical fault data and equipment status information, including feature extraction, feature selection and other steps.

[0224] Data classification: Based on the characteristics and changing trends of abnormal outflows, classification algorithms are used to classify them into different categories.

[0225] Develop treatment measures: Based on the classification results, formulate corresponding treatment measures. For example, if the classification is "network traffic abnormality", measures such as isolating the faulty network port and starting the backup port can be taken to prioritize business recovery; if the classification is "temperature", measures such as intelligent fan switching can be taken to reduce the device temperature.

[0226] Data storage and analysis: Store the processed exception data for subsequent analysis and prediction.

[0227] Data storage and analysis: Store the processed exception data for subsequent analysis and prediction.

[0228] To sum up, this application can detect abnormal conditions in a timely manner and take corresponding treatment measures through real-time monitoring and comparison of switch information. Through automatic uploading and management of switch information, it realizes deep mining of historical data and predictive maintenance. Through automatic processing and optimization adjustment of switch information, it improves the stability and efficiency of the data center.

[0229] The foregoing is merely a list of specific embodiments of the present application, intended to enable those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application is not limited to the embodiments shown herein, but is intended to conform to the broadest scope consistent with the principles and novel features of the present application.

Claims

1. A device information monitoring method, characterized in that: include: Collect device information; Preprocessing the collected device information, wherein the preprocessing includes: comparing the collected device information with a set reference value to obtain the operating status of the device, and using an adaptive weighted time series analysis model to learn historical traffic data in the device information, and dynamically adjusting the weight of each time series based on the prediction error of the historical traffic data and the learning results to predict the future operating status of the device; The pre-processed device information is automatically uploaded to the cloud server, and the device information is monitored by the cloud server.

2. The method according to claim 1, characterized in that The method of using the adaptive weighted time series analysis model to learn the historical traffic data in the device information includes: Based on the acquired performance data of the device, historical network traffic data of the device is acquired from the performance data; The adaptive weighted time series analysis model arranged on the device is used to process the time series characteristics of the historical traffic data through a sliding time window to capture the multiple time scale characteristics of the traffic data.

3. The method according to claim 2, characterized in that The processing of the time series characteristics of the historical traffic data through the sliding time window includes: Preprocessing the historical data includes: data cleaning, removing outliers, and filling missing values; Selecting a time window length and dividing the pre-processed historical traffic data into a plurality of continuous time windows in chronological order; The data within each time window is processed and analyzed, and the data processing results within each time window are output or visualized.

4. The method according to claim 1, wherein The method dynamically adjusts the weights of each time series based on the prediction error and learning results of historical traffic data to predict the future operating status of the equipment, including: Outputting a prediction error and a learning result of the time series characteristics of the historical traffic data using the adaptive weighted time series analysis model; Based on the dynamic weight adjustment mechanism introduced in the adaptive weight time series analysis model, the weights of each time series are dynamically adjusted through the prediction error and learning results output by the adaptive weight time series analysis model, so that the adaptive weight time series analysis model can adapt to different traffic change patterns.

5. The method according to claim 1, wherein The collected device information is compared with a set reference value to obtain the operating status of the device, including: According to the historical data and actual operation status of the switch, the corresponding comparison reference frame is set, and the collected device information is compared with the set reference value; If the comparison result shows that the device information is not within the set reference value range, it is determined that the device operation status is abnormal; otherwise, it is determined that the device operation status is normal.

6. The method according to claim 5, characterized in that Also includes: If the equipment operation status is determined to be abnormal, the fault data and equipment status information are pre-processed to extract abnormal features; Based on abnormal characteristics and change trends, classification algorithms are used to classify faults, and corresponding processing strategies are formulated based on the classification results.

7. The method according to claim 1, characterized in that The acquisition device information includes: Collect switch configuration information through various methods, including web crawlers and device APIs; Collect switch performance data through network traffic analysis tools; Collect alarm information by analyzing the switch's log files or receiving the device's alarm signals.

8. The method according to claim 1, characterized in that Automatically uploading the pre-processed device information to a cloud server and monitoring the device information through the cloud server includes: Organizing the processed device information into a unified data format and automatically uploading it to the cloud server through regular batch uploading or real-time synchronization of data streams; After receiving the device information, the cloud server analyzes and manages the historical data in the device information through cloud storage and computing resources to monitor the device information.

9. The method according to claim 8, characterized in that Also includes: The uploaded data is encrypted, wherein the encryption process includes: symmetric encryption algorithm and asymmetric encryption algorithm.

10. A device information monitoring apparatus for implementing the device information monitoring method according to any one of claims 1 to 9, characterized in that: include: A collection module, which is used to collect device information; a data processing module for preprocessing the collected device information, wherein the preprocessing includes: comparing the collected device information with a set reference value to obtain the operating status of the device; using an adaptive weighted time series analysis model to learn the historical flow data in the device information; and dynamically adjusting the weight of each time series based on the prediction error of the historical flow data and the learning results to predict the future operating status of the device; The cloud upload module is used to automatically upload the pre-processed device information to the cloud server and monitor the device information through the cloud server.

Images